[Test] (Review): border-colour clamping, a readback across a mip gap, and blending on renderbuffer formats with no exact VkFormat

HeadlessGL reported the literal "<unset>" as the backend name when the variable was
not set, while MG_ConfigLoader::InitBackendType defaults it to DirectGLES and brought
DirectGLES up. Every ctest entry sets the variable through its ENVIRONMENT property,
which is why nothing ever noticed - but run straight from an adb shell, where nothing
sets it, the name matched neither backend and every case gated on DirectGLES skipped
as though DirectGLES were not running. On an Adreno 830 that silently disabled the
gl_ViewportIndex emulation control and the DirectGLES-only attachment scenarios.

CMakeLists.txt

@@ -335,6 +335,7 @@ set(SOURCE_FILES
     MobileGL/MG_Impl/GLImpl/Program/ProgramInterface.cpp
     MobileGL/MG_Impl/GLImpl/Program/GL_ProgramPipeline.cpp
     MobileGL/MG_Impl/GLImpl/Texture/GL_Texture.cpp
+    MobileGL/MG_Impl/GLImpl/Debug/GL_Debug.cpp
     MobileGL/MG_Impl/GLImpl/Texture/Validators.cpp
     MobileGL/MG_Impl/GLImpl/Texture/ProxyTexture.cpp
     MobileGL/MG_Impl/GLImpl/VertexArray/GL_VertexArray.cpp
@@ -392,6 +393,7 @@ set(SOURCE_FILES
     MobileGL/MG_State/GLState/TextureState/TextureObject2DCube.cpp
     MobileGL/MG_State/GLState/TextureState/TextureObject3D.cpp
     MobileGL/MG_State/GLState/TextureState/TextureObjectBuffer.cpp
+    MobileGL/MG_State/GLState/TextureState/TextureObjectView.cpp
     MobileGL/MG_State/GLState/TextureState/TextureUnit.cpp
     MobileGL/MG_State/GLState/TextureState/TextureState.cpp
     MobileGL/MG_State/GLState/ProgramState/ProgramObject.cpp

MobileGL/Config.h

@@ -69,6 +69,13 @@ namespace MobileGL::MG_Config {
     struct FeaturesTable {
         // MOBILEGL_DISABLE_TIMERQUERY: do not advertise or use GPU timer queries.
         Bool DisableTimerQuery = false;
+        // MOBILEGL_ENABLE_GLES_TEXTURE_VIEW: advertise GL_ARB_texture_view on DirectGLES when
+        // the host ES driver has EXT/OES_texture_view. Off by default: the host extension is
+        // present on Adreno 830 and the functional half of KHR-GL4{2,3}.texture_view still fails
+        // there, because the view's ES internalformat is normalized independently of the storage
+        // it aliases (see BackendObject_DirectGLES::BuildAdvertisedExtensions). The flag exists
+        // so that work can be done without editing the gate.
+        Bool EnableGlesTextureView = false;
         // MOBILEGL_ENABLE_SPIRV_VALIDATION: validate generated and transformed SPIR-V.
         // Disabled by default because validation is a diagnostics-only cost.
         Bool EnableSpirvValidation = false;
@@ -148,6 +155,11 @@ namespace MobileGL::MG_Config {
         // per-draw glBufferSubData path instead of the persistent-mapped ring allocator
         // (negative control / driver-bug escape hatch).
         Bool DisableUboRing = false;
+        // MOBILEGL_DISABLE_UNPACK_RING: force DirectGLES texture uploads back to
+        // glTexSubImage from the client pointer instead of staging them through the
+        // persistent-mapped unpack-PBO ring (negative control / driver-bug escape
+        // hatch).
+        Bool DisableUnpackRing = false;
         // MOBILEGL_ESPRYT_FORCE_DS_READBACK_EMULATION: make DirectGLES skip the native ES
         // depth/stencil reads and always go through the shader-sampling emulation. Core GL
         // ES has no depth or stencil readback, but some drivers accept it anyway (Mesa does,

MobileGL/ConfigLoader.cpp

@@ -162,6 +162,7 @@ namespace MobileGL::MG_ConfigLoader {
     inline void InitFeatures() {
         auto& features = MG_Config::Features;
         features.DisableTimerQuery = QueryEnvFlag("MOBILEGL_DISABLE_TIMERQUERY");
+        features.EnableGlesTextureView = QueryEnvFlag("MOBILEGL_ENABLE_GLES_TEXTURE_VIEW");
         features.EnableSpirvValidation = QueryEnvFlag("MOBILEGL_ENABLE_SPIRV_VALIDATION");
         features.UseAngle = QueryEnvFlag("MOBILEGL_USE_ANGLE");
 #if defined(MOBILEGL_TRACE_ANGLE_VARIANTS)
@@ -182,6 +183,7 @@ namespace MobileGL::MG_ConfigLoader {
         features.CoherentAsFlush = QueryEnvFlag("MOBILEGL_COHERENT_AS_FLUSH");
         features.TraceSkipAutodestroy = QueryEnvFlag("MOBILEGL_TRACE_SKIP_AUTODESTROY");
         features.DisableUboRing = QueryEnvFlag("MOBILEGL_DISABLE_UBO_RING");
+        features.DisableUnpackRing = QueryEnvFlag("MOBILEGL_DISABLE_UNPACK_RING");
         features.EsprytForceDepthStencilReadbackEmulation =
             QueryEnvFlag("MOBILEGL_ESPRYT_FORCE_DS_READBACK_EMULATION");
         features.RelaxedSemantics = QueryEnvFlag("MOBILEGL_RELAXED_SEMANTICS");

MobileGL/MG_Backend/BackendObject.h

@@ -301,6 +301,12 @@ namespace MobileGL {
 
         struct DynamicBackendParameters {
             SizeT UniformBufferOffsetAlignment = 256;
+            // GL_SHADER_STORAGE_BUFFER_OFFSET_ALIGNMENT, which is a SEPARATE limit from the
+            // uniform one and is routinely larger: Adreno 830 reports 32 for uniform buffers and
+            // 64 for storage buffers. Answering the storage query with the uniform value let an
+            // application bind a storage range at an offset the driver cannot address, which it
+            // accepted without error and then wrote somewhere else entirely.
+            SizeT ShaderStorageBufferOffsetAlignment = 256;
             // GL_MAX_TEXTURE_MAX_ANISOTROPY_EXT. 1.0 means the backend cannot filter anisotropically,
             // which is also why the extension is not advertised in that case.
             Float MaxTextureMaxAnisotropy = 1.0f;
@@ -383,6 +389,19 @@ namespace MobileGL {
             // where there is no device to be honest about and BuildTBuiltInResource still has to
             // hand glslang a workable gl_MaxClipDistances.
             Int MaxClipDistances = 8;
+            // GL_MAX_CULL_DISTANCES and GL_MAX_COMBINED_CLIP_AND_CULL_DISTANCES, under exactly
+            // the contract stated for MaxClipDistances above: ZERO IS A LEGAL ANSWER and a
+            // backend that cannot host a cull distance MUST report it. The failure this prevents
+            // is worse than the clip one, because cull distance discards the whole primitive:
+            // glslang bounds gl_CullDistance[i] against maxCullDistances and expands
+            // gl_MaxCullDistances from it, SPIRV-Cross then emits
+            // `#extension GL_EXT_clip_cull_distance : require` into the ESSL, and a host driver
+            // without that extension rejects the program in an info log nobody surfaces. These
+            // used to be bare 8s inside BuildTBuiltInResource with no backend consulted at all.
+            // The DEFAULTS are the GL 4.5 core minimums for the same reason MaxClipDistances'
+            // is: they describe the no-backend case (standalone compiles, unit tests).
+            Int MaxCullDistances = 8;
+            Int MaxCombinedClipAndCullDistances = 8;
             Int MaxViewports = 16;
             // GL_LAYER_PROVOKING_VERTEX / GL_VIEWPORT_INDEX_PROVOKING_VERTEX: which vertex of a
             // primitive supplies gl_Layer and gl_ViewportIndex. GL 4.6 table 23.65 makes

MobileGL/MG_Backend/DirectGLES/BackendObject_DirectGLES.cpp

@@ -307,6 +307,23 @@ namespace MobileGL::MG_Backend::DirectGLES {
             return capabilities.MaxColorTextureSamples;
         }
 
+        // The RENDERBUFFER twin, and it is a different set of pnames on purpose.
+        // GL_MAX_{COLOR,DEPTH}_TEXTURE_SAMPLES bound multisample TEXTURES; a renderbuffer is
+        // bounded by GL_MAX_SAMPLES (GL 4.6 core 9.2.4), with GL_MAX_INTEGER_SAMPLES for the
+        // integer formats. Using the texture ceilings here - which is what the renderbuffer probe
+        // did - is not merely untidy: the two texture pnames are ES 3.1 state, so on an ES 3.0
+        // context the loader's rejected-probe clamp leaves them at 1 (see the multisample clamps
+        // in the GLES loader) and the walk below would never run past one sample, recording {1}
+        // for EVERY colour format while GL_MAX_SAMPLES - ES 3.0 core, so genuinely answered -
+        // reports 4. Once the frontend validates against this list, that would reject every
+        // multisample renderbuffer on such a context.
+        Int GetGLESRenderbufferFormatMaxSamples(const MG_External::GLESCapabilities& capabilities,
+                                                GLenum imageFormat) {
+            const Bool isInteger = imageFormat == GL_RED_INTEGER || imageFormat == GL_RG_INTEGER ||
+                                   imageFormat == GL_RGB_INTEGER || imageFormat == GL_RGBA_INTEGER;
+            return isInteger ? capabilities.MaxIntegerSamples : capabilities.MaxSamples;
+        }
+
         Bool ProbeFramebufferCompletenessForTexture(const MG_External::GLESFunctionsTable& gl, TextureTarget target,
                                                     GLuint texture, TextureInternalFormat format) {
             GLuint framebuffer = 0;
@@ -717,7 +734,7 @@ namespace MobileGL::MG_Backend::DirectGLES {
                         AddFullFormatCaps(cache, renderbufferTargetIndex, formatIndex,
                                           GetRenderbufferFeatureCaps(logicalFormat));
                         const Int maxSamples =
-                            GetGLESFormatMaxSamples(capabilities, logicalFormat, nativeInfo.ImageFormat);
+                            GetGLESRenderbufferFormatMaxSamples(capabilities, nativeInfo.ImageFormat);
                         cache.SampleCounts[renderbufferTargetIndex][formatIndex] =
                             ProbeRenderbufferSampleCounts(gl, nativeInfo.InternalFormat, logicalFormat, maxSamples);
                     } else {
@@ -731,7 +748,7 @@ namespace MobileGL::MG_Backend::DirectGLES {
                         LogGLESFormatCaveat(logicalFormat, renderbufferTargetIndex, renderbufferFallbackInfo);
                     }
                     const Int maxSamples =
-                        GetGLESFormatMaxSamples(capabilities, logicalFormat, renderbufferFallbackInfo.ImageFormat);
+                        GetGLESRenderbufferFormatMaxSamples(capabilities, renderbufferFallbackInfo.ImageFormat);
                     cache.SampleCounts[renderbufferTargetIndex][formatIndex] = ProbeRenderbufferSampleCounts(
                         gl, renderbufferFallbackInfo.InternalFormat, logicalFormat, maxSamples);
                 }
@@ -749,11 +766,11 @@ namespace MobileGL::MG_Backend::DirectGLES {
                 .ExtraVendor = Nullopt,              // Extra vendor
                 .RendererGLInfo =
                     {
-                        .TargetGLVersion = {4, 0, 0},   // GL target version
+                        .TargetGLVersion = {4, 6, 0},   // GL target version
                         .TargetGLSLVersion = {4, 6, 0}, // Target Shading Language Version
                         // Baseline advertisement (no runtime capabilities yet); reconciled once
                         // the ES capabilities exist, see UpdateAdvertisedCapabilityExtensions.
-                        .Extensions = BuildAdvertisedExtensions(false, false, false, false),
+                        .Extensions = BuildAdvertisedExtensions(false, false, false, false, false, false),
                         .IsCompatibilityProfile = false // Is Compatibility Profile
                     },
                 .StaticBackendCapability = {.AllowVSOnlyPrograms = false} // Backend Capability
@@ -777,7 +794,8 @@ namespace MobileGL::MG_Backend::DirectGLES {
             MutableRendererInfo().RendererGLInfo.Extensions = BuildAdvertisedExtensions(
                 AreTimerQueriesSupported(), capabilities.SupportsTextureFilterAnisotropy,
                 capabilities.SupportsDrawIndirect,
-                capabilities.SupportsDrawIndirect && capabilities.SupportsBaseInstance);
+                capabilities.SupportsDrawIndirect && capabilities.SupportsBaseInstance,
+                capabilities.SupportsTextureView, capabilities.SupportsTextureCubeMapArray);
         }
     } // namespace
 
@@ -990,9 +1008,16 @@ namespace MobileGL::MG_Backend::DirectGLES {
 
     Vector<GLExtension> BuildAdvertisedExtensions(Bool timerQueriesSupported, Bool anisotropicFilteringSupported,
                                                   Bool drawIndirectSupported,
-                                                  Bool nonZeroIndirectBaseInstanceSupported) {
+                                                  Bool nonZeroIndirectBaseInstanceSupported,
+                                                  Bool textureViewSupported, Bool cubeMapArraySupported) {
         Vector<GLExtension> extensions = {
-            V_OpenGL30, V_OpenGL31, V_OpenGL32, V_OpenGL33, V_OpenGL40, E_GL_ARB_draw_buffers_blend,
+            // The version tokens have to reach the version the backend actually claims:
+            // TargetGLVersion is {4,6,0}, and a list that stopped at OpenGL40 told an
+            // application feature-detecting off these tokens the opposite of what
+            // GL_MAJOR_VERSION / GL_MINOR_VERSION told it.
+            V_OpenGL30, V_OpenGL31, V_OpenGL32, V_OpenGL33, V_OpenGL40, V_OpenGL41, V_OpenGL42, V_OpenGL43,
+            V_OpenGL44, V_OpenGL45, V_OpenGL46,
+            E_GL_ARB_draw_buffers_blend,
             E_GL_ARB_compute_shader, E_GL_ARB_shader_storage_buffer_object, E_GL_ARB_shader_image_load_store,
             E_GL_ARB_clear_buffer_object, E_GL_ARB_program_interface_query, E_GL_ARB_framebuffer_object, E_GL_EXT_framebuffer_object,
             E_GL_ARB_depth_texture, E_GL_ARB_buffer_storage, E_GL_ARB_texture_storage,
@@ -1017,6 +1042,77 @@ namespace MobileGL::MG_Backend::DirectGLES {
             // has had the same texture parameter since ES 3.1, which every device MobileGL
             // runs on provides.
             E_GL_ARB_stencil_texturing,
+            // Core since 3.2 and implemented here on both backends - glDrawElementsBaseVertex,
+            // glDrawRangeElementsBaseVertex, glDrawElementsInstancedBaseVertex and
+            // glMultiDrawElementsBaseVertex all reach real per-draw vertex rebasing. The string
+            // was simply never emitted, which left KHR-GL4*.draw_elements_base_vertex_tests
+            // NotSupported on a feature that works.
+            E_GL_ARB_draw_elements_base_vertex,
+            // The whole sync-object family is real and core since 3.2: glFenceSync, glIsSync,
+            // glDeleteSync, glClientWaitSync, glWaitSync and glGetSynciv all live in GLImpl over a
+            // backend fence (a host GLsync here, a VkFence on DirectVulkan), and glGetInteger64v
+            // answers GL_MAX_SERVER_WAIT_TIMEOUT. The string matters for the same reason
+            // ARB_uniform_buffer_object's does: LWJGL builds GLCapabilities from the extension
+            // list, and a caller that finds GL_ARB_sync missing never resolves the entry points -
+            // then calls through null if it uses fences anyway. Nothing in the CTS gates on this
+            // string, so it is advertised on the strength of the implementation, not a test unlock.
+            E_GL_ARB_sync,
+            // Atomic counters, core since 4.2. glGetActiveAtomicCounterBufferiv and the whole
+            // GL_ATOMIC_COUNTER_BUFFER_* query family are real in GLImpl, and SyncAtomicCounterBuffers
+            // re-issues the counter buffer as an SSBO binding in the range reserved at the top of
+            // the ES driver's shader-storage points, so a counter dispatch reads and writes the
+            // buffer the application bound. DirectVulkan reaches the same place through its own
+            // descriptor resolution, so the string is symmetric.
+            E_GL_ARB_shader_atomic_counters,
+            // glVertexAttribDivisor, core since 3.3 and real on both backends. Applications
+            // (Better Clouds' GLCompat among them) accept the extension string as an
+            // ALTERNATIVE to a 3.3 context when deciding whether instanced rendering is
+            // available, so withholding it makes MobileGL look less capable than it is.
+            E_GL_ARB_instanced_arrays,
+            // The whole of KHR_debug lives in GLImpl - the message log, the group stack and the
+            // object-label table are MobileGL's own state, not the host driver's - so it is as
+            // available here as it is on DirectVulkan, which has advertised it all along.
+            E_GL_KHR_debug,
+            // Core GL 3.0-4.3 plumbing that has been real here for as long as the backend has
+            // existed, and that was simply never named. None of these unlocks a single CTS case -
+            // the conformance suite reaches all of them through the version - so they are
+            // advertised for the OTHER consumer of this list: LWJGL builds GLCapabilities from the
+            // string set, and an application that gates its ENTRY POINTS on the string rather than
+            // on the version never resolves them and then calls through null. Each is backed by
+            // the entry points named beside it.
+            //
+            // glBindVertexArray / glGenVertexArrays / glDeleteVertexArrays / glIsVertexArray.
+            E_GL_ARB_vertex_array_object,
+            // The 14 glSamplerParameter* / glGetSamplerParameter* entry points, including the
+            // integer-valued Iiv/Iuiv forms.
+            E_GL_ARB_sampler_objects,
+            // glMapBufferRange + glFlushMappedBufferRange, which ARB_buffer_storage's persistent
+            // maps are already built on top of.
+            E_GL_ARB_map_buffer_range,
+            // glCopyBufferSubData plus the GL_COPY_READ_BUFFER / GL_COPY_WRITE_BUFFER targets.
+            E_GL_ARB_copy_buffer,
+            // glCopyImageSubData, wired to a real backend hook on both backends.
+            E_GL_ARB_copy_image,
+            // GL_TEXTURE_SWIZZLE_{R,G,B,A,RGBA}, which this backend syncs through to the ES
+            // driver's identical parameters.
+            E_GL_ARB_texture_swizzle,
+            // GL_INT_2_10_10_10_REV / GL_UNSIGNED_INT_2_10_10_10_REV on glVertexAttribPointer plus
+            // the eight glVertexAttribP* entry points.
+            E_GL_ARB_vertex_type_2_10_10_10_rev,
+            // The R/RG internal formats. Named separately from the float ones because an
+            // application may check either.
+            E_GL_ARB_texture_rg,
+            // GL_DEPTH_COMPONENT32F and GL_DEPTH32F_STENCIL8.
+            E_GL_ARB_depth_buffer_float,
+            // The floating-point colour formats. Unlike the rest of this block this string DOES
+            // gate CTS cases - KHR-GL4*.internalformat.texture2d.*{16f,32f} is keyed on it with no
+            // core-version fallback, so eight cases per version list were NotSupported on formats
+            // the backend has always had.
+            E_GL_ARB_texture_float,
+            // glViewportArrayv / glViewportIndexedf{,v} / glScissorArrayv / glScissorIndexed{,v} /
+            // glDepthRangeArrayv / glDepthRangeIndexed / glGetFloati_v / glGetDoublei_v, over the
+            // 16 viewports GL_MAX_VIEWPORTS reports and the per-viewport routing emulation.
+            E_GL_ARB_viewport_array,
             // Advertised with GL_NUM_PROGRAM_BINARY_FORMATS = 0, which the
             // extension explicitly permits. It is also the only thing that
             // exposes glProgramParameteri before GL 4.1.
@@ -1065,6 +1161,47 @@ namespace MobileGL::MG_Backend::DirectGLES {
         if (timerQueriesSupported && !MG_Config::Features.DisableTimerQuery) {
             extensions.push_back(E_GL_ARB_timer_query);
         }
+        // Cube map arrays are core from GL 4.0 and from ES 3.2, but on a pre-ES-3.2 driver without
+        // EXT/OES_texture_cube_map_array there is nothing underneath: the texture gets no storage
+        // and a samplerCubeArray shader does not even compile, which is exactly what the POST
+        // reports. So the string follows the host capability rather than the version.
+        //
+        // Named for the application's benefit rather than the suite's: measured on Adreno 830,
+        // KHR-GL43.texture_gather.plain-gather-*-cube-array already passed without the string, so
+        // this unlocks no conformance case. It is advertised because the feature is real and
+        // because an application that feature-detects cube map arrays off the string (rather than
+        // off the 4.0 version) would otherwise decline a path this backend serves.
+        if (cubeMapArraySupported) {
+            extensions.push_back(E_GL_ARB_texture_cube_map_array);
+        }
+        // Only advertised when the host ES driver has EXT/OES_texture_view. ES has no core
+        // texture views at any version and no honest emulation exists: a view is a SECOND NAME
+        // over the SAME storage, so that writes through either are visible through the other and
+        // the two carry independent per-texture parameters at the same time - which is exactly
+        // what applications use it for (Better Clouds samples one D24S8 through its own name with
+        // DEPTH_STENCIL_TEXTURE_MODE = STENCIL_INDEX and through a view with DEPTH_COMPONENT, in
+        // a single shading pass). A copy-based fallback satisfies neither half, and fails
+        // silently; withholding the string and answering glTextureView with INVALID_OPERATION is
+        // the only behaviour that cannot be mistaken for success.
+        //
+        // The host extension is necessary and NOT sufficient, which is why this second gate
+        // exists. Adreno 830 has EXT_texture_view, and on it the whole functional half of
+        // KHR-GL4{2,3}.texture_view fails: base_and_max_levels, reference_counting and
+        // view_sampling Fail and view_classes crashes, while only the two pure-API cases
+        // (errors, gettexparameter - neither of which touches the host view) pass. The cause is
+        // known and is MobileGL's, not the driver's: SyncTextureViewToBackend normalizes the
+        // VIEW's ES internalformat independently of the storage it aliases, so whenever the two
+        // land on different renderability carriers the host rejects the pair, the error is
+        // swallowed, and the view is left as a storage-less name that samples as zeros.
+        // DirectVulkan builds the view as a second VkImageView over one VkImage and has no such
+        // seam - it passes 5 of the 7 cases on the same device - so the string stays there.
+        //
+        // Until that reconciliation exists, advertising here would be the same lie the comment
+        // above refuses to tell, just with an extra prerequisite met. Set
+        // MOBILEGL_ENABLE_GLES_TEXTURE_VIEW=1 to re-enable it for that work.
+        if (textureViewSupported && MG_Config::Features.EnableGlesTextureView) {
+            extensions.push_back(E_GL_ARB_texture_view);
+        }
         // Only advertised when the host ES driver actually filters anisotropically: the sampler
         // state is accepted regardless, but forwarding it would be a no-op without the extension,
         // and an app that trusts the string (LWJGL builds GLCapabilities from it) would silently
@@ -1204,6 +1341,8 @@ namespace MobileGL::MG_Backend::DirectGLES {
 
     void BackendObject_DirectGLES::UpdateDynamicBackendParameters() {
         m_dynamicParameters.UniformBufferOffsetAlignment = m_GLESCapabilities.UniformBufferOffsetAlignment;
+        m_dynamicParameters.ShaderStorageBufferOffsetAlignment =
+            m_GLESCapabilities.ShaderStorageBufferOffsetAlignment;
         m_dynamicParameters.MaxTextureMaxAnisotropy = m_GLESCapabilities.MaxTextureMaxAnisotropy;
         m_dynamicParameters.AliasedLineWidthRangeMin = m_GLESCapabilities.AliasedLineWidthRangeMin;
         m_dynamicParameters.AliasedLineWidthRangeMax = m_GLESCapabilities.AliasedLineWidthRangeMax;
@@ -1343,6 +1482,11 @@ namespace MobileGL::MG_Backend::DirectGLES {
         m_dynamicParameters.MaxDrawBuffers = m_GLESCapabilities.MaxDrawBuffers;
         m_dynamicParameters.MaxColorAttachments = m_GLESCapabilities.MaxColorAttachments;
         m_dynamicParameters.MaxClipDistances = m_GLESCapabilities.MaxClipDistances;
+        // The loader already gated both on GL_EXT_clip_cull_distance and left 0 without it, which
+        // is the answer that keeps glslang from accepting a gl_CullDistance the ESSL compiler
+        // would reject.
+        m_dynamicParameters.MaxCullDistances = m_GLESCapabilities.MaxCullDistances;
+        m_dynamicParameters.MaxCombinedClipAndCullDistances = m_GLESCapabilities.MaxCombinedClipAndCullDistances;
         m_dynamicParameters.MaxViewports = m_GLESCapabilities.MaxViewports;
         // Whatever the driver said about which vertex supplies gl_Layer, and GL_UNDEFINED_VERTEX
         // for gl_ViewportIndex on every driver without GL_OES_viewport_array - which is both test

MobileGL/MG_Backend/DirectGLES/BackendObject_DirectGLES.h

@@ -78,11 +78,12 @@ namespace MobileGL::MG_Backend::DirectGLES {
 
     // The full OpenGL extension list Espryt advertises (glGetString(GL_EXTENSIONS))
     // for a device whose timer queries / anisotropic filtering / native indirect draws /
-    // non-zero indirect baseInstance semantics are (or are not) usable.
+    // non-zero indirect baseInstance semantics / EXT-OES texture views are (or are not) usable.
     // The MOBILEGL_DISABLE_TIMERQUERY escape hatch is applied inside.
     Vector<GLExtension> BuildAdvertisedExtensions(Bool timerQueriesSupported, Bool anisotropicFilteringSupported,
                                                   Bool drawIndirectSupported,
-                                                  Bool nonZeroIndirectBaseInstanceSupported);
+                                                  Bool nonZeroIndirectBaseInstanceSupported,
+                                                  Bool textureViewSupported, Bool cubeMapArraySupported);
 
     // Format: <OpenGL ES Renderer>, OpenGL ES <Major>.<Minor> — the exact string an
     // initialized backend returns from GetBackendAPIVersionString (and that ends up

MobileGL/MG_Backend/DirectGLES/DirectGLES.cpp

@@ -20,6 +20,7 @@
 #include <MG_State/GLState/TextureState/TextureObjectBuffer.h>
 #include <MG_Impl/GLImpl/Framebuffer/GL_Framebuffer.h>
 #include <MG_Util/BackendLoaders/OpenGL/Loader.h>
+#include <MG_Util/SelfTest/DriverBugProbes.h>
 #include <MG_Util/Converters/GLToStr/GLEnumConverter.h>
 #include <MG_Util/Converters/MGToGL/TextureEnumConverter.h>
 #include <MG_Util/Converters/MGToGL/FramebufferEnumConverter.h>
@@ -350,9 +351,20 @@ namespace MobileGL::MG_Backend::DirectGLES {
 #ifdef TRACY_ENABLE
             ZoneScopedC(TRACY_ZONECOLOR_BACKEND);
 #endif
-            // Only sync up to the high-water mark of app-touched points; the fixed array is 36
+            // Only sync up to the high-water mark of app-touched points; the fixed array is 84
             // deep but apps bind a handful, so the never-touched tail is already at GL default 0.
             auto bindingPointCnt = MG_State::pGLContext->GetTouchedBufferBindingPointCount(target);
+            // ...and never past what the ES driver itself can hold. MobileGL advertises the GL 4.5
+            // minimum of 84 uniform binding points while the ES 3.2 minimum is 72, so a frontend
+            // index in that gap would reach glBindBufferBase as GL_INVALID_VALUE. Nothing is lost
+            // by stopping: this frontend-indexed pass exists for the compute path, and the
+            // per-program rebind in BindCurrentProgramWithResources - which is what actually feeds
+            // a shader - remaps every block a program declares onto a compacted ES point, so a
+            // block bound at GL point 83 still reaches its shader.
+            if (target == BufferTarget::Uniform && g_GLESCapabilities.MaxUniformBufferBindings > 0) {
+                bindingPointCnt = std::min(bindingPointCnt,
+                                           static_cast<SizeT>(g_GLESCapabilities.MaxUniformBufferBindings));
+            }
             for (SizeT i = 0; i < bindingPointCnt; ++i) {
                 auto& point = MG_State::pGLContext->GetBufferBindingPoint(target, i);
                 auto& obj = point.GetBoundObject();
@@ -1082,19 +1094,48 @@ namespace MobileGL::MG_Backend::DirectGLES {
             ZoneScopedC(TRACY_ZONECOLOR_BACKEND);
 #endif
             auto* backendTextureSlot = g_backendTextureObjects.Find(textureObject.get());
-            auto& backendObj = backendTextureSlot ? *backendTextureSlot
-                                                  : g_backendTextureObjects.GetOrCreate(textureObject);
-            if (!backendObj) {
-                backendObj = MakeShared<BackendTextureObject>();
+            auto& backendSlot = backendTextureSlot ? *backendTextureSlot
+                                                   : g_backendTextureObjects.GetOrCreate(textureObject);
+            if (!backendSlot) {
+                backendSlot = MakeShared<BackendTextureObject>();
             }
+
+            // A by-VALUE copy of the twin for the duration of the syncs below. `backendSlot` is a
+            // reference INTO the open-addressed registry, and syncing can RE-ENTER this function:
+            // a texture created by glTextureView has to sync the texture whose storage it views
+            // first (SyncTextureViewToBackend), and that nested call may insert, grow the map and
+            // relocate every entry - leaving the reference dangling. Holding the object itself
+            // keeps the calls below working on the right twin regardless; the slot is re-resolved
+            // at the end for the reference this function returns.
+            const SharedPtr<BackendTextureObject> backendObj = backendSlot;
+
             if (imageBindableStorageRequired) {
                 backendObj->RequireImageBindableStorage(textureObject);
             }
             backendObj->SyncTextureParamsToBackend(textureObject);
             backendObj->SyncBuiltinSamplerToBackend(textureObject);
             backendObj->SyncMipmapsToBackend(textureObject);
+            // The storage sync may RE-MINT the driver texture - a fresh glTexStorage after a
+            // shape change, an image-bindable widening, or the glTextureView that an
+            // ARB_texture_view view is created on - which discards every parameter the two calls
+            // above just pushed. Re-push them here rather than leaving it to the next sync: the
+            // very next thing that happens is usually the draw this sync was run for, and until
+            // the filters land the new texture is at the ES defaults, which for a single-level or
+            // integer texture is not merely mis-filtered but INCOMPLETE, i.e. it samples zero.
+            if (backendObj->NeedsParameterResync()) {
+                backendObj->SyncTextureParamsToBackend(textureObject);
+                backendObj->SyncBuiltinSamplerToBackend(textureObject);
+            }
 
-            return backendObj;
+            auto* refreshedSlot = g_backendTextureObjects.Find(textureObject.get());
+            auto& refreshedBackendObj = refreshedSlot ? *refreshedSlot
+                                                      : g_backendTextureObjects.GetOrCreate(textureObject);
+            if (!refreshedBackendObj) {
+                // A collection ran during the nested sync and took this slot with it; put the
+                // twin the caller is about to use back, rather than handing back an empty one.
+                refreshedBackendObj = backendObj;
+            }
+            return refreshedBackendObj;
         }
 
         // Identity snapshot of what one texture unit has bound: the object in every binding
@@ -2279,6 +2320,29 @@ namespace MobileGL::MG_Backend::DirectGLES {
                 }
             }
 
+            if (tailSpanDirty) { // Sample shading (ARB_sample_shading; ES 3.2 core)
+                // Both halves are gated on the same entry point rather than on a version check:
+                // GL_SAMPLE_SHADING and glMinSampleShading arrived together (ES 3.2 core /
+                // OES_sample_shading), so a null pointer means glEnable(GL_SAMPLE_SHADING) would
+                // only push an INVALID_ENUM into the driver's queue. This is NOT part of the
+                // SYNC_CAPABILITY block above for exactly that reason - that macro has nowhere to
+                // put a guard.
+                if (g_GLESFuncs.glMinSampleShading) {
+                    if (forceFullPush ||
+                        parameters.SampleShadingEnabled != g_syncedRenderStateParameters.SampleShadingEnabled) {
+                        if (parameters.SampleShadingEnabled) {
+                            g_GLESFuncs.glEnable(GL_SAMPLE_SHADING);
+                        } else {
+                            g_GLESFuncs.glDisable(GL_SAMPLE_SHADING);
+                        }
+                    }
+                    if (forceFullPush || parameters.MinSampleShadingValue !=
+                                             g_syncedRenderStateParameters.MinSampleShadingValue) {
+                        g_GLESFuncs.glMinSampleShading(parameters.MinSampleShadingValue);
+                    }
+                }
+            }
+
             g_syncedRenderStateVersion = currentRenderStateVersion;
             // Byte copy, not member copy: it also clones the frontend struct's padding bytes,
             // which is what lets the span memcmps above answer "unchanged" exactly instead of
@@ -2431,9 +2495,34 @@ namespace MobileGL::MG_Backend::DirectGLES {
                 // `layout(vertices = N) out` - so a glPatchParameteri between two draws makes the
                 // built program wrong. -1 is "this program needed no such stage", which compares
                 // equal to itself and costs every other program one integer test.
+                //
+                // GL_PATCH_DEFAULT_{OUTER,INNER}_LEVEL are baked into the same stage for the same
+                // reason (ES has neither the state nor an entry point), so glPatchParameterfv
+                // makes it stale too. Both level comparisons sit INSIDE the >= 0 guard: a program
+                // with a control stage of its own - which is nearly all of them - still pays only
+                // the one integer test.
+                //
+                // Compared by BIT PATTERN, matching what DirectVulkan hashes into its module key.
+                // A float compare here would never settle for a NaN level - NaN != NaN - and every
+                // draw of that program would re-transpile, re-compile and re-link a byte-identical
+                // shader. glPatchParameterfv accepts NaN by design.
+                //
+                // The gl_PerVertex MEMBER SET needs no clause of its own here, and that asymmetry
+                // with DirectVulkan is deliberate rather than an omission. It can only change with
+                // the evaluation stage, i.e. across a relink - which the link-version test at the
+                // top of this condition already catches - and this backend never invents the shape
+                // in the first place: AttachPassthroughTessControlStage extracts the member text
+                // out of the neighbouring stages' emitted ESSL on every rebuild
+                // (ExtractPerVertexBlockMembers, "mirrored, never invented"). DirectVulkan needs
+                // the mask in its key precisely because it does NOT mirror - it redeclares from a
+                // member set it has to be told.
                 (twin->GetPassthroughTessControlPatchVertices() >= 0 &&
-                 twin->GetPassthroughTessControlPatchVertices() !=
-                     static_cast<Int>(MG_State::pGLContext->GetPatchVertices()))) {
+                 (twin->GetPassthroughTessControlPatchVertices() !=
+                      static_cast<Int>(MG_State::pGLContext->GetPatchVertices()) ||
+                  !BitwiseEqual(twin->GetPassthroughTessControlOuterLevel(),
+                                MG_State::pGLContext->GetPatchDefaultOuterLevel()) ||
+                  !BitwiseEqual(twin->GetPassthroughTessControlInnerLevel(),
+                                MG_State::pGLContext->GetPatchDefaultInnerLevel())))) {
                 twin->SyncToBackend(currentProgram);
             }
             g_currentDrawFrontendProgram = currentProgram.get();
@@ -3494,6 +3583,18 @@ namespace MobileGL::MG_Backend::DirectGLES {
                                                const SharedPtr<MG_State::GLState::BufferObject>& drawIndirectBuffer,
                                                GLsizei drawcount, GLsizei stride, const char* label) {
         (void)label;
+        // An indirect command's firstIndex/count live in GPU memory, so the substitution has
+        // to rewrite the whole element array buffer rather than this draw's range - which is
+        // exactly what it does when no CPU-known count is handed to it. Held for the whole
+        // command loop so every command in the batch reads the rewritten copy.
+        //
+        // firstIndex counts ELEMENTS, so it survives a widened copy untouched; what does not
+        // survive is the type and the element size, which are re-taken from the substitution
+        // below for both the native and the CPU-unrolled path.
+        const ScopedRestartIndexSubstitution restart(type, /*count=*/0, /*indices=*/nullptr);
+        if (!restart.DrawIsValid()) return;
+        type = restart.IndexType();
+        indexSize = MG_Util::GetGLTypeSize(type);
         const Bool useNative = drawIndirectBuffer != nullptr && SupportsNativeIndirectDraws();
         if (useNative) {
             // gl_BaseInstance must observe GPU-written command fields; expose the indirect
@@ -3799,29 +3900,308 @@ namespace MobileGL::MG_Backend::DirectGLES {
         }
     }
 
-    // GLES core supports only GL_PRIMITIVE_RESTART_FIXED_INDEX (fixed all-ones value). If the app
-    // enabled the arbitrary GL_PRIMITIVE_RESTART with a non-fixed index, hard-fail at this draw with
-    // the reason (a fallback would silently drop restarts and corrupt geometry).
-    void CheckPrimitiveRestartSupported(GLenum indexType) {
+    // ---------------------------------------------------------------------------
+    // Arbitrary-index primitive restart
+    //
+    // Desktop GL restarts on whatever index glPrimitiveRestartIndex named; GLES core only
+    // ever restarts on the all-ones value of the index type. When the two agree - which
+    // includes every GL_PRIMITIVE_RESTART_FIXED_INDEX user - the render state push at
+    // SyncRenderState is the whole implementation and nothing here does any work. When they
+    // disagree the index DATA is rewritten into a scratch element array buffer.
+    //
+    // This used to throw instead. A throw here unwinds a C++ exception through the C GL ABI
+    // and takes the process down - the same hazard GL_Texture.cpp and RenderState.cpp
+    // already call out - so an application that merely asked for a legal desktop feature
+    // died rather than got an error.
+    // ---------------------------------------------------------------------------
+
+    namespace {
+        struct RestartScratchBuffer {
+            Uint id = 0;
+            SizeT capacity = 0;
+        };
+
+        RestartScratchBuffer g_restartIndices;
+        Vector<Uint8> g_restartStaging;
+
+        // Past this the rewrite would stage and re-upload hundreds of megabytes on EVERY
+        // draw (the copy is not memoised, exactly as on the Vulkan side). Decline instead of
+        // trying: a draw that renders nothing is recoverable, a stall of that size is not.
+        constexpr SizeT kMaxRestartRewriteBytes = SizeT{1} << 26; // 64 MiB
+
+        // The index type one step wider than this one, or 0 when there is none. Widening is how
+        // an all-ones value that is a REAL vertex index keeps its meaning while the all-ones
+        // value of the destination type serves as the restart sentinel: a source that cannot
+        // spell 0xFFFF cannot collide with a 16-bit sentinel, and likewise 8 -> 16.
+        GLenum WiderIndexType(GLenum indexType) {
+            switch (indexType) {
+            case GL_UNSIGNED_BYTE: return GL_UNSIGNED_SHORT;
+            case GL_UNSIGNED_SHORT: return GL_UNSIGNED_INT;
+            default: return 0;
+            }
+        }
+
+        Uint32 ReadIndex(const Uint8* source, SizeT i, SizeT indexSize) {
+            switch (indexSize) {
+            case 1: return source[i];
+            case 2: {
+                Uint16 narrow = 0;
+                std::memcpy(&narrow, source + i * 2, sizeof(narrow));
+                return narrow;
+            }
+            default: {
+                Uint32 wide = 0;
+                std::memcpy(&wide, source + i * 4, sizeof(wide));
+                return wide;
+            }
+            }
+        }
+
+        void WriteIndex(Uint8* destination, SizeT i, SizeT indexSize, Uint32 value) {
+            switch (indexSize) {
+            case 1: destination[i] = static_cast<Uint8>(value); break;
+            case 2: {
+                const Uint16 narrow = static_cast<Uint16>(value);
+                std::memcpy(destination + i * 2, &narrow, sizeof(narrow));
+                break;
+            }
+            default: std::memcpy(destination + i * 4, &value, sizeof(value)); break;
+            }
+        }
+
+        // True when any index in the range already holds the type's all-ones value, i.e. when
+        // that value is doing double duty as a real vertex index and so cannot also be the
+        // restart sentinel. Only asked on the rare substitution path.
+        Bool ContainsFixedRestartIndex(const Uint8* source, SizeT indexCount, SizeT indexSize,
+                                       Uint32 fixedMax) {
+            for (SizeT i = 0; i < indexCount; ++i) {
+                if (ReadIndex(source, i, indexSize) == fixedMax) return true;
+            }
+            return false;
+        }
+
+        // Copies index data, replacing every occurrence of the application's restart index with
+        // the all-ones value of the DESTINATION type - the only one GLES restarts on. The
+        // destination may be wider than the source, which is what makes the copy lossless: a
+        // source index equal to the source's all-ones value zero-extends to something the wider
+        // sentinel can never equal, so it stays the vertex it was.
+        //
+        // Same width in and out is the degenerate case, used when the source contains no
+        // all-ones index at all (nothing to protect) or when there is no wider type to move to.
+        // In that last case only - a GL_UNSIGNED_INT stream that really does use index
+        // 0xFFFFFFFF while asking to restart on a different one - a legal index has to be
+        // nudged to 0xFFFFFFFE, because 32 bits cannot hold both meanings. The caller logs it;
+        // it is the one input this feature cannot represent.
+        void RewriteRestartIndices(const Uint8* source, SizeT indexCount, SizeT sourceIndexSize,
+                                   SizeT destinationIndexSize, Uint32 applicationRestartIndex,
+                                   Uint32 destinationFixedMax, Vector<Uint8>& output) {
+            output.resize(indexCount * destinationIndexSize);
+            for (SizeT i = 0; i < indexCount; ++i) {
+                Uint32 value = ReadIndex(source, i, sourceIndexSize);
+                if (value == applicationRestartIndex) {
+                    value = destinationFixedMax;
+                } else if (value == destinationFixedMax) {
+                    // Only reachable when no widening was possible; see above.
+                    value = destinationFixedMax - 1;
+                }
+                WriteIndex(output.data(), i, destinationIndexSize, value);
+            }
+        }
+
+        // Whole-buffer respecify through the manager-wide staging target, so binding it
+        // disturbs no VAO state. glBufferData orphans the previous store, so the upload
+        // never waits on a draw still reading the old contents out of the same name.
+        Bool UploadRestartScratch(SizeT bytes, const void* data) {
+            if (g_restartIndices.id == 0) {
+                GLuint id = 0;
+                g_GLESFuncs.glGenBuffers(1, &id);
+                if (id == 0) return false;
+                g_restartIndices.id = id;
+                g_restartIndices.capacity = 0;
+            }
+            BufferImpl::BindBufferId(BufferImpl::TempBufferTarget, g_restartIndices.id);
+            SizeT capacity = g_restartIndices.capacity == 0 ? bytes : g_restartIndices.capacity;
+            while (capacity < bytes) capacity *= 2;
+            g_GLESFuncs.glBufferData(BufferImpl::TempBufferTarget, static_cast<GLsizeiptr>(capacity), nullptr,
+                                     GL_STREAM_DRAW);
+            g_restartIndices.capacity = capacity;
+            if (data != nullptr && bytes != 0) {
+                g_GLESFuncs.glBufferSubData(BufferImpl::TempBufferTarget, 0, static_cast<GLsizeiptr>(bytes), data);
+            }
+            return true;
+        }
+
+        const SharedPtr<MG_State::GLState::BufferObject>& BoundElementArrayBuffer() {
+            static const SharedPtr<MG_State::GLState::BufferObject> none;
+            const auto& vao = MG_State::pGLContext->GetBoundVertexArray();
+            if (!vao) return none;
+            return vao->GetIndexBufferBindingSlot().GetBoundObject();
+        }
+
+        // The GL name PrepareForDraw left on GL_ELEMENT_ARRAY_BUFFER, i.e. what the
+        // substitution has to put back.
+        Uint BoundElementArrayBufferId() {
+            const auto& ibo = BoundElementArrayBuffer();
+            if (!ibo) return 0;
+            const auto* resource = BufferImpl::EnsureBufferResource(ibo);
+            return resource ? resource->id : 0;
+        }
+    } // namespace
+
+    RestartSubstitutionKind ResolveRestartSubstitution(GLenum indexType) {
         if (!MG_State::pGLContext->IsCapabilityEnabled(CapabilityInput::PrimitiveRestart) ||
             MG_State::pGLContext->IsCapabilityEnabled(CapabilityInput::PrimitiveRestartFixedIndex)) {
+            return RestartSubstitutionKind::None;
+        }
+        const Uint32 fixedMax = MG_Util::FixedRestartIndexForGLType(indexType);
+        if (fixedMax == 0) return RestartSubstitutionKind::None;
+        const Uint32 restartIndex = MG_State::pGLContext->GetPrimitiveRestartIndex();
+        if (restartIndex == fixedMax) return RestartSubstitutionKind::None;
+        // Strictly greater, never truncated. GL 4.6 core 10.3.6 compares the fetched index
+        // zero-extended against the full 32-bit state, so an index this type cannot hold matches
+        // nothing. Truncating instead - glPrimitiveRestartIndex(0x100) over GL_UNSIGNED_BYTE data
+        // becoming "restart on 0" - turns the most common index in any mesh into a restart.
+        if (restartIndex > fixedMax) return RestartSubstitutionKind::SuppressRestart;
+        return RestartSubstitutionKind::RewriteIndices;
+    }
+
+    void OnRestartSubstitutionContextDestroyed() {
+        g_restartIndices = {};
+        g_restartStaging.clear();
+        g_restartStaging.shrink_to_fit();
+    }
+
+    ScopedSuppressedPrimitiveRestart::ScopedSuppressedPrimitiveRestart(RestartSubstitutionKind kind) {
+        if (kind != RestartSubstitutionKind::SuppressRestart) return;
+        // SyncRenderState turned the driver's fixed-index restart on because GL_PRIMITIVE_RESTART
+        // is enabled; for this draw's index type it would restart on a value the application
+        // never named. Toggled directly rather than through the render-state shadow, and put back
+        // in the destructor, so the shadow stays true and the next draw pays nothing.
+        g_GLESFuncs.glDisable(GL_PRIMITIVE_RESTART_FIXED_INDEX);
+        m_suppressed = true;
+    }
+
+    ScopedSuppressedPrimitiveRestart::~ScopedSuppressedPrimitiveRestart() {
+        if (!m_suppressed) return;
+        g_GLESFuncs.glEnable(GL_PRIMITIVE_RESTART_FIXED_INDEX);
+    }
+
+    ScopedRestartIndexSubstitution::ScopedRestartIndexSubstitution(GLenum indexType, GLsizei count,
+                                                                   const void* indices)
+        : m_kind(ResolveRestartSubstitution(indexType)), m_capOverride(m_kind), m_indices(indices),
+          m_indexType(indexType) {
+        if (m_kind != RestartSubstitutionKind::RewriteIndices) {
             return;
         }
-        Uint32 fixedMax = 0;
-        switch (indexType) {
-        case GL_UNSIGNED_BYTE: fixedMax = 0xFFu; break;
-        case GL_UNSIGNED_SHORT: fixedMax = 0xFFFFu; break;
-        case GL_UNSIGNED_INT: fixedMax = 0xFFFFFFFFu; break;
-        default: return;
+        const SizeT sourceIndexSize = MG_Util::GetGLTypeSize(indexType);
+        const Uint32 fixedMax = MG_Util::FixedRestartIndexForGLType(indexType);
+        const Uint32 applicationRestartIndex = MG_State::pGLContext->GetPrimitiveRestartIndex();
+        const auto& indexBuffer = BoundElementArrayBuffer();
+
+        const Uint8* source = nullptr;
+        SizeT indexCount = 0;
+        SizeT sourceByteOffset = 0;
+
+        if (indexBuffer) {
+            // The WHOLE buffer is rewritten, not just this draw's range, so that every index
+            // keeps its position: an indirect draw's firstIndex lives in GPU memory and cannot be
+            // adjusted from here. It is an ELEMENT index, so it survives widening unchanged.
+            const SizeT sizeBytes = indexBuffer->GetSize();
+            if (sizeBytes < sourceIndexSize) {
+                return; // Nothing to restart on; let the driver see the draw unchanged.
+            }
+            if (sizeBytes > kMaxRestartRewriteBytes) {
+                MGLOG_E_ONCE("Draw skipped: GL_PRIMITIVE_RESTART with restart index %u needs the %zu-byte element "
+                             "array buffer rewritten every draw, which is past the %zu-byte ceiling. Use "
+                             "GL_PRIMITIVE_RESTART_FIXED_INDEX, or set glPrimitiveRestartIndex to the all-ones "
+                             "value of the index type.",
+                             applicationRestartIndex, sizeBytes, kMaxRestartRewriteBytes);
+                m_valid = false;
+                return;
+            }
+            // The shadow is the source of truth for CPU reads, but a persistent map or a
+            // shader write may have moved past it since the last sync.
+            indexBuffer->SyncPersistentMappedRange();
+            indexBuffer->SyncGpuWrites();
+            source = indexBuffer->MappedData();
+            if (source == nullptr) {
+                MGLOG_E_ONCE("Draw skipped: GL_PRIMITIVE_RESTART with restart index %u needs a CPU-readable copy of "
+                             "the bound element array buffer and none is available.",
+                             applicationRestartIndex);
+                m_valid = false;
+                return;
+            }
+            indexCount = sizeBytes / sourceIndexSize;
+            sourceByteOffset = reinterpret_cast<SizeT>(indices);
+        } else {
+            // No element array buffer: `indices` is a client pointer, so only the draw's own
+            // range is readable and an indirect draw has nothing to read at all.
+            if (count <= 0 || indices == nullptr || sourceIndexSize == 0) {
+                MGLOG_E_ONCE("Draw skipped: GL_PRIMITIVE_RESTART with restart index %u needs either a bound element "
+                             "array buffer or a client index array with a CPU-known count.",
+                             applicationRestartIndex);
+                m_valid = false;
+                return;
+            }
+            if (static_cast<SizeT>(count) * sourceIndexSize > kMaxRestartRewriteBytes) {
+                MGLOG_E_ONCE("Draw skipped: GL_PRIMITIVE_RESTART index rewrite of %zu bytes is past the %zu-byte "
+                             "ceiling.",
+                             static_cast<SizeT>(count) * sourceIndexSize, kMaxRestartRewriteBytes);
+                m_valid = false;
+                return;
+            }
+            source = static_cast<const Uint8*>(indices);
+            indexCount = static_cast<SizeT>(count);
         }
-        const Uint32 restartIndex = MG_State::pGLContext->GetPrimitiveRestartIndex();
-        if (restartIndex != fixedMax) {
-            THROW_EXCEPTION("GL_PRIMITIVE_RESTART with an arbitrary restart index (" + std::to_string(restartIndex) +
-                            ") is not supported by the GLES backend, which only restarts on the fixed index value (" +
-                            std::to_string(fixedMax) +
-                            ") for this index type; use GL_PRIMITIVE_RESTART_FIXED_INDEX or set glPrimitiveRestartIndex "
-                            "to that value.");
+
+        // Widen only when the source really does use the all-ones value as a vertex index -
+        // otherwise the sentinel is free and the copy stays the caller's width, which keeps the
+        // common substitution allocation-for-allocation identical to the narrow form.
+        GLenum destinationType = indexType;
+        SizeT destinationIndexSize = sourceIndexSize;
+        if (ContainsFixedRestartIndex(source, indexCount, sourceIndexSize, fixedMax)) {
+            const GLenum wider = WiderIndexType(indexType);
+            // An element-array offset that is not a whole number of indices cannot be rescaled
+            // into the widened copy, so such a draw keeps the narrow (lossy) form.
+            const Bool offsetIsWholeIndices = sourceIndexSize != 0 && (sourceByteOffset % sourceIndexSize) == 0;
+            if (wider != 0 && offsetIsWholeIndices &&
+                indexCount * MG_Util::GetGLTypeSize(wider) <= kMaxRestartRewriteBytes) {
+                destinationType = wider;
+                destinationIndexSize = MG_Util::GetGLTypeSize(wider);
+            } else {
+                MGLOG_E_ONCE("GL_PRIMITIVE_RESTART with restart index %u over index data that also uses the "
+                             "all-ones index %u: this index type cannot spell both, so every all-ones index is "
+                             "drawn as %u instead. Use GL_PRIMITIVE_RESTART_FIXED_INDEX, or keep the all-ones "
+                             "value out of the index data.",
+                             applicationRestartIndex, fixedMax, fixedMax - 1);
+            }
         }
+
+        const Uint32 destinationFixedMax = MG_Util::FixedRestartIndexForGLType(destinationType);
+        RewriteRestartIndices(source, indexCount, sourceIndexSize, destinationIndexSize, applicationRestartIndex,
+                              destinationFixedMax, g_restartStaging);
+
+        if (!UploadRestartScratch(g_restartStaging.size(), g_restartStaging.data())) {
+            MGLOG_E_ONCE("Draw skipped: could not allocate the scratch element array buffer for GL_PRIMITIVE_RESTART "
+                         "index substitution.");
+            m_valid = false;
+            return;
+        }
+        m_previousBinding = BoundElementArrayBufferId();
+        BufferImpl::BindBufferId(GL_ELEMENT_ARRAY_BUFFER, g_restartIndices.id);
+        m_substituted = true;
+        m_indexType = destinationType;
+        // The rewritten copy starts at byte 0 of the scratch buffer and holds one
+        // destination-width element per source element, so an EBO-sourced draw keeps its ELEMENT
+        // offset (rescaled to the new width) and a client-memory draw reads from the front.
+        m_indices = indexBuffer
+                        ? reinterpret_cast<const void*>((sourceByteOffset / sourceIndexSize) * destinationIndexSize)
+                        : nullptr;
+    }
+
+    ScopedRestartIndexSubstitution::~ScopedRestartIndexSubstitution() {
+        if (!m_substituted) return;
+        BufferImpl::BindBufferId(GL_ELEMENT_ARRAY_BUFFER, m_previousBinding);
     }
 
     void DrawElements(GLenum mode, GLsizei count, GLenum type, const void* indices) {
@@ -3830,9 +4210,10 @@ namespace MobileGL::MG_Backend::DirectGLES {
 #endif
         DrawSyncFlags syncBit = DrawSyncBit::IndexBuffer;
         PrepareForDraw(syncBit);
-        CheckPrimitiveRestartSupported(type);
+        const ScopedRestartIndexSubstitution restart(type, count, indices);
+        if (!restart.DrawIsValid()) return;
         ForEachViewportRoutingPass([&] {
-            g_GLESFuncs.glDrawElements(mode, count, type, indices);
+            g_GLESFuncs.glDrawElements(mode, count, restart.IndexType(), restart.Indices());
         });
     }
 
@@ -3860,10 +4241,11 @@ namespace MobileGL::MG_Backend::DirectGLES {
 #endif
         DrawSyncFlags syncBit = DrawSyncBit::IndexBuffer;
         PrepareForDraw(syncBit);
-        CheckPrimitiveRestartSupported(type);
+        const ScopedRestartIndexSubstitution restart(type, count, indices);
+        if (!restart.DrawIsValid()) return;
         SetCurrentBaseVertex(basevertex);
         ForEachViewportRoutingPass([&] {
-            g_GLESFuncs.glDrawElementsBaseVertex(mode, count, type, indices, basevertex);
+            g_GLESFuncs.glDrawElementsBaseVertex(mode, count, restart.IndexType(), restart.Indices(), basevertex);
         });
         SetCurrentBaseVertex(0);
     }
@@ -4128,9 +4510,12 @@ namespace MobileGL::MG_Backend::DirectGLES {
                                      const void* indices, GLint basevertex) {
         DrawSyncFlags syncBit = DrawSyncBit::IndexBuffer;
         PrepareForDraw(syncBit);
+        const ScopedRestartIndexSubstitution restart(type, count, indices);
+        if (!restart.DrawIsValid()) return;
         SetCurrentBaseVertex(basevertex);
         ForEachViewportRoutingPass([&] {
-            g_GLESFuncs.glDrawRangeElementsBaseVertex(mode, start, end, count, type, indices, basevertex);
+            g_GLESFuncs.glDrawRangeElementsBaseVertex(mode, start, end, count, restart.IndexType(), restart.Indices(),
+                                                      basevertex);
         });
         SetCurrentBaseVertex(0);
     }
@@ -4138,8 +4523,10 @@ namespace MobileGL::MG_Backend::DirectGLES {
     void DrawRangeElements(GLenum mode, GLuint start, GLuint end, GLsizei count, GLenum type, const void* indices) {
         DrawSyncFlags syncBit = DrawSyncBit::IndexBuffer;
         PrepareForDraw(syncBit);
+        const ScopedRestartIndexSubstitution restart(type, count, indices);
+        if (!restart.DrawIsValid()) return;
         ForEachViewportRoutingPass([&] {
-            g_GLESFuncs.glDrawRangeElements(mode, start, end, count, type, indices);
+            g_GLESFuncs.glDrawRangeElements(mode, start, end, count, restart.IndexType(), restart.Indices());
         });
     }
 
@@ -4161,14 +4548,18 @@ namespace MobileGL::MG_Backend::DirectGLES {
         DrawSyncFlags syncBit = DrawSyncBit::IndexBuffer | DrawSyncBit::Instancing;
         const VertexArrayImpl::ScopedFetchBaseInstance fetchScope(EmulatedFetchBaseInstance(baseinstance));
         PrepareForDraw(syncBit);
+        const ScopedRestartIndexSubstitution restart(type, count, indices);
+        if (!restart.DrawIsValid()) return;
         SetCurrentBaseInstance(baseinstance);
         SetCurrentBaseVertex(basevertex);
         ForEachViewportRoutingPass([&] {
             if (UseNativeBaseInstance()) {
-                g_GLESFuncs.glDrawElementsInstancedBaseVertexBaseInstanceEXT(mode, count, type, indices, instancecount,
+                g_GLESFuncs.glDrawElementsInstancedBaseVertexBaseInstanceEXT(mode, count, restart.IndexType(),
+                                                                            restart.Indices(), instancecount,
                                                                             basevertex, baseinstance);
             } else {
-                g_GLESFuncs.glDrawElementsInstancedBaseVertex(mode, count, type, indices, instancecount, basevertex);
+                g_GLESFuncs.glDrawElementsInstancedBaseVertex(mode, count, restart.IndexType(), restart.Indices(),
+                                                              instancecount, basevertex);
             }
         });
         SetCurrentBaseVertex(0);
@@ -4179,9 +4570,12 @@ namespace MobileGL::MG_Backend::DirectGLES {
                                          GLsizei instancecount, GLint basevertex) {
         DrawSyncFlags syncBit = DrawSyncBit::IndexBuffer | DrawSyncBit::Instancing;
         PrepareForDraw(syncBit);
+        const ScopedRestartIndexSubstitution restart(type, count, indices);
+        if (!restart.DrawIsValid()) return;
         SetCurrentBaseVertex(basevertex);
         ForEachViewportRoutingPass([&] {
-            g_GLESFuncs.glDrawElementsInstancedBaseVertex(mode, count, type, indices, instancecount, basevertex);
+            g_GLESFuncs.glDrawElementsInstancedBaseVertex(mode, count, type, restart.Indices(), instancecount,
+                                                          basevertex);
         });
         SetCurrentBaseVertex(0);
     }
@@ -4191,13 +4585,15 @@ namespace MobileGL::MG_Backend::DirectGLES {
         DrawSyncFlags syncBit = DrawSyncBit::IndexBuffer | DrawSyncBit::Instancing;
         const VertexArrayImpl::ScopedFetchBaseInstance fetchScope(EmulatedFetchBaseInstance(baseinstance));
         PrepareForDraw(syncBit);
+        const ScopedRestartIndexSubstitution restart(type, count, indices);
+        if (!restart.DrawIsValid()) return;
         SetCurrentBaseInstance(baseinstance);
         ForEachViewportRoutingPass([&] {
             if (UseNativeBaseInstance()) {
-                g_GLESFuncs.glDrawElementsInstancedBaseInstanceEXT(mode, count, type, indices, instancecount,
-                                                                  baseinstance);
+                g_GLESFuncs.glDrawElementsInstancedBaseInstanceEXT(mode, count, restart.IndexType(), restart.Indices(),
+                                                                  instancecount, baseinstance);
             } else {
-                g_GLESFuncs.glDrawElementsInstanced(mode, count, type, indices, instancecount);
+                g_GLESFuncs.glDrawElementsInstanced(mode, count, restart.IndexType(), restart.Indices(), instancecount);
             }
         });
         SetCurrentBaseInstance(0);
@@ -4206,8 +4602,10 @@ namespace MobileGL::MG_Backend::DirectGLES {
     void DrawElementsInstanced(GLenum mode, GLsizei count, GLenum type, const void* indices, GLsizei instancecount) {
         DrawSyncFlags syncBit = DrawSyncBit::IndexBuffer | DrawSyncBit::Instancing;
         PrepareForDraw(syncBit);
+        const ScopedRestartIndexSubstitution restart(type, count, indices);
+        if (!restart.DrawIsValid()) return;
         ForEachViewportRoutingPass([&] {
-            g_GLESFuncs.glDrawElementsInstanced(mode, count, type, indices, instancecount);
+            g_GLESFuncs.glDrawElementsInstanced(mode, count, restart.IndexType(), restart.Indices(), instancecount);
         });
     }
 
@@ -5112,6 +5510,164 @@ namespace MobileGL::MG_Backend::DirectGLES {
         DrainBlitErrors();
     }
 
+    // ---- glBlitFramebuffer onto a non-zero array layer -------------------------------------
+    //
+    // Some drivers write to layer 0 whatever layer the DRAW framebuffer's
+    // glFramebufferTextureLayer attachment names, and raise no error doing it (Adreno 830;
+    // SelfTest::ProbeBlitIgnoresDestinationArrayLayer measures it, with the destination-layer-0
+    // case as the control). glCopyImageSubData takes the destination layer as an argument rather
+    // than reading it off an attachment, and honours it on the same driver - so a blit that is a
+    // plain 1:1 copy is issued that way instead.
+    //
+    // ONLY a plain 1:1 copy. glCopyImageSubData cannot scale, flip, convert format or resolve
+    // samples, and it is not clipped by the scissor, so every one of those is a reason to hand
+    // the call back to the driver rather than quietly perform a different operation. Those blits
+    // still land on the wrong layer; a once-per-process line says so rather than leaving it to be
+    // rediscovered.
+    //
+    // Per aspect, not all-or-nothing: the returned mask is the bits this performed itself, and
+    // the caller passes the rest to the driver. A COLOR|DEPTH blit whose colour half scales and
+    // whose depth half does not still gets its depth half repaired.
+    static GLbitfield BlitLayeredDestinationAspects(
+        const SharedPtr<MG_State::GLState::FramebufferObject>& readFramebuffer,
+        const SharedPtr<MG_State::GLState::FramebufferObject>& drawFramebuffer, GLint srcX0, GLint srcY0,
+        GLint srcX1, GLint srcY1, GLint dstX0, GLint dstY0, GLint dstX1, GLint dstY1, GLbitfield mask) {
+        if (mask == 0 || !readFramebuffer || !drawFramebuffer) return 0;
+        if (!g_GLESFuncs.glCopyImageSubData) return 0;
+        if (!MG_Util::SelfTest::BlitIgnoresDestinationArrayLayer(g_GLESFuncs)) return 0;
+
+        // The default framebuffer has no layers to get wrong, and a blit between the two halves
+        // of the same framebuffer object is not a shape this substitutes for.
+        const Int width = srcX1 - srcX0;
+        const Int height = srcY1 - srcY0;
+        const Bool oneToOne = width > 0 && height > 0 && (dstX1 - dstX0) == width && (dstY1 - dstY0) == height;
+        // The scissor clips a blit and does not clip a copy, so an enabled scissor makes the two
+        // different operations no matter how the rectangles line up.
+        const Bool scissorEnabled =
+            (RenderStateImpl::g_syncedRenderStateParameters.ScissorTestEnabledMask & 1u) != 0;
+
+        using MobileGL::FramebufferAttachmentType;
+        struct AspectPlan {
+            GLbitfield bit;
+            FramebufferAttachmentType source;
+            FramebufferAttachmentType destination;
+        };
+        // The colour aspect follows glReadBuffer on the read side and draw buffer 0 on the write
+        // side, which is the only draw buffer a blit onto a layered destination can be pinned to
+        // here: a blit writes EVERY enabled draw buffer, so a framebuffer with more than one is
+        // left to the driver rather than half-repaired.
+        const auto& drawBuffers = drawFramebuffer->GetDrawBuffers();
+        // GL_NONE is what an unwritten draw-buffer slot holds, and it is a different value from
+        // the "no such attachment" one - counting it as enabled made every framebuffer look like
+        // it had eight and sent every colour blit to the driver.
+        //
+        // The buffer is found rather than assumed to be slot 0: a blit writes every ENABLED draw
+        // buffer, and glDrawBuffers(GL_NONE, GL_NONE, GL_NONE, GL_COLOR_ATTACHMENT0) leaves slot
+        // 0 empty while still naming exactly one destination.
+        Int enabledDrawBuffers = 0;
+        FramebufferAttachmentType colorDestination = FramebufferAttachmentType::None;
+        for (const FramebufferAttachmentType buffer : drawBuffers) {
+            if (buffer != FramebufferAttachmentType::Unknown && buffer != FramebufferAttachmentType::None) {
+                ++enabledDrawBuffers;
+                if (enabledDrawBuffers == 1) colorDestination = buffer;
+            }
+        }
+        const AspectPlan plans[] = {
+            {GL_COLOR_BUFFER_BIT, readFramebuffer->GetReadBuffer(), colorDestination},
+            {GL_DEPTH_BUFFER_BIT, FramebufferAttachmentType::Depth, FramebufferAttachmentType::Depth},
+            {GL_STENCIL_BUFFER_BIT, FramebufferAttachmentType::Stencil, FramebufferAttachmentType::Stencil},
+        };
+
+        GLbitfield handled = 0;
+        for (const AspectPlan& plan : plans) {
+            if ((mask & plan.bit) == 0) continue;
+            if (plan.source == FramebufferAttachmentType::Unknown ||
+                plan.destination == FramebufferAttachmentType::Unknown ||
+                plan.source == FramebufferAttachmentType::None ||
+                plan.destination == FramebufferAttachmentType::None) {
+                continue;
+            }
+            const auto& sourceAttachment = readFramebuffer->GetAttachment(plan.source);
+            const auto& destinationAttachment = drawFramebuffer->GetAttachment(plan.destination);
+            // Renderbuffers have no layers, so a destination that is one cannot be hitting this.
+            if (!sourceAttachment.IsTexture() || !destinationAttachment.IsTexture()) continue;
+            // Layer 0 is the case the driver gets right, and a LAYERED attachment (glFramebufferTexture
+            // with no layer) blits its layer 0 by spec - neither is this defect.
+            if (destinationAttachment.GetTextureLayer() == 0) continue;
+            if (destinationAttachment.IsLayered() || sourceAttachment.IsLayered()) continue;
+
+            const auto& sourceTexture = sourceAttachment.GetTexture();
+            const auto& destinationTexture = destinationAttachment.GetTexture();
+            if (!sourceTexture || !destinationTexture) continue;
+            // glCopyImageSubData moves texel blocks: same format both ends, or it is a different
+            // operation. Multisample endpoints would additionally have to agree on sample count,
+            // which is a resolve the driver still owns.
+            if (sourceTexture->GetFormat() != destinationTexture->GetFormat()) continue;
+            if (sourceTexture->GetSamples() > 0 || destinationTexture->GetSamples() > 0) continue;
+            // Copying an image region onto itself is undefined for glCopyImageSubData, and a blit
+            // whose source and destination overlap is undefined for GL too - so this is not a
+            // shape to substitute FOR, it is one to leave exactly as the application wrote it.
+            if (sourceTexture == destinationTexture &&
+                sourceAttachment.GetTextureLevel() == destinationAttachment.GetTextureLevel() &&
+                sourceAttachment.GetTextureLayer() == destinationAttachment.GetTextureLayer()) {
+                continue;
+            }
+
+            // A combined depth-stencil texture is ONE image to glCopyImageSubData: it carries both
+            // aspects across whether or not the mask asked for both. Taking only GL_DEPTH_BUFFER_BIT
+            // on a DEPTH24_STENCIL8 destination would overwrite a stencil the application asked to
+            // keep, so the copy is only allowed when the mask covers everything the format holds.
+            const TextureInternalFormat format = destinationTexture->GetFormat();
+            const Bool hasDepth = MG_Util::IsDepthFormatInternalFormat(format);
+            const Bool hasStencil = MG_Util::IsStencilFormatInternalFormat(format);
+            if (hasDepth && (mask & GL_DEPTH_BUFFER_BIT) == 0) continue;
+            if (hasStencil && (mask & GL_STENCIL_BUFFER_BIT) == 0) continue;
+            // ... and having carried both, it must be credited with both, or the caller hands the
+            // stencil half to the driver and it lands on layer 0 after all.
+            const GLbitfield aspectBits =
+                hasDepth || hasStencil
+                    ? static_cast<GLbitfield>((hasDepth ? GL_DEPTH_BUFFER_BIT : 0) |
+                                              (hasStencil ? GL_STENCIL_BUFFER_BIT : 0))
+                    : static_cast<GLbitfield>(GL_COLOR_BUFFER_BIT);
+            if ((handled & aspectBits) == aspectBits) continue;
+
+            if (!oneToOne || scissorEnabled || (plan.bit == GL_COLOR_BUFFER_BIT && enabledDrawBuffers != 1)) {
+                MGLOG_E_ONCE("BlitFramebuffer: this driver ignores a non-zero destination array layer and this "
+                             "blit cannot be expressed as a copy (%s), so it will land on layer 0",
+                             !oneToOne          ? "it scales or flips"
+                             : scissorEnabled   ? "the scissor test is enabled"
+                                                : "the destination has more than one draw buffer");
+                continue;
+            }
+
+            auto backendSource = TextureImpl::SyncTextureObjectToBackend(sourceTexture);
+            auto backendDestination = TextureImpl::SyncTextureObjectToBackend(destinationTexture);
+            if (!backendSource || !backendDestination) continue;
+            const GLuint sourceName = backendSource->GetBackendTextureId();
+            const GLuint destinationName = backendDestination->GetBackendTextureId();
+            if (sourceName == 0 || destinationName == 0) continue;
+            const GLenum sourceTarget = TextureImpl::ConvertTextureTargetToBackendGLEnum(sourceTexture->GetTarget());
+            const GLenum destinationTarget =
+                TextureImpl::ConvertTextureTargetToBackendGLEnum(destinationTexture->GetTarget());
+
+            ClearGLErrors();
+            g_GLESFuncs.glCopyImageSubData(sourceName, sourceTarget, sourceAttachment.GetTextureLevel(), srcX0, srcY0,
+                                           sourceAttachment.GetTextureLayer(), destinationName, destinationTarget,
+                                           destinationAttachment.GetTextureLevel(), dstX0, dstY0,
+                                           destinationAttachment.GetTextureLayer(), width, height, 1);
+            if (const GLenum error = g_GLESFuncs.glGetError(); error != GL_NO_ERROR) {
+                // The driver blit still runs for this aspect - onto the wrong layer, but the
+                // substitute has to leave the call no worse off than it found it.
+                MGLOG_E_ONCE("BlitFramebuffer: the layered-destination copy substitute failed with %s; the "
+                             "driver blit will run instead and land on layer 0",
+                             MG_Util::ConvertGLEnumToString(error).c_str());
+                continue;
+            }
+            handled |= aspectBits;
+        }
+        return handled & mask;
+    }
+
     void BlitFramebuffer(GLint srcX0, GLint srcY0, GLint srcX1, GLint srcY1, GLint dstX0, GLint dstY0, GLint dstX1,
                          GLint dstY1, GLbitfield mask, GLenum filter) {
 #if MOBILEGL_LOG_ACTIVE_LEVEL <= MOBILEGL_LOG_LEVEL_DEBUG && MOBILEGL_ENABLE_SCOPE_MARKER
@@ -5138,7 +5694,15 @@ namespace MobileGL::MG_Backend::DirectGLES {
         });
         MGLOG_D("ES %s(%d, %d, %d, %d, %d, %d, %d, %d, 0x%x, %s)", __func__, srcX0, srcY0, srcX1, srcY1, dstX0, dstY0,
                 dstX1, dstY1, mask, MG_Util::ConvertGLEnumToString(filter).c_str());
-        IssueBlitWithResolveFallback(srcX0, srcY0, srcX1, srcY1, dstX0, dstY0, dstX1, dstY1, mask, filter);
+        // A no-op on every driver that honours a non-zero destination array layer, which is all
+        // of them but the probed one. Whatever it performs itself is taken out of the mask.
+        mask &= ~BlitLayeredDestinationAspects(
+            MG_State::pGLContext->GetFramebufferBindingSlot(FramebufferTarget::Read).GetBoundObject(),
+            MG_State::pGLContext->GetFramebufferBindingSlot(FramebufferTarget::Draw).GetBoundObject(), srcX0, srcY0,
+            srcX1, srcY1, dstX0, dstY0, dstX1, dstY1, mask);
+        if (mask != 0) {
+            IssueBlitWithResolveFallback(srcX0, srcY0, srcX1, srcY1, dstX0, dstY0, dstX1, dstY1, mask, filter);
+        }
         DebugImpl::ErrorLopper::Loop([file = __FILE__, line = __LINE__](auto err) {
             MGLOG_D("ES error (%s:%d): %s", file, line, MG_Util::ConvertGLEnumToString(err).c_str());
         });
@@ -5160,7 +5724,12 @@ namespace MobileGL::MG_Backend::DirectGLES {
 
         MGLOG_D("ES %s(%d, %d, %d, %d, %d, %d, %d, %d, 0x%x, %s)", __func__, srcX0, srcY0, srcX1, srcY1,
                 dstX0, dstY0, dstX1, dstY1, mask, MG_Util::ConvertGLEnumToString(filter).c_str());
-        IssueBlitWithResolveFallback(srcX0, srcY0, srcX1, srcY1, dstX0, dstY0, dstX1, dstY1, mask, filter);
+        // See the DSA-free entry point above: only the probed defect makes this do anything.
+        mask &= ~BlitLayeredDestinationAspects(readFramebuffer, drawFramebuffer, srcX0, srcY0, srcX1, srcY1, dstX0,
+                                               dstY0, dstX1, dstY1, mask);
+        if (mask != 0) {
+            IssueBlitWithResolveFallback(srcX0, srcY0, srcX1, srcY1, dstX0, dstY0, dstX1, dstY1, mask, filter);
+        }
         // Debug-only diagnostics: which GLES depth texture did this blit write?
 #if MOBILEGL_LOG_ACTIVE_LEVEL <= MOBILEGL_LOG_LEVEL_DEBUG
         if (mask & GL_DEPTH_BUFFER_BIT) {
@@ -9784,9 +10353,10 @@ namespace MobileGL::MG_Backend::DirectGLES {
             g_completedFrameSerial.store(completed, std::memory_order_relaxed);
         }
 
-        // After the watermark advanced: retire grown-away UBO-ring stores and record
-        // the frame's ring high-water mark for slot reclamation.
+        // After the watermark advanced: retire grown-away ring stores and record the
+        // frame's ring high-water marks for slot reclamation.
         BufferImpl::UboRingOnPresent();
+        BufferImpl::UnpackRingOnPresent();
         BufferImpl::TrimBufferPool();
     }
 
@@ -9794,6 +10364,7 @@ namespace MobileGL::MG_Backend::DirectGLES {
         BufferImpl::OnBackendContextDestroyed();
         XfbImpl::OnBackendContextDestroyed();
         MultiDrawImpl::OnBackendContextDestroyed();
+        OnRestartSubstitutionContextDestroyed();
         ScratchFBOImpl::OnBackendContextDestroyed();
         ReleasePackedWordScratchTexture();
         FramebufferImpl::InvalidateFramebufferBindingCache();

MobileGL/MG_Backend/DirectGLES/Managers.h

@@ -102,9 +102,98 @@ namespace MobileGL::MG_Backend::DirectGLES {
     // Brings the whole draw-relevant frontend state onto the native ES context and binds
     // the program; every GL draw entry point calls it exactly once before issuing draws.
     void PrepareForDraw(DrawSyncFlags syncBits);
-    // GLES core supports only GL_PRIMITIVE_RESTART_FIXED_INDEX. Throws when the app enabled
-    // the arbitrary GL_PRIMITIVE_RESTART with a non-fixed index for this index type.
-    void CheckPrimitiveRestartSupported(GLenum indexType);
+    // What an indexed draw has to do about primitive restart before it can be issued.
+    //
+    // Desktop GL restarts on an application-chosen index (glPrimitiveRestartIndex under
+    // GL_PRIMITIVE_RESTART); GLES core restarts only on the all-ones value of the index type
+    // (GL_PRIMITIVE_RESTART_FIXED_INDEX), which the render-state push enables for BOTH caps.
+    // That leaves three cases, and the difference between the last two is not cosmetic - one
+    // adds restarts, the other has to take away restarts the driver would otherwise make.
+    enum class RestartSubstitutionKind : Uint8 {
+        // Nothing to do: restart is off, the fixed-index cap is on, or the application's
+        // restart index already IS the type's all-ones value. The overwhelmingly common answer.
+        None,
+        // The application's index is representable in this index type and differs from the
+        // all-ones value: the index DATA has to be rewritten so the driver restarts where the
+        // application asked.
+        RewriteIndices,
+        // The application's index cannot be held by this index type at all. GL 4.6 core 10.3.6
+        // compares the fetched index, zero-extended, against the full 32-bit
+        // PRIMITIVE_RESTART_INDEX, so no index can match and the draw restarts NOWHERE - but the
+        // render-state push has already enabled the driver's fixed-index restart, so the
+        // all-ones value has to be un-restarted for the duration of the draw.
+        SuppressRestart,
+    };
+    RestartSubstitutionKind ResolveRestartSubstitution(GLenum indexType);
+
+    // Turns the driver's fixed-index restart off for one draw and back on afterwards, for the
+    // SuppressRestart case above. Separate from the substitution below because the multi-draw
+    // tiers need it on its own: they rewrite the index stream themselves and only ever need the
+    // cap half. Inert for every other kind, and it never touches the render-state shadow - it
+    // puts the driver back exactly where SyncRenderState left it.
+    class ScopedSuppressedPrimitiveRestart {
+    public:
+        explicit ScopedSuppressedPrimitiveRestart(RestartSubstitutionKind kind);
+        ~ScopedSuppressedPrimitiveRestart();
+        ScopedSuppressedPrimitiveRestart(const ScopedSuppressedPrimitiveRestart&) = delete;
+        ScopedSuppressedPrimitiveRestart& operator=(const ScopedSuppressedPrimitiveRestart&) = delete;
+
+    private:
+        Bool m_suppressed = false;
+    };
+
+    // Swaps in a scratch element array buffer holding a copy of the index data in which the
+    // application's restart index has been replaced by the value GLES restarts on. Inert
+    // (and free) unless ResolveRestartSubstitution asks for it. The swap lives for the
+    // object's lifetime, so it covers every pass of a viewport-routed draw, and the previous
+    // GL_ELEMENT_ARRAY_BUFFER name is restored on destruction - which matters beyond tidiness,
+    // because the VAO twin memoises that it already synced that binding.
+    //
+    // The copy may be WIDER than the source (see IndexType): when the source already contains
+    // the type's all-ones value as an ordinary vertex index, that value cannot double as the
+    // restart sentinel, and widening is the only way to keep both meanings. Callers must
+    // therefore take the index type from this object, not from their own argument.
+    class ScopedRestartIndexSubstitution {
+    public:
+        // count/indices describe the draw's index range when the CPU knows it. Pass
+        // count == 0 for an indirect draw, whose count lives in GPU memory: the whole bound
+        // element array buffer is rewritten instead, so every element keeps its position and
+        // a GPU-resident firstIndex - an ELEMENT index, so it survives widening too - still
+        // addresses the index it named.
+        ScopedRestartIndexSubstitution(GLenum indexType, GLsizei count, const void* indices);
+        ~ScopedRestartIndexSubstitution();
+        ScopedRestartIndexSubstitution(const ScopedRestartIndexSubstitution&) = delete;
+        ScopedRestartIndexSubstitution& operator=(const ScopedRestartIndexSubstitution&) = delete;
+
+        // False only when a substitution was needed and could not be made. The draw must
+        // then be skipped: issuing it would let the driver silently drop every restart and
+        // weld the primitives on either side together, which is worse than drawing nothing.
+        Bool DrawIsValid() const { return m_valid; }
+        // The element-array offset (or client pointer) the draw must use. Identical to what
+        // was passed in unless a substitution was made.
+        const void* Indices() const { return m_indices; }
+        // The index type the draw must be issued with. Identical to the constructor's unless
+        // the copy had to be widened to keep an all-ones vertex index distinguishable from the
+        // restart sentinel.
+        GLenum IndexType() const { return m_indexType; }
+
+    private:
+        // Declared before m_capOverride so it is initialised first (members initialise in
+        // declaration order): the whole decision is made once, and both the cap override and the
+        // constructor body read the same answer.
+        RestartSubstitutionKind m_kind = RestartSubstitutionKind::None;
+        ScopedSuppressedPrimitiveRestart m_capOverride;
+        const void* m_indices = nullptr;
+        GLenum m_indexType = 0;
+        Uint m_previousBinding = 0;
+        Bool m_substituted = false;
+        Bool m_valid = true;
+    };
+
+    // Drops the scratch element array buffer the substitution above stages through. Like
+    // MultiDrawImpl's scratch names it is abandoned rather than deleted: the name belongs to
+    // the dead ES context, and deleting it would target whatever its successor handed out.
+    void OnRestartSubstitutionContextDestroyed();
     // Feed the current program's gl_BaseInstance / gl_DrawID / gl_BaseVertex emulation
     // uniforms. All are no-ops when the program does not read the corresponding builtin.
     void SetCurrentBaseInstance(Uint32 baseInstance);
@@ -510,6 +599,40 @@ namespace MobileGL::MG_Backend::DirectGLES {
         // Present()-time upkeep: records the frame's high-water mark for reclamation
         // and deletes grown-away ring stores once the GPU is done with them.
         void UboRingOnPresent();
+
+        // --- Texture unpack-PBO ring ----------------------------------------------
+        // The same persistent-mapped bump allocator, staging TEXTURE UPLOADS. A
+        // glTexSubImage from client memory hands the driver a pointer it must read
+        // before the call returns, so the copy has to be ordered against whatever GPU
+        // work still reads the destination texture: Mali resolves that by BLOCKING the
+        // calling thread (osup_sync_object_wait) instead of ghosting, and Minecraft
+        // re-uploads animated atlas sprites and the lightmap every tick into textures
+        // the in-flight frame is still sampling. Staging the bytes into a
+        // GPU-visible unpack PBO and passing an OFFSET instead lets the driver queue
+        // the copy in the command stream with no CPU wait at all.
+        //
+        // Same reclamation contract as the UBO ring: no ring bytes are recycled before
+        // the frame that referenced them completed on the GPU, so a staged block stays
+        // intact for as long as the queued transfer can still be reading it. The store
+        // therefore settles at roughly (bytes staged per frame) x (frames in flight),
+        // which is what to watch if this ring ever shows up in an RSS regression: it
+        // grows on demand from 4 MiB and is capped, not unbounded.
+        //
+        // False when the feature is disabled (MOBILEGL_DISABLE_UNPACK_RING),
+        // EXT_buffer_storage / fences are missing, the ES context is not current, or
+        // ring creation already failed under this context. Callers then upload from
+        // the client pointer exactly as before.
+        Bool UnpackRingAvailable();
+        // Bump-allocate `size` bytes aligned to 64 (a PBO-sourced glTexSubImage only
+        // owes the driver the pixel type's own alignment). Grows the ring when the
+        // in-flight span would be overrun; false when the request exceeds the ring's
+        // size cap or storage (re)creation fails.
+        Bool UnpackRingAllocate(SizeT size, SizeT& outOffset);
+        void* UnpackRingMappedPtr();
+        Uint UnpackRingBufferId();
+        // Largest single staging request the ring can ever satisfy.
+        SizeT UnpackRingMaxBytes();
+        void UnpackRingOnPresent();
     } // namespace BufferImpl
 
     namespace VertexArrayImpl {
@@ -792,6 +915,18 @@ namespace MobileGL::MG_Backend::DirectGLES {
             BackendTextureObject(const BackendTextureObject&) = delete;
             BackendTextureObject& operator=(const BackendTextureObject&) = delete;
             void SyncMipmapsToBackend(const SharedPtr<MG_State::GLState::ITextureObject>& stateTextureObject);
+            // The storage half of the sync for a texture created by glTextureView. Instead of
+            // allocating storage and replaying uploads, it makes this object's ES name BE a view
+            // of the storage texture's ES name (EXT/OES_texture_view), which is what gives the
+            // two names one image and independent per-texture parameters at the same time. The
+            // parameter and sampler halves are unchanged and run on this name as on any other.
+            void SyncTextureViewToBackend(const SharedPtr<MG_State::GLState::ITextureObject>& stateTextureObject);
+            void StampViewSyncKeys(const SharedPtr<MG_State::GLState::ITextureObject>& stateTextureObject);
+            // The storage half of the sync for a texture created by glTextureView. Instead of
+            // allocating storage and replaying uploads, it makes this object's ES name BE a view
+            // of the storage texture's ES name (EXT/OES_texture_view), which is what gives the
+            // two names one image and independent per-texture parameters at the same time. The
+            // parameter and sampler halves are unchanged and run on this name as on any other.
             void SyncBuiltinSamplerToBackend(const SharedPtr<MG_State::GLState::ITextureObject>& stateTextureObject);
             void SyncTextureParamsToBackend(const SharedPtr<MG_State::GLState::ITextureObject>& stateTextureObject);
             // Marks the texture as one whose ES storage has to be image-bindable, which for a
@@ -824,6 +959,10 @@ namespace MobileGL::MG_Backend::DirectGLES {
             // `contextId`/`samplingGeneration` are the frontend context's current
             // values, hoisted by the caller so a per-draw list walk reads them once
             // instead of per texture. `t` must be the live frontend texture.
+            // True while a driver-side re-mint has left the parameter caches describing a texture
+            // that no longer exists; SyncTextureObjectToBackend re-pushes them in the same sync.
+            Bool NeedsParameterResync() const { return m_forceTextureParamsResync || m_forceSamplerResync; }
+
             Bool IsDrawSyncClean(const MG_State::GLState::ITextureObject* t, Uint64 contextId,
                                  Uint64 samplingGeneration) const {
                 if (!m_isInitialized || m_syncedShapeContextId == 0 || m_syncedShapeContextId != contextId ||
@@ -867,6 +1006,17 @@ namespace MobileGL::MG_Backend::DirectGLES {
             // it is only the IMAGE binding ES cannot spell - and the private name below carries
             // the split the shader was rewritten against. 0 when this texture takes no split.
             Uint m_bufferImageSplitViewId = 0;
+            // For a texture created by glTextureView: the ES name of the storage texture this
+            // one was last made a view OF. EXT_texture_view may be called only once per name, so
+            // a storage texture that got re-minted underneath (RecreateBackendTexture) has to be
+            // detected here and answered with a fresh name for the view as well - otherwise the
+            // view would keep aliasing storage that no longer exists.
+            Uint m_viewSourceBackendTextureId = 0;
+            // For a texture created by glTextureView: the ES name of the storage texture this
+            // one was last made a view OF. EXT_texture_view may be called only once per name, so
+            // a storage texture that got re-minted underneath (RecreateBackendTexture) has to be
+            // detected here and answered with a fresh name for the view as well - otherwise the
+            // view would keep aliasing storage that no longer exists.
             // ES context generation the id was created under; a dtor running after
             // that context died must not delete a foreign (recycled) name.
             Uint m_contextGeneration = 0;
@@ -900,7 +1050,14 @@ namespace MobileGL::MG_Backend::DirectGLES {
             Uint16 m_syncedShapeParamsVersion = 0;
             SamplerParameters m_cacheSamplerParameters;
             UintVec2 m_cacheLodRange = {0, 1000};
+            // All three representations plus the form, because none of them alone identifies the
+            // border colour the driver texture is holding: two integer borders can share one float
+            // (anything differing above 2^24), and a Float -> Int transition can leave every number
+            // unchanged while still needing a different driver entry point.
             FloatVec4 m_cacheBorderColor = {0.0f, 0.0f, 0.0f, 0.0f};
+            IntVec4 m_cacheBorderColorI = {0, 0, 0, 0};
+            UintVec4 m_cacheBorderColorUI = {0, 0, 0, 0};
+            BorderColorForm m_cacheBorderColorForm = BorderColorForm::Float;
             Vec4<TextureSwizzleParam> m_cacheSwizzleParams = {TextureSwizzleParam::Red, TextureSwizzleParam::Green,
                                                               TextureSwizzleParam::Blue, TextureSwizzleParam::Alpha};
             // GL_DEPTH_STENCIL_TEXTURE_MODE. GL_DEPTH_COMPONENT is the GL and ES default, so a
@@ -915,6 +1072,15 @@ namespace MobileGL::MG_Backend::DirectGLES {
             // parameter already pushed onto it: the params-version early-out has to be overridden
             // once, or an unchanged version would skip the re-push forever.
             Bool m_forceTextureParamsResync = false;
+            // The same problem for the FILTER state, which lives in m_cacheSamplerParameters and
+            // is gated on the frontend sampler's version rather than on the params version. A
+            // re-mint leaves that cache describing values the new driver texture never received,
+            // and an unchanged sampler version would then skip re-pushing them forever. This
+            // matters more than mis-filtering: ES makes a texture INCOMPLETE when its filters do
+            // not suit its level set (any integer texture with a non-NEAREST filter, or a
+            // single-level texture with a mipmapping filter), and an incomplete texture samples
+            // (0, 0, 0, 1) rather than its contents.
+            Bool m_forceSamplerResync = false;
         };
 
         void ActivateTextureUnit(Uint unit);
@@ -1377,6 +1543,17 @@ namespace MobileGL::MG_Backend::DirectGLES {
             Int GetPassthroughTessControlPatchVertices() const {
                 return m_passthroughTessControlPatchVertices;
             }
+            // GL_PATCH_DEFAULT_{OUTER,INNER}_LEVEL the same synthesized stage was built with, for
+            // the same reason: ES has neither the state nor an entry point to forward it to, so
+            // glPatchParameterfv's values are compiled in as literals and a program built with one
+            // set is stale for another. Meaningless (and never read) when the patch-vertices field
+            // above is -1, which is the gate the draw path tests first.
+            const FloatVec4& GetPassthroughTessControlOuterLevel() const {
+                return m_passthroughTessControlOuterLevel;
+            }
+            const FloatVec2& GetPassthroughTessControlInnerLevel() const {
+                return m_passthroughTessControlInnerLevel;
+            }
 
             Bool HasGlobalUboBlock() const { return m_globalUboBackendBlockIndex >= 0; }
             const Vector<Int>& GetUniformBlockBackendIndices() const { return m_uniformBlockBackendIndices; }
@@ -1477,6 +1654,10 @@ namespace MobileGL::MG_Backend::DirectGLES {
             // all); otherwise the GL_PATCH_VERTICES the synthesized pass-through stage was built
             // with. See GetPassthroughTessControlPatchVertices.
             Int m_passthroughTessControlPatchVertices = -1;
+            // The default tessellation levels baked into that same stage. Only meaningful while
+            // the field above is not -1.
+            FloatVec4 m_passthroughTessControlOuterLevel = FloatVec4(1.0f, 1.0f, 1.0f, 1.0f);
+            FloatVec2 m_passthroughTessControlInnerLevel = FloatVec2(1.0f, 1.0f);
             Bool m_isInitialized = false;
             Bool m_backendProgramUsable = false;
             // Set by SyncToBackend every time it relinks the driver program, cleared by the

MobileGL/MG_Backend/DirectGLES/MultiDraw.cpp

@@ -29,16 +29,21 @@ namespace MobileGL::MG_Backend::DirectGLES::MultiDrawImpl {
             }
         }
 
-        // The all-ones value of an index type, which is what GL restarts on once
-        // primitive restart is in play. CheckPrimitiveRestartSupported has already
-        // rejected the arbitrary-index form of GL_PRIMITIVE_RESTART, so an enabled
-        // restart always restarts here and nowhere else.
+        // The index value this batch restarts on, compared at 32 bits against the zero-extended
+        // source index. Normally the all-ones value of the source type, which is what
+        // GL_PRIMITIVE_RESTART_FIXED_INDEX and GLES both restart on; with desktop
+        // GL_PRIMITIVE_RESTART it is instead whatever glPrimitiveRestartIndex named. The rebased
+        // tier turns whichever it is into 0xFFFFFFFF in its widened stream, which is what the
+        // driver restarts on.
+        //
+        // No truncation, deliberately, and the same rule ResolveRestartSubstitution applies: a
+        // restart index the source type cannot hold simply matches nothing, so returning it
+        // verbatim is already "this batch restarts nowhere".
         Uint32 RestartSentinelFor(GLenum type) {
-            switch (type) {
-            case GL_UNSIGNED_BYTE: return 0xFFu;
-            case GL_UNSIGNED_SHORT: return 0xFFFFu;
-            default: return 0xFFFFFFFFu;
+            if (ResolveRestartSubstitution(type) != RestartSubstitutionKind::None) {
+                return MG_State::pGLContext->GetPrimitiveRestartIndex();
             }
+            return MG_Util::FixedRestartIndexForGLType(type);
         }
 
         Bool RestartActive() {
@@ -275,10 +280,20 @@ namespace MobileGL::MG_Backend::DirectGLES::MultiDrawImpl {
         // its remaining feasibility checks inside its implementation, where the data it
         // has to walk is already in hand.
         GLESMultiDrawMode ResolveTierForBatch(Bool programReadsDrawID, Bool perSubDrawBaseVertex,
-                                              Bool hasIndexBuffer) {
+                                              Bool hasIndexBuffer, Bool arbitraryRestart) {
             ResolveTierOnce();
             GLESMultiDrawMode tier = g_resolvedTier;
 
+            // Desktop GL_PRIMITIVE_RESTART restarts on an application-chosen index; the driver
+            // only ever restarts on the all-ones value. Every tier but the rebased one hands
+            // the application's own index data to the driver, which would then see no restarts
+            // at all and weld the primitives together. The rebased tier is the one that
+            // REWRITES the stream, and RestartSentinelFor already tells it which value to
+            // translate, so it is the only tier this batch can take.
+            if (arbitraryRestart) {
+                return GLESMultiDrawMode::DrawElements;
+            }
+
             // Batched tiers issue one driver entry for the whole batch, so the emulated
             // gl_DrawID uniform can only hold one value across every sub-draw. A program
             // that reads gl_DrawID gets an unrolled tier, which feeds each sub-draw its
@@ -488,6 +503,16 @@ namespace MobileGL::MG_Backend::DirectGLES::MultiDrawImpl {
 
             const Bool restartActive = RestartActive();
             const Uint32 restartSentinel = RestartSentinelFor(type);
+            // Widening to GL_UNSIGNED_INT gives a UBYTE/USHORT source a sentinel it can never
+            // spell, so those batches are lossless. A UINT source that already uses 0xFFFFFFFF as
+            // a real vertex index while restarting on a different one is the one shape 32 bits
+            // cannot express - the same corner the single-draw substitution reports.
+            if (restartActive && indexSize == 4 && restartSentinel != 0xFFFFFFFFu) {
+                MGLOG_E_ONCE("GL_PRIMITIVE_RESTART with restart index %u over GL_UNSIGNED_INT multi-draw indices: "
+                             "any index that is already 0xFFFFFFFF will restart too, because the rewritten stream "
+                             "has no wider sentinel to move to.",
+                             restartSentinel);
+            }
             g_indexStaging.resize(total);
             SizeT cursor = 0;
             for (GLsizei i = 0; i < drawcount; ++i) {
@@ -852,8 +877,14 @@ void main() {
     void DrawElementsBatch(GLenum mode, const GLsizei* count, GLenum type, const GLvoid* const* indices,
                            GLsizei drawcount, const GLint* basevertex) {
         if (drawcount <= 0 || !count || !indices) return;
-        // State-independent and possibly throwing, so it runs before any GL work.
-        CheckPrimitiveRestartSupported(type);
+        // Read before any GL work, because it decides the tier below: a desktop restart index
+        // the driver does not know about can only be honoured by the tier that rewrites the
+        // index stream (see ResolveTierForBatch). A restart index this index type cannot hold
+        // needs no rewrite at all - nothing can match it - but it does need the driver's own
+        // fixed-index restart held off for the batch, which is what the scope below does.
+        const RestartSubstitutionKind restartKind = ResolveRestartSubstitution(type);
+        const Bool arbitraryRestart = restartKind == RestartSubstitutionKind::RewriteIndices;
+        const ScopedSuppressedPrimitiveRestart restartCapOverride(restartKind);
 
         const Bool hasIndexBuffer = BoundIndexBuffer() != nullptr;
 
@@ -889,7 +920,8 @@ void main() {
         // the tier choice and the per-sub-draw feeds use those, not the guess above.
         const Bool feedDrawID = CurrentProgramReadsDrawID();
         const Bool feedBaseVertex = basevertex != nullptr && CurrentProgramReadsBaseVertex();
-        const GLESMultiDrawMode tier = ResolveTierForBatch(feedDrawID, feedBaseVertex, hasIndexBuffer);
+        const GLESMultiDrawMode tier =
+            ResolveTierForBatch(feedDrawID, feedBaseVertex, hasIndexBuffer, arbitraryRestart);
 
         Bool drawn = false;
         switch (tier) {
@@ -921,8 +953,10 @@ void main() {
         // Every tier above may decline a batch whose shape it cannot express. The two
         // below are the floor: a base-vertex replay where the driver has one, and the
         // rewritten index stream where it does not. Both are safe for any batch these
-        // entry points can receive.
-        if (!drawn) {
+        // entry points can receive - except that the base-vertex replay hands the
+        // application's own indices to the driver, which cannot restart on a desktop
+        // restart index, so that batch has only the rewriting floor.
+        if (!drawn && !arbitraryRestart) {
             drawn = RunBaseVertexLoop(mode, count, type, indices, drawcount, basevertex, feedDrawID, feedBaseVertex);
         }
         if (!drawn) {

MobileGL/MG_Backend/DirectGLES/Utils.cpp

@@ -26,6 +26,7 @@
 #include <cmath>
 #include <cctype>
 #include <cstring>
+#include <format>
 #include <regex>
 
 namespace MobileGL::MG_Backend::DirectGLES {
@@ -836,7 +837,9 @@ namespace MobileGL::MG_Backend::DirectGLES {
 
         String BuildPassthroughTessControlEssl(const Uint esslVersion, const Uint patchVertices,
                                                const String& inPerVertexMembers,
-                                               const String& outPerVertexMembers) {
+                                               const String& outPerVertexMembers,
+                                               const FloatVec4& defaultOuterLevel,
+                                               const FloatVec2& defaultInnerLevel) {
 #ifdef TRACY_ENABLE
             ZoneScopedC(TRACY_ZONECOLOR_BACKEND);
 #endif
@@ -866,12 +869,14 @@ namespace MobileGL::MG_Backend::DirectGLES {
             // was declined before this was ever called (ModuleReadsLocatedInput), and gl_PointSize
             // from a tessellation stage is a separate capability on both targets.
             source += "    gl_out[gl_InvocationID].gl_Position = gl_in[gl_InvocationID].gl_Position;\n";
-            source += "    gl_TessLevelOuter[0] = 1.0;\n";
-            source += "    gl_TessLevelOuter[1] = 1.0;\n";
-            source += "    gl_TessLevelOuter[2] = 1.0;\n";
-            source += "    gl_TessLevelOuter[3] = 1.0;\n";
-            source += "    gl_TessLevelInner[0] = 1.0;\n";
-            source += "    gl_TessLevelInner[1] = 1.0;\n";
+            for (Uint i = 0; i < 4; ++i) {
+                source += "    gl_TessLevelOuter[" + std::to_string(i) +
+                          "] = " + MG_Util::ShaderTranspiler::TessellationLevelLiteral(defaultOuterLevel[i]) + ";\n";
+            }
+            for (Uint i = 0; i < 2; ++i) {
+                source += "    gl_TessLevelInner[" + std::to_string(i) +
+                          "] = " + MG_Util::ShaderTranspiler::TessellationLevelLiteral(defaultInnerLevel[i]) + ";\n";
+            }
             source += "}\n";
             return source;
         }

MobileGL/MG_Backend/DirectGLES/Utils.h

@@ -368,11 +368,11 @@ namespace MobileGL::MG_Backend::DirectGLES {
         //
         // All four outer levels and both inner levels are written unconditionally: writing a
         // level the evaluation stage's domain does not use is legal and ignored, and it saves
-        // this from having to know the domain. They are literal 1.0 because that is the GL
-        // default and glPatchParameterfv - their only setter - is a stub in this frontend
-        // (MG_Impl/GLImpl/Exporting/Definitions.cpp). Implementing that entry point means making
-        // the levels a parameter here AND part of what makes a built program stale, exactly as
-        // PATCH_VERTICES already is; the two must move together, so they are named together.
+        // this from having to know the domain. They are the GL_PATCH_DEFAULT_OUTER_LEVEL /
+        // GL_PATCH_DEFAULT_INNER_LEVEL state, baked in as literals - ES has no such state and no
+        // glPatchParameterfv to forward to, so compiling them in is the only way to honour them.
+        // That makes them part of what a built program is stale against, exactly as PATCH_VERTICES
+        // is: see the staleness clause in DirectGLES.cpp's SyncCurrentProgram, which compares both.
         //
         // The same stage, for the same reason, that DirectVulkan synthesizes in
         // ProgramFactory::BuildPassthroughTessControlSource - Vulkan likewise requires both
@@ -382,7 +382,9 @@ namespace MobileGL::MG_Backend::DirectGLES {
         // VkShaderModule against a driver shader object.
         String BuildPassthroughTessControlEssl(Uint esslVersion, Uint patchVertices,
                                                const String& inPerVertexMembers,
-                                               const String& outPerVertexMembers);
+                                               const String& outPerVertexMembers,
+                                               const FloatVec4& defaultOuterLevel,
+                                               const FloatVec2& defaultInnerLevel);
         // Prefix of the writeonly half a read+write image uniform is split into (see
         // SplitReadWriteImageUniforms); the suffix is the image's own (already access-tagged) name.
         constexpr const char* IMAGE_WRITE_ALIAS_PREFIX = "mg_imageWrite_";

MobileGL/MG_Backend/DirectVulkan/BackendObject_DirectVulkan.cpp

@@ -500,11 +500,11 @@ namespace MobileGL::MG_Backend::DirectVulkan {
             .RendererName = "Magma",
             .BackendName = "Direct (Vulkan)",
             .ExtraVendor = Nullopt,
-            .RendererGLInfo = {.TargetGLVersion = {4, 0, 0},
+            .RendererGLInfo = {.TargetGLVersion = {4, 6, 0},
                                .TargetGLSLVersion = {4, 6, 0},
                                // Baseline advertisement (no runtime-gated capabilities); a live
                                // backend reconciles its copy in UpdateAdvertisedExtensions.
-                               .Extensions = BuildAdvertisedExtensions(false, false, false, false),
+                               .Extensions = BuildAdvertisedExtensions(false, false, false, false, false),
                                .IsCompatibilityProfile = false},
             .StaticBackendCapability = {.AllowVSOnlyPrograms = false}};
         return rendererInfo;
@@ -512,9 +512,16 @@ namespace MobileGL::MG_Backend::DirectVulkan {
 
     Vector<GLExtension> BuildAdvertisedExtensions(Bool shaderSubgroupSupported, Bool timerQueriesSupported,
                                                   Bool anisotropicFilteringSupported,
-                                                  Bool nonZeroIndirectBaseInstanceSupported) {
+                                                  Bool nonZeroIndirectBaseInstanceSupported,
+                                                  Bool cubeMapArraySupported) {
         Vector<GLExtension> extensions = {
-            V_OpenGL30, V_OpenGL31, V_OpenGL32, V_OpenGL33, V_OpenGL40, E_GL_ARB_draw_buffers_blend,
+            // The version tokens have to reach the version the backend actually claims:
+            // TargetGLVersion is {4,6,0}, and a list that stopped at OpenGL40 told an
+            // application feature-detecting off these tokens the opposite of what
+            // GL_MAJOR_VERSION / GL_MINOR_VERSION told it.
+            V_OpenGL30, V_OpenGL31, V_OpenGL32, V_OpenGL33, V_OpenGL40, V_OpenGL41, V_OpenGL42, V_OpenGL43,
+            V_OpenGL44, V_OpenGL45, V_OpenGL46,
+            E_GL_ARB_draw_buffers_blend,
             E_GL_ARB_compute_shader, E_GL_ARB_shader_storage_buffer_object, E_GL_ARB_shader_image_load_store,
             E_GL_ARB_clear_buffer_object, E_GL_ARB_program_interface_query, E_GL_ARB_framebuffer_object, E_GL_ARB_draw_indirect,
             E_GL_ARB_multi_draw_indirect,
@@ -533,6 +540,79 @@ namespace MobileGL::MG_Backend::DirectVulkan {
             // Sampling the stencil aspect through DEPTH_STENCIL_TEXTURE_MODE. Core from 4.3,
             // so on a 4.0 context the string is the only way to reach it.
             E_GL_ARB_stencil_texturing,
+            // Unconditional, unlike DirectGLES: a GL texture view is a second set of VkImageViews
+            // over the same VkImage with a sub-range and possibly a reinterpreted VkFormat, which
+            // is core Vulkan on every device MobileGL runs on. Format-reinterpreting views need
+            // VK_IMAGE_CREATE_MUTABLE_FORMAT_BIT on the image, which SyncTextureResource sets for
+            // every immutable-storage texture (see the comment there).
+            E_GL_ARB_texture_view,
+            // Core since 3.2 and implemented here on both backends - glDrawElementsBaseVertex,
+            // glDrawRangeElementsBaseVertex, glDrawElementsInstancedBaseVertex and
+            // glMultiDrawElementsBaseVertex all reach real per-draw vertex rebasing. The string
+            // was simply never emitted, which left KHR-GL4*.draw_elements_base_vertex_tests
+            // NotSupported on a feature that works.
+            E_GL_ARB_draw_elements_base_vertex,
+            // The whole sync-object family is real and core since 3.2: glFenceSync, glIsSync,
+            // glDeleteSync, glClientWaitSync, glWaitSync and glGetSynciv all live in GLImpl over a
+            // backend fence (a VkFence here, an EGLSync/GLsync on DirectGLES), and glGetInteger64v
+            // answers GL_MAX_SERVER_WAIT_TIMEOUT. The string matters for the same reason
+            // ARB_uniform_buffer_object's does: LWJGL builds GLCapabilities from the extension
+            // list, and a caller that finds GL_ARB_sync missing never resolves the entry points -
+            // then calls through null if it uses fences anyway. Nothing in the CTS gates on this
+            // string, so it is advertised on the strength of the implementation, not a test unlock.
+            E_GL_ARB_sync,
+            // Atomic counters, core since 4.2. glGetActiveAtomicCounterBufferiv and the whole
+            // GL_ATOMIC_COUNTER_BUFFER_* query family are real in GLImpl, and the counter buffer
+            // now reaches the shader on BOTH backends - Magma resolves the lowered
+            // gl_AtomicCounterBlock_<N> from the atomic-counter binding points rather than the
+            // shader-storage ones (see ResolveStorageBufferDescriptor). Withheld here until that
+            // landed, because the counter silently read whatever was bound as SSBO N instead.
+            E_GL_ARB_shader_atomic_counters,
+            // glVertexAttribDivisor, core since 3.3 and real on both backends. Applications
+            // (Better Clouds' GLCompat among them) accept the extension string as an
+            // ALTERNATIVE to a 3.3 context when deciding whether instanced rendering is
+            // available, so withholding it makes MobileGL look less capable than it is.
+            E_GL_ARB_instanced_arrays,
+            // Core GL 3.0-4.3 plumbing that has been real here for as long as the backend has
+            // existed, and that was simply never named. None of these unlocks a single CTS case -
+            // the conformance suite reaches all of them through the version - so they are
+            // advertised for the OTHER consumer of this list: LWJGL builds GLCapabilities from the
+            // string set, and an application that gates its ENTRY POINTS on the string rather than
+            // on the version never resolves them and then calls through null. Each is backed by
+            // the entry points named beside it. Kept identical to the DirectGLES block so the two
+            // backends do not disagree about what MobileGL is.
+            //
+            // glBindVertexArray / glGenVertexArrays / glDeleteVertexArrays / glIsVertexArray.
+            E_GL_ARB_vertex_array_object,
+            // The 14 glSamplerParameter* / glGetSamplerParameter* entry points, including the
+            // integer-valued Iiv/Iuiv forms.
+            E_GL_ARB_sampler_objects,
+            // glMapBufferRange + glFlushMappedBufferRange, which ARB_buffer_storage's persistent
+            // maps are already built on top of.
+            E_GL_ARB_map_buffer_range,
+            // glCopyBufferSubData plus the GL_COPY_READ_BUFFER / GL_COPY_WRITE_BUFFER targets.
+            E_GL_ARB_copy_buffer,
+            // glCopyImageSubData, wired to a real backend hook on both backends.
+            E_GL_ARB_copy_image,
+            // GL_TEXTURE_SWIZZLE_{R,G,B,A,RGBA}, which map onto a VkImageView's component swizzle.
+            E_GL_ARB_texture_swizzle,
+            // GL_INT_2_10_10_10_REV / GL_UNSIGNED_INT_2_10_10_10_REV on glVertexAttribPointer plus
+            // the eight glVertexAttribP* entry points.
+            E_GL_ARB_vertex_type_2_10_10_10_rev,
+            // The R/RG internal formats. Named separately from the float ones because an
+            // application may check either.
+            E_GL_ARB_texture_rg,
+            // GL_DEPTH_COMPONENT32F and GL_DEPTH32F_STENCIL8.
+            E_GL_ARB_depth_buffer_float,
+            // The floating-point colour formats. Unlike the rest of this block this string DOES
+            // gate CTS cases - KHR-GL4*.internalformat.texture2d.*{16f,32f} is keyed on it with no
+            // core-version fallback, so eight cases per version list were NotSupported on formats
+            // the backend has always had.
+            E_GL_ARB_texture_float,
+            // glViewportArrayv / glViewportIndexedf{,v} / glScissorArrayv / glScissorIndexed{,v} /
+            // glDepthRangeArrayv / glDepthRangeIndexed / glGetFloati_v / glGetDoublei_v, over the
+            // 16 viewports GL_MAX_VIEWPORTS reports.
+            E_GL_ARB_viewport_array,
             // Advertised with GL_NUM_PROGRAM_BINARY_FORMATS = 0, which the
             // extension explicitly permits. It is also the only thing that
             // exposes glProgramParameteri before GL 4.1.
@@ -585,6 +665,18 @@ namespace MobileGL::MG_Backend::DirectVulkan {
             extensions.push_back(E_GL_EXT_texture_filter_anisotropic);
             extensions.push_back(E_GL_ARB_texture_filter_anisotropic);
         }
+        // A cube map array is a 6n-layer VkImage viewed as VK_IMAGE_VIEW_TYPE_CUBE_ARRAY, and that
+        // view type cannot be created without the imageCubeArray device feature - so the string
+        // follows the feature, not the version, exactly as the per-layer attachment bit does.
+        //
+        // Named for the application's benefit rather than the suite's: measured on Adreno 830,
+        // KHR-GL43.texture_gather.plain-gather-*-cube-array already passed without the string, so
+        // this unlocks no conformance case. It is advertised because the feature is real and
+        // because an application that feature-detects cube map arrays off the string (rather than
+        // off the 4.0 version) would otherwise decline a path this backend serves.
+        if (cubeMapArraySupported) {
+            extensions.push_back(E_GL_ARB_texture_cube_map_array);
+        }
         return extensions;
     }
 
@@ -715,7 +807,8 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         m_rendererInfo.RendererGLInfo.Extensions = BuildAdvertisedExtensions(
             subgroupSupportAdvertised, pVulkanRenderer && pVulkanRenderer->IsTimerQuerySupported(),
             pVulkanRenderer && pVulkanRenderer->IsSamplerAnisotropySupported(),
-            pVulkanRenderer && pVulkanRenderer->IsNonZeroIndirectBaseInstanceSupported());
+            pVulkanRenderer && pVulkanRenderer->IsNonZeroIndirectBaseInstanceSupported(),
+            m_vulkanCaps.SupportsImageCubeArray);
     }
 
     void BackendObject_DirectVulkan::UpdateDynamicBackendParameters() {
@@ -763,6 +856,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
 
         static constexpr SizeT kMaxAdvertisedShaderStorageBlockSize = 512ull * 1024ull * 1024ull;
         m_dynamicParameters.UniformBufferOffsetAlignment = m_vulkanCaps.UniformBufferOffsetAlignment;
+        m_dynamicParameters.ShaderStorageBufferOffsetAlignment = m_vulkanCaps.ShaderStorageBufferOffsetAlignment;
         m_dynamicParameters.AliasedLineWidthRangeMin = m_vulkanCaps.AliasedLineWidthRangeMin;
         m_dynamicParameters.AliasedLineWidthRangeMax = m_vulkanCaps.AliasedLineWidthRangeMax;
         // Without the samplerAnisotropy feature the limit is unusable, so report 1.0 (no anisotropy)
@@ -912,6 +1006,19 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         // it the limit describes a capacity no shader may use, so report none.
         m_dynamicParameters.MaxClipDistances =
             m_vulkanCaps.SupportsShaderClipDistance ? std::max(m_vulkanCaps.MaxClipDistances, 0) : 0;
+        // The cull pair, gated on its own feature. shaderCullDistance is separate from
+        // shaderClipDistance and VulkanRenderer enables it independently, so it gets its own
+        // gate rather than riding on the clip one.
+        m_dynamicParameters.MaxCullDistances =
+            m_vulkanCaps.SupportsShaderCullDistance ? std::max(m_vulkanCaps.MaxCullDistances, 0) : 0;
+        // GL 4.6 core 11.1.3.10: the combined limit is at least as large as either half. A device
+        // with only one of the two features must not report a combined capacity that implies the
+        // other, so the gate is "either feature" and the value never drops below what is enabled.
+        m_dynamicParameters.MaxCombinedClipAndCullDistances =
+            (m_vulkanCaps.SupportsShaderClipDistance || m_vulkanCaps.SupportsShaderCullDistance)
+                ? std::max({m_vulkanCaps.MaxCombinedClipAndCullDistances, m_dynamicParameters.MaxClipDistances,
+                            m_dynamicParameters.MaxCullDistances})
+                : 0;
         m_dynamicParameters.MaxViewports = m_vulkanCaps.MaxViewports;
         // Assigned explicitly rather than left to the struct's defaults, like every other
         // parameter here, so a second fill cannot inherit a stale value. GL_UNDEFINED_VERTEX is

MobileGL/MG_Backend/DirectVulkan/BackendObject_DirectVulkan.h

@@ -75,7 +75,8 @@ namespace MobileGL::MG_Backend::DirectVulkan {
     // the detected device support (passing an already-gated value is harmless).
     Vector<GLExtension> BuildAdvertisedExtensions(Bool shaderSubgroupSupported, Bool timerQueriesSupported,
                                                   Bool anisotropicFilteringSupported,
-                                                  Bool nonZeroIndirectBaseInstanceSupported);
+                                                  Bool nonZeroIndirectBaseInstanceSupported,
+                                                  Bool cubeMapArraySupported);
 
     // Format: <GPU Name>, Vulkan <Vulkan Version>, Driver <Driver Version> — the exact
     // string an initialized backend returns from GetBackendAPIVersionString (and that

MobileGL/MG_Backend/DirectVulkan/Renderer/PipelineFactory.cpp

@@ -201,11 +201,15 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         XXHASH_VERIFY(XXH64_update(m_hashState, &payload.renderPass, sizeof(payload.renderPass)));
         XXHASH_VERIFY(XXH64_update(m_hashState, &payload.colorAttachmentCount, sizeof(payload.colorAttachmentCount)));
         XXHASH_VERIFY(XXH64_update(m_hashState, &payload.rasterizationSamples, sizeof(payload.rasterizationSamples)));
+        XXHASH_VERIFY(XXH64_update(m_hashState, &payload.sampleShadingEnable, sizeof(payload.sampleShadingEnable)));
+        XXHASH_VERIFY(XXH64_update(m_hashState, &payload.minSampleShading, sizeof(payload.minSampleShading)));
         XXHASH_VERIFY(XXH64_update(m_hashState, &payload.subpass, sizeof(payload.subpass)));
         XXHASH_VERIFY(XXH64_update(m_hashState, &payload.topology, sizeof(payload.topology)));
         XXHASH_VERIFY(
             XXH64_update(m_hashState, &payload.primitiveRestartEnable, sizeof(payload.primitiveRestartEnable)));
         XXHASH_VERIFY(XXH64_update(m_hashState, &payload.patchControlPoints, sizeof(payload.patchControlPoints)));
+        XXHASH_VERIFY(XXH64_update(m_hashState, &payload.passthroughTessControlKey,
+                                   sizeof(payload.passthroughTessControlKey)));
         XXHASH_VERIFY(XXH64_update(m_hashState, &payload.viewportCount, sizeof(payload.viewportCount)));
         XXHASH_VERIFY(XXH64_update(m_hashState, &payload.polygonMode, sizeof(payload.polygonMode)));
         XXHASH_VERIFY(XXH64_update(m_hashState, &payload.cullMode, sizeof(payload.cullMode)));
@@ -435,6 +439,10 @@ namespace MobileGL::MG_Backend::DirectVulkan {
 
         VkPipelineMultisampleStateCreateInfo ms{VK_STRUCTURE_TYPE_PIPELINE_MULTISAMPLE_STATE_CREATE_INFO};
         ms.rasterizationSamples = payload.rasterizationSamples;
+        ms.sampleShadingEnable = payload.sampleShadingEnable ? VK_TRUE : VK_FALSE;
+        // Ignored by Vulkan unless sampleShadingEnable is set, but written unconditionally so the
+        // struct's bytes match the hash the payload was keyed by.
+        ms.minSampleShading = payload.minSampleShading;
 
         VkPipelineDepthStencilStateCreateInfo depthStencil{VK_STRUCTURE_TYPE_PIPELINE_DEPTH_STENCIL_STATE_CREATE_INFO};
         depthStencil.depthTestEnable = payload.depthTestEnable ? VK_TRUE : VK_FALSE;

MobileGL/MG_Backend/DirectVulkan/Renderer/PipelineFactory.h

@@ -37,11 +37,24 @@ namespace MobileGL::MG_Backend::DirectVulkan {
             VkRenderPass renderPass = VK_NULL_HANDLE;
             Uint32 colorAttachmentCount = 1;
             VkSampleCountFlagBits rasterizationSamples = VK_SAMPLE_COUNT_1_BIT;
+            // glEnable(GL_SAMPLE_SHADING) + glMinSampleShading, which Vulkan bakes into the
+            // pipeline rather than exposing as dynamic state - so both are part of the pipeline's
+            // identity and both are hashed. The renderer leaves the enable false unless the
+            // device's sampleRateShading feature was enabled
+            // (VUID-VkPipelineMultisampleStateCreateInfo-sampleShadingEnable-00784).
+            Bool sampleShadingEnable = false;
+            Float minSampleShading = 0.0f;
             Uint32 subpass = 0;
             VkPrimitiveTopology topology = VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST;
             Bool primitiveRestartEnable = false;
             // GL_PATCH_VERTICES; only read for a PATCH_LIST topology.
             Uint32 patchControlPoints = 3;
+            // ProgramFactory::ComputePassthroughTessControlKey of the synthesized pass-through
+            // tessellation control stage below, or 0 when this pipeline has none. Hashed, because
+            // the levels glPatchParameterfv set are compiled INTO that module and are not a
+            // function of the program or of patchControlPoints - see the note on
+            // passthroughTessControlStage.
+            Uint64 passthroughTessControlKey = 0;
             // How many of ARB_viewport_array's viewports this pipeline rasterizes into. 1 for
             // every program that never assigns gl_ViewportIndex, which is all of them outside the
             // conformance suite - the wide shape costs a longer vkCmdSetViewport/Scissor per state
@@ -87,8 +100,9 @@ namespace MobileGL::MG_Backend::DirectVulkan {
             // renderer could not build one, and CreatePipeline refuses the pipeline - the same
             // refusal it applies when `stages` itself is half-tessellated.
             //
-            // NOT hashed: it is a pure function of the program and of patchControlPoints, both
-            // of which ComputeHash already mixes in.
+            // NOT hashed directly: it is a pure function of the program, of patchControlPoints and
+            // of the default tessellation levels - the first two of which ComputeHash already
+            // mixes in, and the third of which arrives through passthroughTessControlKey above.
             VkPipelineShaderStageCreateInfo passthroughTessControlStage{};
             const VkPipelineVertexInputStateCreateInfo* vertexInputState = nullptr;
             // Diagnostic only; may be null. Read solely from the pipeline-creation failure path.

MobileGL/MG_Backend/DirectVulkan/Renderer/ProgramFactory.cpp

@@ -13,7 +13,10 @@
 #include "MG_Util/ShaderTranspiler/SpvcSession.h"
 #include "MG_Util/ShaderTranspiler/Types.h"
 #include <algorithm>
+#include <bit>
+#include <cmath>
 #include <cstring>
+#include <format>
 #include <map>
 #include <utility>
 #include <spirv-tools/libspirv.h>
@@ -2186,6 +2189,49 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         }
     }
 
+    SamplerNumericDomain ProgramFactory::UniformTypeToImageNumericDomain(GLenum glType) {
+        switch (glType) {
+        case GL_INT_IMAGE_1D:
+        case GL_INT_IMAGE_2D:
+        case GL_INT_IMAGE_3D:
+        case GL_INT_IMAGE_2D_RECT:
+        case GL_INT_IMAGE_CUBE:
+        case GL_INT_IMAGE_BUFFER:
+        case GL_INT_IMAGE_1D_ARRAY:
+        case GL_INT_IMAGE_2D_ARRAY:
+        case GL_INT_IMAGE_CUBE_MAP_ARRAY:
+        case GL_INT_IMAGE_2D_MULTISAMPLE:
+        case GL_INT_IMAGE_2D_MULTISAMPLE_ARRAY:
+            return SamplerNumericDomain::SignedInteger;
+        case GL_UNSIGNED_INT_IMAGE_1D:
+        case GL_UNSIGNED_INT_IMAGE_2D:
+        case GL_UNSIGNED_INT_IMAGE_3D:
+        case GL_UNSIGNED_INT_IMAGE_2D_RECT:
+        case GL_UNSIGNED_INT_IMAGE_CUBE:
+        case GL_UNSIGNED_INT_IMAGE_BUFFER:
+        case GL_UNSIGNED_INT_IMAGE_1D_ARRAY:
+        case GL_UNSIGNED_INT_IMAGE_2D_ARRAY:
+        case GL_UNSIGNED_INT_IMAGE_CUBE_MAP_ARRAY:
+        case GL_UNSIGNED_INT_IMAGE_2D_MULTISAMPLE:
+        case GL_UNSIGNED_INT_IMAGE_2D_MULTISAMPLE_ARRAY:
+            return SamplerNumericDomain::UnsignedInteger;
+        case GL_IMAGE_1D:
+        case GL_IMAGE_2D:
+        case GL_IMAGE_3D:
+        case GL_IMAGE_2D_RECT:
+        case GL_IMAGE_CUBE:
+        case GL_IMAGE_BUFFER:
+        case GL_IMAGE_1D_ARRAY:
+        case GL_IMAGE_2D_ARRAY:
+        case GL_IMAGE_CUBE_MAP_ARRAY:
+        case GL_IMAGE_2D_MULTISAMPLE:
+        case GL_IMAGE_2D_MULTISAMPLE_ARRAY:
+            return SamplerNumericDomain::Float;
+        default:
+            return SamplerNumericDomain::Unknown;
+        }
+    }
+
     ProgramFactory::HashType ProgramFactory::ComputeHash(const MG_State::GLState::ProgramObject& program,
                                                          CompileOptionFlags flags) const {
         XXHASH_VERIFY(XXH64_reset(m_hashState, m_config.CacheVersion));
@@ -2302,6 +2348,14 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         case GL_INT_IMAGE_2D_RECT:
         case GL_UNSIGNED_INT_IMAGE_2D_RECT:
             return TextureTarget::TextureRectangle;
+        case GL_SAMPLER_CUBE_MAP_ARRAY:
+        case GL_SAMPLER_CUBE_MAP_ARRAY_SHADOW:
+        case GL_INT_SAMPLER_CUBE_MAP_ARRAY:
+        case GL_UNSIGNED_INT_SAMPLER_CUBE_MAP_ARRAY:
+        case GL_IMAGE_CUBE_MAP_ARRAY:
+        case GL_INT_IMAGE_CUBE_MAP_ARRAY:
+        case GL_UNSIGNED_INT_IMAGE_CUBE_MAP_ARRAY:
+            return TextureTarget::TextureCubeMapArray;
         case GL_SAMPLER_2D:
         case GL_SAMPLER_2D_SHADOW:
         case GL_INT_SAMPLER_2D:
@@ -2948,6 +3002,34 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                                     static_cast<Int>(numericDomain));
                     entry.samplerNumericDomainByBinding[binding] = numericDomain;
                 }
+                // Every other opaque kind records its domain too. Only the combined-image-sampler
+                // path above needs it to pick a sampled view format; the three below need it to
+                // describe the descriptor a binding gets when its unit is UNBOUND, which is legal
+                // GL and must not lose the draw (see UniformManager's Resolve*Descriptor). Left
+                // Unknown, those placeholders would have no way to tell a `samplerBuffer` from a
+                // `usamplerBuffer` - and a texel buffer view whose numeric type disagrees with the
+                // shader's is invalid Vulkan, not merely wrong data.
+                if (descriptorKind == DescriptorBindingKind::UniformTexelBuffer ||
+                    descriptorKind == DescriptorBindingKind::StorageTexelBuffer ||
+                    descriptorKind == DescriptorBindingKind::StorageImage) {
+                    const SamplerNumericDomain opaqueDomain =
+                        descriptorKind == DescriptorBindingKind::UniformTexelBuffer
+                            ? UniformTypeToSamplerNumericDomain(uniformType)
+                            : UniformTypeToImageNumericDomain(uniformType);
+                    MOBILEGL_ASSERT(opaqueDomain != SamplerNumericDomain::Unknown,
+                                    "ProgramFactory::ReflectLayout: failed to resolve numeric domain for '%s' "
+                                    "(uniformType=0x%x)",
+                                    uniformName.c_str(), uniformType);
+                    MOBILEGL_ASSERT(entry.samplerNumericDomainByBinding[binding] ==
+                                            SamplerNumericDomain::Unknown ||
+                                        entry.samplerNumericDomainByBinding[binding] == opaqueDomain,
+                                    "ProgramFactory::ReflectLayout: binding %u ('%s') has conflicting numeric "
+                                    "domains (%d vs %d)",
+                                    binding, uniformName.c_str(),
+                                    static_cast<Int>(entry.samplerNumericDomainByBinding[binding]),
+                                    static_cast<Int>(opaqueDomain));
+                    entry.samplerNumericDomainByBinding[binding] = opaqueDomain;
+                }
                 MOBILEGL_ASSERT(entry.samplerUniformLocationByBinding[binding] < 0 || location < 0 ||
                                     entry.samplerUniformLocationByBinding[binding] == location,
                                 "ProgramFactory::ReflectLayout: texture binding %u maps to conflicting uniform locations (%d vs %d)",
@@ -3515,18 +3597,136 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         }
     }
 
-    String ProgramFactory::BuildPassthroughTessControlSource(Uint32 patchVertices) {
+    Uint64 ProgramFactory::ComputePassthroughTessControlKey(Uint32 patchVertices,
+                                                            const FloatVec4& defaultOuterLevel,
+                                                            const FloatVec2& defaultInnerLevel,
+                                                            Uint32 perVertexMembers) {
+        // A plain 32-byte blob of exactly what the generator reads, hashed once. Deliberately over
+        // the RAW BITS rather than the values: two levels that compare unequal must key apart, and
+        // a NaN level - which glPatchParameterfv accepts - compares unequal to itself.
+        struct Blob {
+            Uint32 patchVertices;
+            Uint32 outerBits[4];
+            Uint32 innerBits[2];
+            Uint32 perVertexMembers;
+        } blob{};
+        blob.patchVertices = patchVertices;
+        for (Uint32 i = 0; i < 4; ++i) blob.outerBits[i] = std::bit_cast<Uint32>(defaultOuterLevel[i]);
+        for (Uint32 i = 0; i < 2; ++i) blob.innerBits[i] = std::bit_cast<Uint32>(defaultInnerLevel[i]);
+        blob.perVertexMembers = perVertexMembers;
+        return XXH64(&blob, sizeof(blob), 0);
+    }
+
+    // The member list a gl_PerVertex redeclaration must spell, derived from the mask. Order is
+    // glslang's declaration order and is load-bearing: a redeclaration whose members are the same
+    // set in a different order is a different block.
+    static String BuildPerVertexMemberDeclarations(Uint32 perVertexMembers) {
+        using Bit = ProgramFactory::PerVertexMemberBit;
+        String members;
+        if (perVertexMembers & static_cast<Uint32>(Bit::Position)) members += "    vec4 gl_Position;\n";
+        if (perVertexMembers & static_cast<Uint32>(Bit::PointSize)) members += "    float gl_PointSize;\n";
+        // Sized at one, not left unsized: an unsized built-in array in a redeclared block is
+        // implicitly sized by use, and this stage never indexes either distance array.
+        if (perVertexMembers & static_cast<Uint32>(Bit::ClipDistance)) members += "    float gl_ClipDistance[1];\n";
+        if (perVertexMembers & static_cast<Uint32>(Bit::CullDistance)) members += "    float gl_CullDistance[1];\n";
+        return members;
+    }
+
+    Uint32 ProgramFactory::ReflectPerVertexInputMembers(const Vector<Uint>& spirv) {
+        // Minimal, self-contained SPIR-V walk. SPIRV-Reflect is deliberately NOT used: for an
+        // array of interface blocks it reports built_in == -1 on the block and leaves every
+        // member's built_in at 0 (which is SpvBuiltInPosition), so a member walk through it reads
+        // "Position, Position, Position" - the same trap ReflectPassthroughTessControlNeed
+        // documents. The decorations below are unambiguous.
+        constexpr SizeT kHeaderWords = 5;
+        constexpr Uint32 kOpName = 5;
+        constexpr Uint32 kOpDecorate = 71;
+        constexpr Uint32 kOpMemberDecorate = 72;
+        constexpr Uint32 kOpTypeArray = 28;
+        constexpr Uint32 kOpTypePointer = 32;
+        constexpr Uint32 kOpVariable = 59;
+        constexpr Uint32 kDecorationBlock = 2;
+        constexpr Uint32 kDecorationBuiltIn = 11;
+        constexpr Uint32 kStorageClassInput = 1;
+        constexpr Uint32 kBuiltInPosition = 0;
+        constexpr Uint32 kBuiltInPointSize = 1;
+        constexpr Uint32 kBuiltInClipDistance = 3;
+        constexpr Uint32 kBuiltInCullDistance = 4;
+        (void)kOpName;
+
+        if (spirv.size() <= kHeaderWords) return 0;
+
+        UnorderedMap<Uint32, Uint32> arrayElementType;             // array id   -> element type id
+        UnorderedMap<Uint32, Pair<Uint32, Uint32>> pointerPointee; // pointer id -> (storage class, pointee)
+        UnorderedMap<Uint32, Uint32> structMembers;                // struct id  -> PerVertexMemberBit mask
+        std::set<Uint32> blockStructs;
+        Vector<Uint32> inputVariablePointerTypes;
+
+        for (SizeT i = kHeaderWords; i < spirv.size();) {
+            const Uint32 wordCount = spirv[i] >> 16;
+            const Uint32 opcode = spirv[i] & 0xFFFFu;
+            if (wordCount == 0 || i + wordCount > spirv.size()) break;
+            const Uint32* words = &spirv[i];
+            switch (opcode) {
+            case kOpTypeArray:
+                if (wordCount >= 4) arrayElementType[words[1]] = words[2];
+                break;
+            case kOpTypePointer:
+                if (wordCount >= 4) pointerPointee[words[1]] = {words[2], words[3]};
+                break;
+            case kOpVariable:
+                if (wordCount >= 4 && words[3] == kStorageClassInput) inputVariablePointerTypes.push_back(words[1]);
+                break;
+            case kOpDecorate:
+                if (wordCount >= 3 && words[2] == kDecorationBlock) blockStructs.insert(words[1]);
+                break;
+            case kOpMemberDecorate:
+                if (wordCount >= 5 && words[3] == kDecorationBuiltIn) {
+                    Uint32 bit = 0;
+                    switch (words[4]) {
+                    case kBuiltInPosition: bit = static_cast<Uint32>(PerVertexMemberBit::Position); break;
+                    case kBuiltInPointSize: bit = static_cast<Uint32>(PerVertexMemberBit::PointSize); break;
+                    case kBuiltInClipDistance: bit = static_cast<Uint32>(PerVertexMemberBit::ClipDistance); break;
+                    case kBuiltInCullDistance: bit = static_cast<Uint32>(PerVertexMemberBit::CullDistance); break;
+                    default: break;
+                    }
+                    structMembers[words[1]] |= bit;
+                }
+                break;
+            default:
+                break;
+            }
+            i += wordCount;
+        }
+
+        // The one Input variable whose type is an array of a Block-decorated struct IS gl_in;
+        // gl_TessCoord and friends are plain scalars/vectors and never match.
+        for (const Uint32 pointerType : inputVariablePointerTypes) {
+            const auto pointer = pointerPointee.find(pointerType);
+            if (pointer == pointerPointee.end()) continue;
+            const auto array = arrayElementType.find(pointer->second.second);
+            if (array == arrayElementType.end()) continue;
+            if (!blockStructs.contains(array->second)) continue;
+            const auto members = structMembers.find(array->second);
+            if (members == structMembers.end()) continue;
+            return members->second;
+        }
+        return 0;
+    }
+
+    String ProgramFactory::BuildPassthroughTessControlSource(Uint32 patchVertices,
+                                                             const FloatVec4& defaultOuterLevel,
+                                                             const FloatVec2& defaultInnerLevel,
+                                                             Uint32 perVertexMembers) {
         // The stage GL 4.6 core 11.2.2 describes when a program has an evaluation shader and no
         // control shader: "the input patch is passed through unmodified", the output patch has
         // as many vertices as the input one (PATCH_VERTICES), and the levels come from the
         // PATCH_DEFAULT_OUTER_LEVEL / PATCH_DEFAULT_INNER_LEVEL state.
         //
-        // Those two levels default to 1.0 and are baked here as literals because
-        // glPatchParameterfv - their only setter - is not implemented in this frontend (it is a
-        // stub in MG_Impl/GLImpl/Exporting/Definitions.cpp). Implementing that entry point means
-        // making the levels a parameter of this source AND of the cache key in
-        // GetOrCreatePassthroughTessControlStage; the two must move together, so they are named
-        // together here.
+        // Those two levels are baked in as literals - Vulkan has no equivalent dynamic state, so
+        // compiling them in is the only way to honour glPatchParameterfv. That makes them part of
+        // this module's identity: GetOrCreatePassthroughTessControlStage keys its cache on them,
+        // and PipelineFactory hashes them into the pipeline key. The three must move together.
         //
         // gl_out carries gl_Position and nothing else on purpose. The evaluation stage that
         // reads it was linked against the VERTEX stage directly, so its input gl_PerVertex holds
@@ -3540,60 +3740,91 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         // this from having to know the domain.
         String source = "#version 450 core\n";
         source += "layout(vertices = " + std::to_string(patchVertices) + ") out;\n";
-        // gl_in and gl_out are redeclared to the exact gl_PerVertex the FRONTEND's linked programs
-        // carry - gl_Position, gl_PointSize, gl_ClipDistance[1], in that order - because Vulkan
-        // matches built-in interface blocks by their whole shape, and the two obvious spellings
-        // are both wrong:
+        // gl_in and gl_out are redeclared to the exact gl_PerVertex the NEIGHBOURING EVALUATION
+        // STAGE carries, because Vulkan matches built-in interface blocks by their whole shape,
+        // and the two obvious spellings are both wrong:
         //   * narrowing the block to gl_Position alone makes the evaluation stage read a patch of
         //     zeroes (degenerate triangles, nothing rasterized), and
-        //   * taking glslang's DEFAULT block for a standalone control stage yields FOUR members -
-        //     it appends gl_CullDistance - where a linked vertex+evaluation program has three.
-        // PassthroughTessControlTest.MatchesTheFrontendPerVertexBlock is the latch: it links a
-        // vertex+evaluation program through this same compiler and fails if the two shapes ever
-        // stop agreeing, rather than letting the mismatch show up as a black frame.
+        //   * taking glslang's DEFAULT block for a standalone control stage yields whatever THIS
+        //     source's #version implies, which is unrelated to the evaluation stage's.
+        //
+        // The member set is a PARAMETER rather than a constant, and that is the whole point: it
+        // was hardcoded to {gl_Position, gl_PointSize, gl_ClipDistance[1]}, which is the shape a
+        // program carries only below #version 450. glslang appends gl_CullDistance to the block
+        // from 450 upward, so every 450/460 program - and every ESSL program, which the source
+        // processor rewrites to "#version 460 core" - carried FOUR members against this stage's
+        // three and got the black-frame-no-error case described above. The mask comes from
+        // ReflectPerVertexInputMembers, read off the evaluation stage's own SPIR-V.
+        // PassthroughTessControlTest.MatchesTheFrontendPerVertexBlock is the latch, and it now
+        // links the program at both 430 and 460.
         //
         // Only gl_Position is written. gl_PointSize is declared but left alone deliberately:
         // writing it from a tessellation stage requires the shaderTessellationAndGeometryPointSize
         // feature, which this renderer does not enable, so a program whose evaluation stage reads
         // gl_in[].gl_PointSize gets an undefined point size instead of the vertex stage's - a gap
         // this trades for not making every tessellated pipeline depend on an optional feature.
-        source += "in gl_PerVertex {\n"
-                  "    vec4 gl_Position;\n"
-                  "    float gl_PointSize;\n"
-                  "    float gl_ClipDistance[1];\n"
-                  "} gl_in[gl_MaxPatchVertices];\n";
-        source += "out gl_PerVertex {\n"
-                  "    vec4 gl_Position;\n"
-                  "    float gl_PointSize;\n"
-                  "    float gl_ClipDistance[1];\n"
-                  "} gl_out[];\n";
+        const String perVertexBody = BuildPerVertexMemberDeclarations(perVertexMembers);
+        source += "in gl_PerVertex {\n" + perVertexBody + "} gl_in[gl_MaxPatchVertices];\n";
+        source += "out gl_PerVertex {\n" + perVertexBody + "} gl_out[];\n";
         source += "void main() {\n";
         source += "    gl_out[gl_InvocationID].gl_Position = gl_in[gl_InvocationID].gl_Position;\n";
-        source += "    gl_TessLevelOuter[0] = 1.0;\n";
-        source += "    gl_TessLevelOuter[1] = 1.0;\n";
-        source += "    gl_TessLevelOuter[2] = 1.0;\n";
-        source += "    gl_TessLevelOuter[3] = 1.0;\n";
-        source += "    gl_TessLevelInner[0] = 1.0;\n";
-        source += "    gl_TessLevelInner[1] = 1.0;\n";
+        for (Uint32 i = 0; i < 4; ++i) {
+            source += "    gl_TessLevelOuter[" + std::to_string(i) +
+                      "] = " + MG_Util::ShaderTranspiler::TessellationLevelLiteral(defaultOuterLevel[i]) + ";\n";
+        }
+        for (Uint32 i = 0; i < 2; ++i) {
+            source += "    gl_TessLevelInner[" + std::to_string(i) +
+                      "] = " + MG_Util::ShaderTranspiler::TessellationLevelLiteral(defaultInnerLevel[i]) + ";\n";
+        }
         source += "}\n";
         return source;
     }
 
-    VkPipelineShaderStageCreateInfo ProgramFactory::GetOrCreatePassthroughTessControlStage(Uint32 patchVertices) {
+    VkPipelineShaderStageCreateInfo ProgramFactory::GetOrCreatePassthroughTessControlStage(
+        Uint32 patchVertices, const FloatVec4& defaultOuterLevel, const FloatVec2& defaultInnerLevel,
+        Uint32 perVertexMembers) {
+        // Everything compiled into the stage, folded into one key. The patch size alone stopped
+        // being enough once glPatchParameterfv could change the levels: two modules that differ
+        // only in a baked-in level are different modules, and pipelines built from either may be
+        // alive at the same time. The gl_PerVertex member set joins it for the same reason - two
+        // programs at different GLSL versions need differently-shaped blocks.
+        const Uint64 key =
+            ComputePassthroughTessControlKey(patchVertices, defaultOuterLevel, defaultInnerLevel, perVertexMembers);
         // A cached VK_NULL_HANDLE is a remembered failure, not a miss: returning it keeps a
         // generator that cannot compile from re-running glslang on every draw.
-        const auto cached = m_passthroughTessControlStages.find(patchVertices);
+        const auto cached = m_passthroughTessControlStages.find(key);
         if (cached != m_passthroughTessControlStages.end()) {
             return cached->second;
         }
 
+        // The key stopped being bounded when the levels joined it: patchVertices alone could only
+        // take 32 values, but six unclamped application floats can take any number, and an
+        // application that ramps a level per frame would retain one VkShaderModule per frame for
+        // the lifetime of the device. Flushed wholesale rather than aged: a module is not
+        // referenced by the pipelines built from it (Vulkan copies what it needs at
+        // vkCreateGraphicsPipelines), everything here runs on the GL thread, and an application
+        // that can overflow this cap is already recompiling every frame - so the flush costs it
+        // nothing it was not paying anyway.
+        if (m_passthroughTessControlStages.size() >= kMaxPassthroughTessControlStages) {
+            MGLOG_D("ProgramFactory: flushing %zu pass-through tessellation control stages; the application has "
+                    "used more than %zu distinct (patch size, default level) combinations",
+                    m_passthroughTessControlStages.size(), kMaxPassthroughTessControlStages);
+            for (auto& entry : m_passthroughTessControlStages) {
+                if (entry.second.module != VK_NULL_HANDLE) {
+                    vkDestroyShaderModule(m_device, entry.second.module, nullptr);
+                }
+            }
+            m_passthroughTessControlStages.clear();
+        }
+
         VkPipelineShaderStageCreateInfo stage{VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO};
         stage.stage = VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT;
         stage.module = VK_NULL_HANDLE;
         stage.pName = "main";
 
         using namespace MG_Util::ShaderTranspiler;
-        const String source = BuildPassthroughTessControlSource(patchVertices);
+        const String source =
+            BuildPassthroughTessControlSource(patchVertices, defaultOuterLevel, defaultInnerLevel, perVertexMembers);
         // Same compile configuration as every other stage of every other program: this runs on
         // the GL thread (the draw path), so the live compile env is the right one, and flags=0
         // is the Vulkan-targeting form (CompileForOpenGL is what the GLES backend adds).
@@ -3607,7 +3838,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
             MGLOG_E("ProgramFactory: could not compile the pass-through tessellation control stage for "
                     "patchVertices=%u; a program with an evaluation stage and no control stage cannot draw. %s",
                     patchVertices, compiled.error().log.c_str());
-            m_passthroughTessControlStages.emplace(patchVertices, stage);
+            m_passthroughTessControlStages.emplace(key, stage);
             return stage;
         }
 
@@ -3617,7 +3848,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         if (!linked) {
             MGLOG_E("ProgramFactory: could not link the pass-through tessellation control stage for "
                     "patchVertices=%u. %s", patchVertices, linked.error().log.c_str());
-            m_passthroughTessControlStages.emplace(patchVertices, stage);
+            m_passthroughTessControlStages.emplace(key, stage);
             return stage;
         }
 
@@ -3626,7 +3857,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         if (!binary || binary.value().empty() || binary.value().front().empty()) {
             MGLOG_E("ProgramFactory: could not generate SPIR-V for the pass-through tessellation control stage "
                     "for patchVertices=%u", patchVertices);
-            m_passthroughTessControlStages.emplace(patchVertices, stage);
+            m_passthroughTessControlStages.emplace(key, stage);
             return stage;
         }
 
@@ -3648,14 +3879,14 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         if (result != VK_SUCCESS) {
             MGLOG_E("ProgramFactory: vkCreateShaderModule failed (%d) for the pass-through tessellation control "
                     "stage for patchVertices=%u", static_cast<Int>(result), patchVertices);
-            m_passthroughTessControlStages.emplace(patchVertices, stage);
+            m_passthroughTessControlStages.emplace(key, stage);
             return stage;
         }
 
         stage.module = module;
         MGLOG_D("ProgramFactory: built the pass-through tessellation control stage for patchVertices=%u "
                 "(GL 4.6 11.2.2; Vulkan has no fixed-function equivalent)", patchVertices);
-        m_passthroughTessControlStages.emplace(patchVertices, stage);
+        m_passthroughTessControlStages.emplace(key, stage);
         return stage;
     }
 
@@ -3665,6 +3896,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         VkProgramObject& entry) const {
         entry.needsPassthroughTessControl = false;
         entry.passthroughTessControlEmulatable = false;
+        entry.passthroughPerVertexMembers = 0;
 
         Bool hasTessEval = false;
         Bool hasTessControl = false;
@@ -3684,6 +3916,17 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         if (tessEvalModuleIndex >= spirv.size() || spirv[tessEvalModuleIndex].empty()) return;
         const auto& module = spirv[tessEvalModuleIndex];
 
+        // The shape the synthesized control stage has to redeclare. Read here because this is the
+        // only place that holds the evaluation stage's module; a zero mask means the walk found
+        // no input per-vertex block at all, in which case the pre-450 shape is the safe stand-in
+        // (it is what every program carried before gl_CullDistance joined the block).
+        const Uint32 perVertexMembers = ReflectPerVertexInputMembers(module);
+        entry.passthroughPerVertexMembers = perVertexMembers != 0 ? perVertexMembers : kDefaultPerVertexMembers;
+        if (perVertexMembers == 0) {
+            MGLOG_W("ProgramFactory: could not read the evaluation stage's gl_PerVertex block shape; the "
+                    "pass-through control stage falls back to the pre-450 three-member form");
+        }
+
         SpvReflectShaderModule reflectModule{};
         const SpvReflectResult createResult =
             spvReflectCreateShaderModule(module.size() * sizeof(Uint), module.data(), &reflectModule);

MobileGL/MG_Backend/DirectVulkan/Renderer/ProgramFactory.h

@@ -76,6 +76,23 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         using CompileOptionFlags = Flags<CompileOptionBit>;
         using HashType = Uint64;
 
+        // The gl_PerVertex members a pass-through tessellation control stage may have to carry,
+        // in the order glslang declares them - which is the order a redeclaration must use.
+        // Which of them exist is a function of the neighbouring stage's GLSL VERSION
+        // (gl_CullDistance joins the block at #version 450), so the mask is read off that
+        // stage's SPIR-V rather than assumed. See ReflectPerVertexInputMembers.
+        enum class PerVertexMemberBit : Uint32 {
+            Position = 1u << 0,
+            PointSize = 1u << 1,
+            ClipDistance = 1u << 2,
+            CullDistance = 1u << 3,
+        };
+        // What a program parsed below #version 450 carries, and the fallback when a module's
+        // block cannot be read.
+        static constexpr Uint32 kDefaultPerVertexMembers =
+            static_cast<Uint32>(PerVertexMemberBit::Position) | static_cast<Uint32>(PerVertexMemberBit::PointSize) |
+            static_cast<Uint32>(PerVertexMemberBit::ClipDistance);
+
         struct UpdateAfterBindLimits {
             Bool enabled = false;
             Uint32 maxPerStageSamplers = 0;
@@ -194,6 +211,15 @@ namespace MobileGL::MG_Backend::DirectVulkan {
             // instead (PipelineFactory::CreatePipeline refuses the pipeline and the draw is
             // skipped). See ReflectPassthroughTessControlNeed.
             Bool passthroughTessControlEmulatable = false;
+            // Which gl_PerVertex members the evaluation stage's `in gl_PerVertex gl_in[]` block
+            // actually carries, as a PerVertexMemberBit mask read off its SPIR-V. The synthesized
+            // control stage has to redeclare the SAME shape: glslang appends gl_CullDistance to
+            // that block from #version 450 upward, so a 450/460 program - and every ESSL program,
+            // which the source processor rewrites to "#version 460 core" - carries four members
+            // where a 430 program carries three. A fixed three-member pass-through fed the
+            // evaluation stage a differently-shaped block, which is the black-frame-no-error case
+            // this whole family is written around.
+            Uint32 passthroughPerVertexMembers = 0;
             // Frame-boundary counter value of the last GetOrCreateProgram hit; drives
             // cache eviction (see OnFrameBoundary). Mutable: the draw snapshot's memoised
             // entry pointer re-stamps use through a const reference (StampProgramUse).
@@ -249,6 +275,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                 writesViewportIndexBuiltin = other.writesViewportIndexBuiltin;
                 needsPassthroughTessControl = other.needsPassthroughTessControl;
                 passthroughTessControlEmulatable = other.passthroughTessControlEmulatable;
+                passthroughPerVertexMembers = other.passthroughPerVertexMembers;
                 lastUsedFrame = other.lastUsedFrame;
                 other.hash = 0;
                 other.descriptorSetLayout = VK_NULL_HANDLE;
@@ -267,6 +294,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                 other.writesViewportIndexBuiltin = false;
                 other.needsPassthroughTessControl = false;
                 other.passthroughTessControlEmulatable = false;
+                other.passthroughPerVertexMembers = 0;
                 other.lastUsedFrame = 0;
             }
             VkProgramObject& operator=(VkProgramObject&& other) noexcept {
@@ -311,6 +339,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                 writesViewportIndexBuiltin = other.writesViewportIndexBuiltin;
                 needsPassthroughTessControl = other.needsPassthroughTessControl;
                 passthroughTessControlEmulatable = other.passthroughTessControlEmulatable;
+                passthroughPerVertexMembers = other.passthroughPerVertexMembers;
                 lastUsedFrame = other.lastUsedFrame;
                 other.hash = 0;
                 other.descriptorSetLayout = VK_NULL_HANDLE;
@@ -329,6 +358,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                 other.writesViewportIndexBuiltin = false;
                 other.needsPassthroughTessControl = false;
                 other.passthroughTessControlEmulatable = false;
+                other.passthroughPerVertexMembers = 0;
                 other.lastUsedFrame = 0;
                 return *this;
             }
@@ -448,6 +478,14 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         static VkShaderStageFlagBits ToVkStage(ShaderStage stage);
         static VkFormat ConvertSpirvImageFormatToVkFormat(SpvImageFormat format);
         static SamplerNumericDomain UniformTypeToSamplerNumericDomain(GLenum glType);
+        // The same question for an IMAGE uniform (`image2D`, `uimageBuffer`, ...), which the
+        // sampler form above deliberately does not answer. Kept separate rather than folded in
+        // because the two are asked in different places for different reasons: a sampler's domain
+        // decides a sampled VIEW format, an image's decides what a placeholder descriptor for an
+        // UNBOUND image unit must be (see UniformManager::AcquireUnboundTexelBufferView and
+        // GetUnboundStorageImageTexture) - a formatless `writeonly` declaration reflects no
+        // format at all, and the numeric domain is then the only thing that constrains it.
+        static SamplerNumericDomain UniformTypeToImageNumericDomain(GLenum glType);
         // True when any entry point declares the DepthReplacing execution mode, i.e. the
         // shader assigns gl_FragDepth. Exposed so the blended depth-write quirk's exemption
         // can be pinned by tests. A false negative loses the exemption, so such a shader is
@@ -477,18 +515,40 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         // the caller then has no control stage to inject, and CreatePipeline refuses the
         // pipeline rather than handing the driver a half-tessellated one.
         //
-        // Keyed on the patch size because GL takes the output patch size from PATCH_VERTICES,
-        // which is draw state, not link state - the CTS case that motivated this links at the
-        // default 3 and draws at 4. The pipeline cache already re-keys on patchControlPoints,
-        // so the module a pipeline was built with is part of that pipeline's identity.
-        // Compiling is bounded by the number of distinct patch sizes a program draws with
-        // (MAX_PATCH_VERTICES = 32 in the worst case, one or two in practice) and only ever
-        // happens for the rare program that has no control stage at all.
-        VkPipelineShaderStageCreateInfo GetOrCreatePassthroughTessControlStage(Uint32 patchVertices);
+        // Keyed on the patch size, the six default tessellation levels AND the gl_PerVertex
+        // member set, because all three decide what the generator emits. The size comes from
+        // PATCH_VERTICES and the levels from PATCH_DEFAULT_OUTER_LEVEL / PATCH_DEFAULT_INNER_LEVEL
+        // - draw state rather than link state, and the CTS case that motivated this links at the
+        // default 3 and draws at 4. The member set comes from the neighbouring evaluation stage's
+        // own SPIR-V, so two programs at different GLSL versions need different modules. The
+        // pipeline cache re-keys on the same inputs, so the module a pipeline was built with is
+        // part of that pipeline's identity. Compiling is bounded by the number of distinct
+        // (size, levels, members) combinations a program draws with - one or two in practice -
+        // and only ever happens for the rare program that has no control stage at all.
+        VkPipelineShaderStageCreateInfo GetOrCreatePassthroughTessControlStage(Uint32 patchVertices,
+                                                                              const FloatVec4& defaultOuterLevel,
+                                                                              const FloatVec2& defaultInnerLevel,
+                                                                              Uint32 perVertexMembers);
 
         // Source of the module above. Exposed for tests: the generated GLSL is the whole
         // contract with the evaluation stage, so it is worth pinning independently of a device.
-        static String BuildPassthroughTessControlSource(Uint32 patchVertices);
+        static String BuildPassthroughTessControlSource(Uint32 patchVertices, const FloatVec4& defaultOuterLevel,
+                                                        const FloatVec2& defaultInnerLevel, Uint32 perVertexMembers);
+
+        // The identity of one such module: everything the generator bakes in, folded into a
+        // 64-bit key over the raw bits (so -0.0 and +0.0 key apart, which is harmless, and NaN
+        // keys to itself, which is what matters). Shared with PipelineFactory, which mixes the
+        // same value into the pipeline hash so a pipeline can never be handed a module built for
+        // different levels or a different block shape.
+        static Uint64 ComputePassthroughTessControlKey(Uint32 patchVertices, const FloatVec4& defaultOuterLevel,
+                                                       const FloatVec2& defaultInnerLevel, Uint32 perVertexMembers);
+
+        // The PerVertexMemberBit mask of the INPUT per-vertex block a module declares, read
+        // straight out of its SPIR-V (OpMemberDecorate ... BuiltIn on the struct behind the one
+        // Input variable that is an array of a Block-decorated struct). Zero when the module has
+        // no such block. Exposed for tests, which is the only way to pin the shape agreement
+        // without a device.
+        static Uint32 ReflectPerVertexInputMembers(const Vector<Uint>& spirv);
 
     private:
         struct ProgramLookupCache {
@@ -548,11 +608,20 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         // See GetCacheStructureEpoch(). Starts at 1 so a zero-initialized memo can never match.
         Uint64 m_cacheStructureEpoch = 1;
         IEvictionObserver* m_evictionObserver = nullptr;
-        // Pass-through tessellation control stages by input patch size. Never evicted: at most
-        // MAX_PATCH_VERTICES entries exist for the lifetime of the device, and every pipeline
-        // ever built from one keeps referencing its module. A failed build is cached as
+        // Pass-through tessellation control stages by the identity of what was compiled into
+        // them - the input patch size and the six default tessellation levels, folded into one
+        // 64-bit key by ComputePassthroughTessControlKey (the levels are float state, so the map
+        // cannot simply be keyed on the patch size any more). A failed build is cached as
         // VK_NULL_HANDLE so a broken generator costs one compile, not one per draw.
-        UnorderedMap<Uint32, VkPipelineShaderStageCreateInfo> m_passthroughTessControlStages;
+        //
+        // Hard-capped, because the key is application-controlled: glPatchParameterfv clamps
+        // nothing, so an application that recomputes a level per frame mints a new key per frame.
+        // Reaching the cap destroys every module and starts over (see the flush in
+        // GetOrCreatePassthroughTessControlStage); the cap is far above what any program that
+        // holds its levels still will ever need. The gl_PerVertex member set is in the key too
+        // and adds only a handful of values, so it does not move the cap in practice.
+        static constexpr SizeT kMaxPassthroughTessControlStages = 64;
+        UnorderedMap<Uint64, VkPipelineShaderStageCreateInfo> m_passthroughTessControlStages;
         static inline XXH64_state_t* m_hashState = XXH64_createState();
     };
 } // namespace MobileGL::MG_Backend::DirectVulkan

MobileGL/MG_Backend/DirectVulkan/Renderer/UniformManager.cpp

@@ -11,13 +11,19 @@
 #include "MG_Backend/DirectVulkan/DirectVulkanResourceState.h"
 #include "MG_State/GLState/Core.h"
 #include "MG_State/GLState/ProgramState/ProgramObject.h"
+#include "MG_State/GLState/TextureState/TextureObject1D.h"
 #include "MG_State/GLState/TextureState/TextureObject2D.h"
+#include "MG_State/GLState/TextureState/TextureObject2DCube.h"
+#include "MG_State/GLState/TextureState/TextureObject3D.h"
 #include "MG_State/GLState/TextureState/TextureObjectBuffer.h"
+#include "MG_State/GLState/TextureState/TextureObjectStubs.h"
 #include "MG_Util/Converters/GLToMG/TextureEnumConverter.h"
 #include "MG_Util/Converters/MGToStr/FramebufferEnumConverter.h"
 #include "MG_Util/Converters/MGToVk/TextureEnumConverter.h"
 #include "MG_Util/Metrics/TextureMetrics.h"
+#include "MG_Util/ShaderTranspiler/Types.h"
 #include <Config.h>
+#include <vulkan/utility/vk_format_utils.h>
 #include <algorithm>
 #include <cstdio>
 #include <cstdlib>
@@ -27,6 +33,136 @@
 namespace MobileGL::MG_Backend::DirectVulkan {
     namespace {
         constexpr Uint kFallbackTexture2DExternalIndex = 0xFFFFFF00u;
+        // One id for every storage-image placeholder. They are never reachable through GL - no
+        // glGenTextures ever hands this out, and nothing looks a placeholder up by name - so the
+        // id only has to stay clear of the application's, exactly like the sampled fallback's.
+        constexpr Uint kUnboundStorageImageExternalIndex = 0xFFFFFF01u;
+
+        // The R32 member of each numeric class. Every one of the three is a MANDATORY-support
+        // format for uniform texel buffers, storage texel buffers and storage images alike
+        // (Vulkan 1.0, "Required Format Support"), which is what makes them a fallback that
+        // cannot itself fail for want of device features.
+        VkFormat PlaceholderFormatForNumericDomain(SamplerNumericDomain numericDomain) {
+            switch (numericDomain) {
+            case SamplerNumericDomain::Float:
+                return VK_FORMAT_R32_SFLOAT;
+            case SamplerNumericDomain::SignedInteger:
+                return VK_FORMAT_R32_SINT;
+            case SamplerNumericDomain::UnsignedInteger:
+                return VK_FORMAT_R32_UINT;
+            case SamplerNumericDomain::Unknown:
+                break;
+            }
+            return VK_FORMAT_UNDEFINED;
+        }
+
+        Bool BufferFormatSupportsFeature(VkPhysicalDevice physicalDevice, VkFormat format,
+                                         VkFormatFeatureFlags requiredFeature) {
+            if (physicalDevice == VK_NULL_HANDLE || format == VK_FORMAT_UNDEFINED) {
+                return false;
+            }
+            VkFormatProperties properties{};
+            vkGetPhysicalDeviceFormatProperties(physicalDevice, format, &properties);
+            return (properties.bufferFeatures & requiredFeature) == requiredFeature;
+        }
+
+        // Reverse of MG_Util::ConvertTextureInternalFormatToVkEnum. A placeholder texture is
+        // built through the ordinary frontend texture object (that is what gets it an image with
+        // STORAGE usage, a GENERAL transition and a view, for free), and that object is described
+        // by a GL internal format - while everything upstream of here speaks VkFormat. Scanned
+        // rather than tabulated: it runs once per (target, format) placeholder ever created, the
+        // enum is ~70 entries, and a second hand-written table is a second thing to drift.
+        // Ascending order matters: the sized formats precede the unsized aliases, so a scan
+        // answers with the sized one.
+        TextureInternalFormat InternalFormatForVkFormat(VkFormat format) {
+            if (format == VK_FORMAT_UNDEFINED) {
+                return TextureInternalFormat::Unknown;
+            }
+            for (Int index = 0; index < static_cast<Int>(TextureInternalFormat::TextureInternalFormatCount);
+                 ++index) {
+                const auto candidate = static_cast<TextureInternalFormat>(index);
+                if (MG_Util::ConvertTextureInternalFormatToVkEnum(candidate) == format) {
+                    return candidate;
+                }
+            }
+            return TextureInternalFormat::Unknown;
+        }
+
+        // What a 1x1 placeholder of a given target has to allocate for the backend to give it the
+        // Vulkan view type that target's image declaration demands (see
+        // VkTextureManager's TryResolveTextureShapeInfo, which reads exactly these two things).
+        struct PlaceholderShape {
+            Array<TextureUploadTarget, 6> uploadTargets{};
+            Uint32 uploadTargetCount = 0;
+            // The GL depth of the single level: the array length for an array target, the depth
+            // for a 3D one, and 6 for a cube map array (one whole cube).
+            Int depth = 1;
+            Bool valid = false;
+        };
+
+        PlaceholderShape PlaceholderShapeForTarget(TextureTarget target) {
+            PlaceholderShape shape{};
+            switch (target) {
+            case TextureTarget::Texture1D:
+                shape = {{TextureUploadTarget::Texture1D}, 1, 1, true};
+                break;
+            case TextureTarget::Texture2D:
+                shape = {{TextureUploadTarget::Texture2D}, 1, 1, true};
+                break;
+            case TextureTarget::TextureRectangle:
+                shape = {{TextureUploadTarget::TextureRectangle}, 1, 1, true};
+                break;
+            case TextureTarget::Texture3D:
+                shape = {{TextureUploadTarget::Texture3D}, 1, 1, true};
+                break;
+            case TextureTarget::Texture1DArray:
+                shape = {{TextureUploadTarget::Texture1DArray}, 1, 1, true};
+                break;
+            case TextureTarget::Texture2DArray:
+                shape = {{TextureUploadTarget::Texture2DArray}, 1, 1, true};
+                break;
+            case TextureTarget::TextureCubeMap:
+                shape = {{TextureUploadTarget::CubeMapPositiveX, TextureUploadTarget::CubeMapNegativeX,
+                          TextureUploadTarget::CubeMapPositiveY, TextureUploadTarget::CubeMapNegativeY,
+                          TextureUploadTarget::CubeMapPositiveZ, TextureUploadTarget::CubeMapNegativeZ},
+                         6, 1, true};
+                break;
+            case TextureTarget::TextureCubeMapArray:
+                // Layers are cube faces, so the count must be a whole number of cubes.
+                shape = {{TextureUploadTarget::CubeMapArray}, 1, 6, true};
+                break;
+            default:
+                // Multisample targets above all: their descriptor needs a multisample view.
+                break;
+            }
+            return shape;
+        }
+
+        // TextureObjectMipmap, not ITextureObject: AllocateStorage and MarkStorageDirty live
+        // there, and every placeholder shape above is one of its subclasses.
+        SharedPtr<MG_State::GLState::TextureObjectMipmap> MakePlaceholderTextureObject(TextureTarget target,
+                                                                                       Uint index) {
+            switch (target) {
+            case TextureTarget::Texture1D:
+                return MakeShared<MG_State::GLState::TextureObject1D>(index);
+            case TextureTarget::Texture2D:
+                return MakeShared<MG_State::GLState::TextureObject2D>(index);
+            case TextureTarget::TextureRectangle:
+                return MakeShared<MG_State::GLState::TextureObjectRectangle>(index);
+            case TextureTarget::Texture3D:
+                return MakeShared<MG_State::GLState::TextureObject3D>(index);
+            case TextureTarget::Texture1DArray:
+                return MakeShared<MG_State::GLState::TextureObject1DArray>(index);
+            case TextureTarget::Texture2DArray:
+                return MakeShared<MG_State::GLState::TextureObject2DArray>(index);
+            case TextureTarget::TextureCubeMap:
+                return MakeShared<MG_State::GLState::TextureObject2DCube>(index);
+            case TextureTarget::TextureCubeMapArray:
+                return MakeShared<MG_State::GLState::TextureObjectCubeMapArray>(index);
+            default:
+                return nullptr;
+            }
+        }
     }
 
     static Bool FindFramebufferAttachmentForTexture(const MG_State::GLState::FramebufferObject& framebuffer,
@@ -47,7 +183,8 @@ namespace MobileGL::MG_Backend::DirectVulkan {
             auto attachedTexture = attachment.GetTexture();
             if (attachedTexture && attachedTexture.get() == &texture) {
                 outAttachment = attachmentType;
-                outLevel = attachment.GetTextureLevel();
+                outLevel = static_cast<Int>(ToStorageMipLevel(attachment.GetTexture().get(),
+                                                             attachment.GetTextureLevel()));
                 return true;
             }
         }
@@ -113,7 +250,8 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         return reflectedFormat != VK_FORMAT_UNDEFINED ? reflectedFormat : resourceFormat;
     }
 
-    Bool UniformManager::Initialize(VkDevice device, VkBufferManager* bufferManager,
+    Bool UniformManager::Initialize(VkDevice device, VkPhysicalDevice physicalDevice,
+                                             VkBufferManager* bufferManager,
                                              ProgramFactory* programFactory,
                                              VkDeviceSize minUniformBufferOffsetAlignment, Uint32 frameCount,
                                              Uint32 maxBindings, Uint32 setsPerFrame,
@@ -121,6 +259,8 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         Shutdown();
 
         MOBILEGL_ASSERT(device != VK_NULL_HANDLE, "UniformDescriptorBinder::Initialize requires valid VkDevice");
+        MOBILEGL_ASSERT(physicalDevice != VK_NULL_HANDLE,
+                        "UniformDescriptorBinder::Initialize requires valid VkPhysicalDevice");
         MOBILEGL_ASSERT(bufferManager != nullptr, "UniformDescriptorBinder::Initialize requires valid buffer manager");
         MOBILEGL_ASSERT(programFactory != nullptr,
                         "UniformDescriptorBinder::Initialize requires valid program factory");
@@ -133,6 +273,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                         "UniformDescriptorBinder::Initialize requires valid sampler manager");
 
         m_device = device;
+        m_physicalDevice = physicalDevice;
         m_bufferManager = bufferManager;
         m_programFactory = programFactory;
         m_minDynamicOffsetAlignment = std::max<VkDeviceSize>(1, minUniformBufferOffsetAlignment);
@@ -171,6 +312,17 @@ namespace MobileGL::MG_Backend::DirectVulkan {
     }
 
     void UniformManager::Shutdown() {
+        // Before the per-frame loop, because these views are NOT owned by any frame slot (see
+        // m_unboundTexelBufferViews) and the loop below is what clears m_device.
+        if (m_device != VK_NULL_HANDLE) {
+            for (const auto& viewEntry : m_unboundTexelBufferViews) {
+                if (viewEntry.second != VK_NULL_HANDLE) {
+                    vkDestroyBufferView(m_device, viewEntry.second, nullptr);
+                }
+            }
+        }
+        m_unboundTexelBufferViews.clear();
+        m_unboundStorageImageTextures.clear();
         for (auto& frame : m_frames) {
             if (m_device != VK_NULL_HANDLE) {
                 for (auto& view : frame.texelBufferViews) {
@@ -197,6 +349,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         m_bufferManager = nullptr;
         m_programFactory = nullptr;
         m_device = VK_NULL_HANDLE;
+        m_physicalDevice = VK_NULL_HANDLE;
         m_minDynamicOffsetAlignment = 1;
         m_frameCount = 0;
         m_maxBindings = 0;
@@ -416,16 +569,25 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                                            numericDomain == SamplerNumericDomain::UnsignedInteger;
         SamplerResolveMemo* viewFormatMemo =
             binding < m_samplerResolveMemo.size() ? &m_samplerResolveMemo[binding] : nullptr;
+        // The format this GL texture presents to the shader. For a texture created by
+        // glTextureView that is the format the VIEW reinterpreted its storage as (GL 4.6 core
+        // 8.18), not the storage image's own - resolving the numeric domain against the latter
+        // would pick a sampled view for a format the shader never declared. The probe is behind
+        // IsTextureView() so nothing about the ordinary per-draw path changes.
+        const VkFormat sampledSourceFormat =
+            texture->IsTextureView()
+                ? m_textureManager->ResolveTextureViewWindow(*texture, *resource).format
+                : resource->format;
         VkFormat sampledViewFormat;
         if (viewFormatMemo != nullptr && viewFormatMemo->viewFormatValid &&
-            viewFormatMemo->viewFormatSource == resource->format &&
+            viewFormatMemo->viewFormatSource == sampledSourceFormat &&
             viewFormatMemo->viewFormatDomain == numericDomain) {
             sampledViewFormat = viewFormatMemo->viewFormat;
         } else {
             sampledViewFormat =
-                VkTextureManager::ResolveSampledImageViewFormat(resource->format, numericDomain);
+                VkTextureManager::ResolveSampledImageViewFormat(sampledSourceFormat, numericDomain);
             if (viewFormatMemo != nullptr) {
-                viewFormatMemo->viewFormatSource = resource->format;
+                viewFormatMemo->viewFormatSource = sampledSourceFormat;
                 viewFormatMemo->viewFormatDomain = numericDomain;
                 viewFormatMemo->viewFormat = sampledViewFormat;
                 viewFormatMemo->viewFormatValid = true;
@@ -440,9 +602,12 @@ namespace MobileGL::MG_Backend::DirectVulkan {
             return false;
         }
         // No reinterpretation requested: bind the depth-or-color aspect view the sync above
-        // already produced instead of re-entering GetOrCreateSampledImageView's sync path.
+        // already produced instead of re-entering GetOrCreateSampledImageView's sync path. A GL
+        // texture view is excluded because resource->sampledView belongs to the texture it VIEWS
+        // - same image, but the storage texture's level range and depth/stencil aspect, which is
+        // exactly the state a view exists to differ on.
         const VkImageView sampledImageView =
-            sampledViewFormat == resource->format
+            (!texture->IsTextureView() && sampledViewFormat == resource->format)
                 ? resource->sampledView
                 : m_textureManager->GetOrCreateSampledImageView(*texture, sampledViewFormat);
         if (sampledImageView == VK_NULL_HANDLE) {
@@ -547,7 +712,12 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                                                             resource->sampledLevelCount),
             .imageView = samplerBindingOverride.imageView != VK_NULL_HANDLE ?
                 samplerBindingOverride.imageView :
-                (resource->sampledView != VK_NULL_HANDLE ? resource->sampledView : resource->fullView),
+                // Same reason as in ResolveSamplerDescriptor: the resource's own views describe
+                // the storage texture, so a view has to be asked for its own.
+                (samplerBindingOverride.texture->IsTextureView()
+                     ? m_textureManager->GetOrCreateSampledImageView(*samplerBindingOverride.texture,
+                                                                     VK_FORMAT_UNDEFINED)
+                     : (resource->sampledView != VK_NULL_HANDLE ? resource->sampledView : resource->fullView)),
             .imageLayout = samplerBindingOverride.imageLayout != VK_IMAGE_LAYOUT_UNDEFINED ?
                 samplerBindingOverride.imageLayout : resource->layout,
         };
@@ -674,11 +844,29 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         MOBILEGL_ASSERT(m_bufferManager != nullptr, "ResolveTexelBufferDescriptor: buffer manager is null");
         MOBILEGL_ASSERT(frameIndex < m_frames.size(), "ResolveTexelBufferDescriptor: frame index out of range");
 
+        MOBILEGL_ASSERT(binding < programObj.samplerNumericDomainByBinding.size(),
+                        "ResolveTexelBufferDescriptor: numeric domain binding %u out of range", binding);
+        const SamplerNumericDomain numericDomain = programObj.samplerNumericDomainByBinding[binding];
+
         SharedPtr<MG_State::GLState::ITextureObject> texture;
         if (!ResolveSamplerTexture(program, programObj, binding, texture) || texture == nullptr) {
-            MGLOG_E_ONCE("ResolveTexelBufferDescriptor: texture buffer binding %u ('%s') is unbound", binding,
-                    programObj.samplerNameByBinding[binding].c_str());
-            return false;
+            // NOT an error, and not a reason to lose the draw. A texture unit with nothing on it
+            // is a legal GL state (4.6 core 8.24): the sampler is incomplete, so a fetch through
+            // it returns undefined values - the same answer the sampled path above gives with its
+            // fallback texture, which a buffer texture simply cannot use because its descriptor is
+            // a VkBufferView. A per-format placeholder view is the equivalent for this kind.
+            const VkBufferView placeholder =
+                AcquireUnboundTexelBufferView(VK_FORMAT_UNDEFINED, numericDomain, false);
+            if (placeholder == VK_NULL_HANDLE) {
+                MGLOG_E_ONCE("ResolveTexelBufferDescriptor: texture buffer binding %u ('%s') is unbound, and the "
+                        "placeholder descriptor could not be created", binding,
+                        programObj.samplerNameByBinding[binding].c_str());
+                return false;
+            }
+            MGLOG_D("ResolveTexelBufferDescriptor: binding %u ('%s') is unbound; using the placeholder descriptor",
+                    binding, programObj.samplerNameByBinding[binding].c_str());
+            outBufferView = placeholder;
+            return true;
         }
 
         if (texture->GetStorageType() != TextureStorageType::Buffer ||
@@ -693,9 +881,21 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         auto* textureBuffer = static_cast<MG_State::GLState::TextureObjectBuffer*>(texture.get());
         const auto& bufferObject = textureBuffer->GetBufferBindingSlot().GetBoundObject();
         if (bufferObject == nullptr) {
-            MGLOG_E_ONCE("ResolveTexelBufferDescriptor: texture buffer binding %u ('%s') has no GL buffer bound",
-                    binding, programObj.samplerNameByBinding[binding].c_str());
-            return false;
+            // A buffer texture with no buffer object attached is INCOMPLETE, not illegal (GL 4.6
+            // core 8.9), and sampling an incomplete texture is undefined - so this too keeps the
+            // draw on a placeholder rather than dropping it.
+            const VkBufferView placeholder =
+                AcquireUnboundTexelBufferView(VK_FORMAT_UNDEFINED, numericDomain, false);
+            if (placeholder == VK_NULL_HANDLE) {
+                MGLOG_E_ONCE("ResolveTexelBufferDescriptor: texture buffer binding %u ('%s') has no GL buffer bound, "
+                        "and the placeholder descriptor could not be created", binding,
+                        programObj.samplerNameByBinding[binding].c_str());
+                return false;
+            }
+            MGLOG_D("ResolveTexelBufferDescriptor: binding %u ('%s') has no attached GL buffer; using the "
+                    "placeholder descriptor", binding, programObj.samplerNameByBinding[binding].c_str());
+            outBufferView = placeholder;
+            return true;
         }
 
         BufferSlice slice{};
@@ -785,12 +985,31 @@ namespace MobileGL::MG_Backend::DirectVulkan {
             return false;
         }
 
+        MOBILEGL_ASSERT(binding < programObj.storageImageFormatByBinding.size(),
+                        "ResolveStorageTexelBufferDescriptor: binding %u has no reflected format slot", binding);
+        MOBILEGL_ASSERT(binding < programObj.samplerNumericDomainByBinding.size(),
+                        "ResolveStorageTexelBufferDescriptor: numeric domain binding %u out of range", binding);
+
         auto& imageBinding = MG_State::pGLContext->GetImageTextureBinding(imageUnit);
         const auto& texture = imageBinding.Texture;
         if (texture == nullptr) {
-            MGLOG_E_ONCE("ResolveStorageTexelBufferDescriptor: image unit %d is unbound for binding %u", imageUnit,
-                    binding);
-            return false;
+            // An image unit with no texture on it is legal GL (4.6 core 8.26): loads return zero
+            // and stores are discarded. Declining here took the whole draw or dispatch with it -
+            // the same shape as the unbound storage block fixed alongside this. A placeholder view
+            // in the shader's own declared format lets the work proceed with the stores landing
+            // nowhere anyone can observe, which is what GL asks for.
+            const VkBufferView placeholder =
+                AcquireUnboundTexelBufferView(programObj.storageImageFormatByBinding[binding],
+                                              programObj.samplerNumericDomainByBinding[binding], true);
+            if (placeholder == VK_NULL_HANDLE) {
+                MGLOG_E_ONCE("ResolveStorageTexelBufferDescriptor: image unit %d is unbound for binding %u, and the "
+                        "placeholder descriptor could not be created", imageUnit, binding);
+                return false;
+            }
+            MGLOG_D("ResolveStorageTexelBufferDescriptor: image unit %d (binding %u) is unbound; using the "
+                    "placeholder descriptor", imageUnit, binding);
+            outBufferView = placeholder;
+            return true;
         }
         if (texture->GetStorageType() != TextureStorageType::Buffer ||
             texture->GetTarget() != TextureTarget::TextureBuffer) {
@@ -805,9 +1024,20 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         auto* textureBuffer = static_cast<MG_State::GLState::TextureObjectBuffer*>(texture.get());
         const auto& bufferObject = textureBuffer->GetBufferBindingSlot().GetBoundObject();
         if (bufferObject == nullptr) {
-            MGLOG_E_ONCE("ResolveStorageTexelBufferDescriptor: texture buffer on image unit %d has no GL buffer bound",
-                    imageUnit);
-            return false;
+            // Incomplete buffer texture, same as the sampled path: legal state, undefined data,
+            // and no reason to drop the work.
+            const VkBufferView placeholder =
+                AcquireUnboundTexelBufferView(programObj.storageImageFormatByBinding[binding],
+                                              programObj.samplerNumericDomainByBinding[binding], true);
+            if (placeholder == VK_NULL_HANDLE) {
+                MGLOG_E_ONCE("ResolveStorageTexelBufferDescriptor: texture buffer on image unit %d has no GL buffer "
+                        "bound, and the placeholder descriptor could not be created", imageUnit);
+                return false;
+            }
+            MGLOG_D("ResolveStorageTexelBufferDescriptor: texture buffer on image unit %d has no attached GL buffer; "
+                    "using the placeholder descriptor", imageUnit);
+            outBufferView = placeholder;
+            return true;
         }
 
         // Unlike the sampled texel buffer, the shader MAY write this one, and those writes land
@@ -833,8 +1063,6 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         // policy as a storage image: a typed `layout(r32ui) uniform uimageBuffer` must be read as
         // r32ui whatever the texture's own attachment format says. Falling back, in order:
         // reflected format, then the bind format, then the texture's attached format.
-        MOBILEGL_ASSERT(binding < programObj.storageImageFormatByBinding.size(),
-                        "ResolveStorageTexelBufferDescriptor: binding %u has no reflected format slot", binding);
         const auto internalFormat = textureBuffer->GetFormat();
         const VkFormat resourceFormat = MG_Util::ConvertTextureInternalFormatToVkEnum(internalFormat);
         const VkFormat reflectedFormat = programObj.storageImageFormatByBinding[binding];
@@ -905,23 +1133,69 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         const Int blockIndex = programObj.storageBlockIndexByBinding[binding];
         MOBILEGL_ASSERT(blockIndex >= 0, "ResolveStorageBufferDescriptor: no SSBO block mapped to binding %u",
                         binding);
+        // An atomic counter is not an SSBO the application ever declared: glslang lowers every
+        // atomic_uint onto a synthesized gl_AtomicCounterBlock_<N> storage block, where N is the
+        // GL ATOMIC-COUNTER binding. That block arrives here auto-mapped to an arbitrary
+        // storage-block slot, so resolving it the SSBO way looked up GL_SHADER_STORAGE_BUFFER
+        // point N' - which is never where glBindBufferBase(GL_ATOMIC_COUNTER_BUFFER, N, ...) put
+        // the buffer. The counter therefore never reached the shader (KHR-GL43
+        // shader_atomic_counters.advanced-usage-*), and when the application also bound an SSBO at
+        // the colliding slot the descriptor silently aliased it, so the dispatch wrote over the
+        // application's own buffer. DirectGLES has always taken this branch explicitly
+        // (SyncAtomicCounterBuffers); this is the same rule in Magma's descriptor resolution.
+        //
+        // Only the SOURCE of the handle differs. The per-counter layout(offset=) is already folded
+        // into the block's SPIR-V member offsets on this path (FlattenAtomicCounterBlockPass is
+        // DirectGLES-only), so everything below - residency, the glBindBufferRange window, the
+        // descriptor fill - is target-agnostic and stays exactly as it was.
+        const String& blockName = programObj.storageBlockNameByBinding[binding];
+        const Int atomicCounterBinding = MG_Util::ShaderTranspiler::AtomicCounterBlockGlBinding(blockName);
+        const Bool isAtomicCounterBlock = atomicCounterBinding >= 0;
+        const BufferTarget bufferTarget =
+            isAtomicCounterBlock ? BufferTarget::AtomicCounter : BufferTarget::ShaderStorage;
         // A block instance array declares one block whose elements take consecutive GL binding
         // points from the declared one (GL 4.6 core 7.8), and the reflection collapses the whole
-        // array to that one block - so the element index IS the offset from its binding.
+        // array to that one block - so the element index IS the offset from its binding. glslang
+        // synthesizes one counter block per GL binding, so a counter block is never an instance
+        // array and `element` is always 0 there; the +element rule stays with the SSBO case.
         const GLuint frontendBinding =
-            GetShaderStorageBlockBinding(program, static_cast<GLuint>(blockIndex)) + element;
+            isAtomicCounterBlock
+                ? static_cast<GLuint>(atomicCounterBinding)
+                : GetShaderStorageBlockBinding(program, static_cast<GLuint>(blockIndex)) + element;
         const Uint32 bindingPointCount =
-            static_cast<Uint32>(MG_State::pGLContext->GetBufferBindingPointCount(BufferTarget::ShaderStorage));
+            static_cast<Uint32>(MG_State::pGLContext->GetBufferBindingPointCount(bufferTarget));
         MOBILEGL_ASSERT(frontendBinding < bindingPointCount,
-                        "ResolveStorageBufferDescriptor: frontend SSBO binding %u out of range for block '%s'",
-                        frontendBinding, programObj.storageBlockNameByBinding[binding].c_str());
+                        "ResolveStorageBufferDescriptor: frontend binding %u out of range for block '%s'",
+                        frontendBinding, blockName.c_str());
 
-        auto& bindingPoint = MG_State::pGLContext->GetBufferBindingPoint(BufferTarget::ShaderStorage, frontendBinding);
+        auto& bindingPoint = MG_State::pGLContext->GetBufferBindingPoint(bufferTarget, frontendBinding);
         const auto& bufferObject = bindingPoint.GetBoundObject();
         if (bufferObject == nullptr) {
-            MGLOG_E_ONCE("ResolveStorageBufferDescriptor: no SSBO bound at frontend binding %u for block '%s'",
-                    frontendBinding, programObj.storageBlockNameByBinding[binding].c_str());
-            return false;
+            // NOT an error, and above all not a reason to lose the draw. GL 4.6 core 7.8 lets a
+            // program declare a shader storage block the application never binds a buffer to:
+            // the block simply has no store, so a read is undefined and a write goes nowhere.
+            // Refusing here used to take the whole draw or dispatch with it (SetupDraw and
+            // DispatchCompute both skip on a false return), which is how AcceleratedRendering's
+            // GUI batch lost its backgrounds: its vertex-transform compute shader declares a
+            // `Meshes` block it only reads when a vertex comes from a cached server mesh, and a
+            // batch of plain GUI blits has no meshes and so binds nothing there. The dispatch
+            // never ran, the transformed vertex buffer stayed as it was, and every hotbar and
+            // container-screen background quad came out degenerate. A shared zero-filled
+            // placeholder puts something legal in the descriptor and lets the draw proceed.
+            const BufferSlice placeholder = m_bufferManager->AcquireUnboundStorageDescriptor();
+            if (!placeholder.IsValid()) {
+                MGLOG_E_ONCE("ResolveStorageBufferDescriptor: no buffer bound at frontend binding %u for block "
+                             "'%s', and the placeholder descriptor could not be created",
+                             frontendBinding, blockName.c_str());
+                return false;
+            }
+            MGLOG_D("ResolveStorageBufferDescriptor: frontend binding %u ('%s') is unbound; using the placeholder "
+                    "descriptor",
+                    frontendBinding, blockName.c_str());
+            outBufferInfo.buffer = placeholder.buffer;
+            outBufferInfo.offset = placeholder.offset;
+            outBufferInfo.range = placeholder.size;
+            return true;
         }
 
         // The shader may write this buffer, and those writes land in GPU memory behind the
@@ -997,8 +1271,40 @@ namespace MobileGL::MG_Backend::DirectVulkan {
 
         auto& imageBinding = MG_State::pGLContext->GetImageTextureBinding(imageUnit);
         if (imageBinding.Texture == nullptr) {
-            MGLOG_E_ONCE("ResolveStorageImageDescriptor: image unit %d is unbound for binding %u", imageUnit, binding);
-            return false;
+            // Legal GL: an image unit with no texture bound makes loads return zero and discards
+            // stores (4.6 core 8.26). It is not a reason to lose the draw, which is what returning
+            // false here did - both SetupDraw and DispatchCompute skip everything on it. The
+            // placeholder is a 1x1 image of the target and format the shader's declaration asks
+            // for, so the descriptor is valid and the stores land where nobody can see them.
+            TextureTarget placeholderTarget = TextureTarget::Unknown;
+            VkFormat placeholderFormat = VK_FORMAT_UNDEFINED;
+            SharedPtr<MG_State::GLState::ITextureObject> placeholder;
+            if (ResolveUnboundStorageImagePlaceholder(programObj, binding, placeholderTarget, placeholderFormat)) {
+                placeholder = GetUnboundStorageImageTexture(placeholderTarget, placeholderFormat);
+            }
+            VkImageView placeholderView = VK_NULL_HANDLE;
+            if (placeholder != nullptr &&
+                m_textureManager->TransitionTextureForStorageImage(commandBuffer, *placeholder)) {
+                // layered=true, layer=0: the placeholder's own view type IS the one the shader's
+                // image declaration demands, and that is exactly what the layered form asks for
+                // (see GetOrCreateStorageImageView, which only narrows the view type when a
+                // non-layered binding names a single layer).
+                placeholderView =
+                    m_textureManager->GetOrCreateStorageImageView(*placeholder, 0, placeholderFormat, true, 0);
+            }
+            if (placeholderView == VK_NULL_HANDLE) {
+                MGLOG_E_ONCE("ResolveStorageImageDescriptor: image unit %d is unbound for binding %u, and no "
+                        "placeholder descriptor could be built (target=%d format=%d)",
+                        imageUnit, binding, static_cast<Int>(placeholderTarget),
+                        static_cast<Int>(placeholderFormat));
+                return false;
+            }
+            MGLOG_D("ResolveStorageImageDescriptor: image unit %d (binding %u) is unbound; using the placeholder "
+                    "descriptor", imageUnit, binding);
+            outImageInfo.sampler = VK_NULL_HANDLE;
+            outImageInfo.imageView = placeholderView;
+            outImageInfo.imageLayout = VK_IMAGE_LAYOUT_GENERAL;
+            return true;
         }
 
         const Bool ready = m_textureManager->TransitionTextureForStorageImage(commandBuffer, *imageBinding.Texture);
@@ -1020,8 +1326,15 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                         binding);
         const VkFormat reflectedFormat = programObj.storageImageFormatByBinding[binding];
         const Bool useBindingFormat = programObj.storageImageUsesBindingFormatByBinding[binding];
+        // The storage's own VkFormat is the wrong reference for a GL texture view: the view
+        // reinterprets it (GL 4.6 core table 8.21), and it is the VIEW's format the shader's
+        // image declaration was written against. Same correction the sampled path makes above.
+        const VkFormat storageImageSourceFormat =
+            imageBinding.Texture->IsTextureView()
+                ? m_textureManager->ResolveTextureViewWindow(*imageBinding.Texture, *resource).format
+                : resource->format;
         const VkFormat viewFormat = ResolveStorageImageViewFormat(
-            reflectedFormat, imageBinding.Format, resource->format, useBindingFormat);
+            reflectedFormat, imageBinding.Format, storageImageSourceFormat, useBindingFormat);
         if (viewFormat == VK_FORMAT_UNDEFINED) {
             MGLOG_E_ONCE("ResolveStorageImageDescriptor: unsupported glBindImageTexture format=0x%x "
                     "for binding=%u imageUnit=%d textureId=%d bindingPolicy=%s",
@@ -1029,8 +1342,16 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                     useBindingFormat ? "true" : "false");
             return false;
         }
+        // glBindImageTexture named a level and a layer of the bound texture; on a GL texture
+        // view both are relative to the view, and the storage image is what the descriptor
+        // actually points at (see ToStorageMipLevel).
+        const Int32 storageImageLayer =
+            imageBinding.Layered != GL_FALSE
+                ? imageBinding.Layer
+                : static_cast<Int32>(ToStorageArrayLayer(imageBinding.Texture.get(), imageBinding.Layer));
         const VkImageView view = m_textureManager->GetOrCreateStorageImageView(
-            *imageBinding.Texture, mipLevel, viewFormat, imageBinding.Layered != GL_FALSE, imageBinding.Layer);
+            *imageBinding.Texture, ToStorageMipLevel(imageBinding.Texture.get(), static_cast<Int>(mipLevel)),
+            viewFormat, imageBinding.Layered != GL_FALSE, storageImageLayer);
         if (view == VK_NULL_HANDLE) {
             MGLOG_E_ONCE("ResolveStorageImageDescriptor: failed to resolve storage view textureId=%d mip=%u "
                     "bindingFormat=0x%x imageFormat=%d reflectedFormat=%d selectedFormat=%d bindingPolicy=%s",
@@ -1075,6 +1396,138 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         return m_fallbackTexture2D;
     }
 
+    VkBufferView UniformManager::AcquireUnboundTexelBufferView(VkFormat declaredFormat,
+                                                               SamplerNumericDomain numericDomain, Bool storage) {
+        MOBILEGL_ASSERT(m_bufferManager != nullptr, "AcquireUnboundTexelBufferView: buffer manager is null");
+        const VkFormatFeatureFlags requiredFeature = storage ? VK_FORMAT_FEATURE_STORAGE_TEXEL_BUFFER_BIT
+                                                             : VK_FORMAT_FEATURE_UNIFORM_TEXEL_BUFFER_BIT;
+        const VkFormat fallbackFormat = PlaceholderFormatForNumericDomain(numericDomain);
+
+        VkFormat format = declaredFormat;
+        if (format == VK_FORMAT_UNDEFINED || !BufferFormatSupportsFeature(m_physicalDevice, format, requiredFeature)) {
+            // The declared format is what a shader that WRITES through this descriptor is
+            // validated against, so it is tried first and kept whenever the device can use it.
+            // Falling back is for the two cases where it cannot be: a sampled texel buffer, which
+            // declares no format at all, and a device that does not list the declared one as a
+            // texel buffer. The fallback stays inside the shader's numeric class, which is the
+            // part the descriptor is checked on for a formatless declaration - and the R32
+            // members of the three classes are mandatory-support formats, so this cannot fail for
+            // want of device features.
+            format = fallbackFormat;
+        }
+        if (format == VK_FORMAT_UNDEFINED || !BufferFormatSupportsFeature(m_physicalDevice, format, requiredFeature)) {
+            MGLOG_E_ONCE("AcquireUnboundTexelBufferView: no usable placeholder format (declared=%d fallback=%d "
+                         "storage=%s)",
+                         static_cast<Int>(declaredFormat), static_cast<Int>(fallbackFormat),
+                         storage ? "true" : "false");
+            return VK_NULL_HANDLE;
+        }
+
+        const Uint64 key = (static_cast<Uint64>(format) << 1) | (storage ? 1ull : 0ull);
+        const auto cached = m_unboundTexelBufferViews.find(key);
+        if (cached != m_unboundTexelBufferViews.end()) {
+            return cached->second;
+        }
+
+        const BufferSlice placeholder = m_bufferManager->AcquireUnboundTexelBufferDescriptor();
+        if (!placeholder.IsValid()) {
+            MGLOG_E_ONCE("AcquireUnboundTexelBufferView: placeholder buffer unavailable");
+            return VK_NULL_HANDLE;
+        }
+        // A buffer view's range must be a whole number of texels of its own format, and the
+        // placeholder is sized for the largest of them - so floor rather than assume.
+        const VkDeviceSize texelSize = std::max<VkDeviceSize>(1, vkuFormatTexelBlockSize(format));
+        const VkDeviceSize range = (placeholder.size / texelSize) * texelSize;
+        if (range == 0) {
+            MGLOG_E_ONCE("AcquireUnboundTexelBufferView: placeholder holds no whole texel of format=%d",
+                         static_cast<Int>(format));
+            return VK_NULL_HANDLE;
+        }
+
+        VkBufferViewCreateInfo viewInfo{};
+        viewInfo.sType = VK_STRUCTURE_TYPE_BUFFER_VIEW_CREATE_INFO;
+        viewInfo.buffer = placeholder.buffer;
+        viewInfo.format = format;
+        viewInfo.offset = placeholder.offset;
+        viewInfo.range = range;
+
+        VkBufferView view = VK_NULL_HANDLE;
+        const VkResult result = vkCreateBufferView(m_device, &viewInfo, nullptr, &view);
+        if (result != VK_SUCCESS || view == VK_NULL_HANDLE) {
+            MGLOG_E_ONCE("AcquireUnboundTexelBufferView: vkCreateBufferView failed result=%d format=%d", result,
+                         static_cast<Int>(format));
+            return VK_NULL_HANDLE;
+        }
+        m_unboundTexelBufferViews.emplace(key, view);
+        MGLOG_D("AcquireUnboundTexelBufferView: created placeholder view format=%d storage=%s",
+                static_cast<Int>(format), storage ? "true" : "false");
+        return view;
+    }
+
+    Bool UniformManager::ResolveUnboundStorageImagePlaceholder(const ProgramFactory::VkProgramObject& programObj,
+                                                               Uint32 binding, TextureTarget& outTarget,
+                                                               VkFormat& outFormat) const {
+        MOBILEGL_ASSERT(binding < programObj.samplerTextureTargetByBinding.size(),
+                        "ResolveUnboundStorageImagePlaceholder: binding %u out of range", binding);
+        MOBILEGL_ASSERT(binding < programObj.storageImageFormatByBinding.size(),
+                        "ResolveUnboundStorageImagePlaceholder: format binding %u out of range", binding);
+        outTarget = programObj.samplerTextureTargetByBinding[binding];
+        // The shader's own format qualifier, exactly as the bound path prefers it over the one
+        // glBindImageTexture named - there is no binding here to name one. A `writeonly` image
+        // may carry no qualifier at all; its numeric class is then the only constraint, and the
+        // R32 member of that class is what carries it (see AcquireUnboundTexelBufferView).
+        outFormat = programObj.storageImageFormatByBinding[binding];
+        if (outFormat == VK_FORMAT_UNDEFINED) {
+            outFormat = PlaceholderFormatForNumericDomain(programObj.samplerNumericDomainByBinding[binding]);
+        }
+        return outFormat != VK_FORMAT_UNDEFINED && PlaceholderShapeForTarget(outTarget).valid;
+    }
+
+    SharedPtr<MG_State::GLState::ITextureObject> UniformManager::GetUnboundStorageImageTexture(
+        TextureTarget target, VkFormat format) const {
+        const Uint64 key = (static_cast<Uint64>(target) << 32) | static_cast<Uint32>(format);
+        const auto cached = m_unboundStorageImageTextures.find(key);
+        if (cached != m_unboundStorageImageTextures.end()) {
+            return cached->second;
+        }
+
+        const PlaceholderShape shape = PlaceholderShapeForTarget(target);
+        if (!shape.valid) {
+            // A multisample image uniform is the case with no answer here: its descriptor demands
+            // a multisample view, and a single-sampled 1x1 image is invalid Vulkan in that slot,
+            // not a degraded picture. The caller declines the binding exactly as it did before.
+            MGLOG_D("GetUnboundStorageImageTexture: no placeholder shape for target=%d", static_cast<Int>(target));
+            return nullptr;
+        }
+        const TextureInternalFormat internalFormat = InternalFormatForVkFormat(format);
+        if (internalFormat == TextureInternalFormat::Unknown) {
+            MGLOG_E_ONCE("GetUnboundStorageImageTexture: no GL internal format matches VkFormat=%d",
+                         static_cast<Int>(format));
+            return nullptr;
+        }
+
+        auto texture = MakePlaceholderTextureObject(target, kUnboundStorageImageExternalIndex);
+        if (texture == nullptr) {
+            return nullptr;
+        }
+        texture->SetInternalFormat(internalFormat);
+        const SizeT texelBytes = MG_Util::GetSizedInternalFormatSizeInBytes(internalFormat);
+        for (Uint32 index = 0; index < shape.uploadTargetCount; ++index) {
+            texture->AllocateStorage(shape.uploadTargets[index], 0,
+                                     {.texelSize = {1, 1, shape.depth},
+                                      .byteSize = texelBytes * static_cast<SizeT>(shape.depth)});
+            // Not dirty: there is deliberately nothing to upload. The image is created and
+            // transitioned to GENERAL by the storage-image preparation pass like any other, and
+            // its contents are exactly as undefined as GL says a fetch through an unbound image
+            // unit is.
+            texture->MarkStorageDirty(shape.uploadTargets[index], 0, false);
+        }
+        m_unboundStorageImageTextures.emplace(key, texture);
+        MGLOG_D("GetUnboundStorageImageTexture: created placeholder target=%d format=%d", static_cast<Int>(target),
+                static_cast<Int>(format));
+        return texture;
+    }
+
     Bool UniformManager::ResolveSampledBinding(const MG_State::GLState::ProgramObject& program,
                                                const ProgramFactory::VkProgramObject& programObj,
                                                Uint32 binding, Uint32 element,
@@ -1260,9 +1713,23 @@ namespace MobileGL::MG_Backend::DirectVulkan {
 
                 auto* texture = MG_State::pGLContext->GetImageTextureBinding(imageUnit).Texture.get();
                 if (texture == nullptr) {
-                    MGLOG_E_ONCE("CollectStorageImageTextures: image unit %d is unbound for binding %u element %u",
-                            imageUnit, binding, element);
-                    return false;
+                    // ResolveStorageImageDescriptor will substitute the placeholder image for this
+                    // binding; include it here for the same reason the sampled walk includes the
+                    // fallback texture - this walk is what gets a storage image created,
+                    // STORAGE-usage-marked and transitioned to GENERAL BEFORE the render pass
+                    // opens, and all three of those are illegal once it has. A target with no
+                    // placeholder shape (multisample) contributes nothing and is declined at
+                    // resolve time exactly as it was.
+                    TextureTarget placeholderTarget = TextureTarget::Unknown;
+                    VkFormat placeholderFormat = VK_FORMAT_UNDEFINED;
+                    if (!ResolveUnboundStorageImagePlaceholder(programObj, binding, placeholderTarget,
+                                                               placeholderFormat)) {
+                        continue;
+                    }
+                    texture = GetUnboundStorageImageTexture(placeholderTarget, placeholderFormat).get();
+                    if (texture == nullptr) {
+                        continue;
+                    }
                 }
                 if (std::find(outTextures.begin(), outTextures.end(), texture) == outTextures.end()) {
                     outTextures.push_back(texture);

MobileGL/MG_Backend/DirectVulkan/Renderer/UniformManager.h

@@ -42,7 +42,10 @@ namespace MobileGL::MG_Backend::DirectVulkan {
             SamplerNumericDomain numericDomain = SamplerNumericDomain::Unknown;
         };
 
-        Bool Initialize(VkDevice device, VkBufferManager* bufferManager,
+        // `physicalDevice` is only ever asked for format properties: a placeholder descriptor for
+        // an unbound texel-buffer binding has to be built from a format the DEVICE accepts as a
+        // texel buffer, and there is no other route to that answer from here.
+        Bool Initialize(VkDevice device, VkPhysicalDevice physicalDevice, VkBufferManager* bufferManager,
                         ProgramFactory* programFactory,
                         VkDeviceSize minUniformBufferOffsetAlignment, Uint32 frameCount,
                         Uint32 maxBindings = 16, Uint32 setsPerFrame = 64,
@@ -177,6 +180,32 @@ namespace MobileGL::MG_Backend::DirectVulkan {
             const MG_State::GLState::ProgramObject& program,
             const ProgramFactory::VkProgramObject& programObj, Uint32 binding, Uint32 element);
         SharedPtr<MG_State::GLState::ITextureObject> GetFallbackTexture(TextureTarget target) const;
+        // ---- placeholders for UNBOUND image-backed descriptors -------------------------
+        // GL lets a program declare `samplerBuffer`, `imageBuffer` or `image2D` and bind nothing
+        // to the unit it names: the fetch is then undefined (GL 4.6 core 8.9 for an incomplete
+        // buffer texture, 8.26 for an image unit with no texture) - undefined VALUES, not a
+        // dropped draw. Vulkan has no unwritten descriptor, so something valid has to sit in the
+        // set or the whole draw or dispatch is lost, which is what these two build. Same shape as
+        // VkBufferManager::AcquireUnboundStorageDescriptor, one level up: per FORMAT rather than
+        // one shared object, because a descriptor whose format disagrees with the shader's
+        // declaration is invalid Vulkan even when nothing ever reads it.
+        //
+        // `declaredFormat` is the format the SHADER declared (VK_FORMAT_UNDEFINED for a sampled
+        // texel buffer, which never carries one, or for a formatless `writeonly` image);
+        // `numericDomain` decides the format when there is no declaration and is the fallback
+        // class when the device cannot use the declared one as a texel buffer.
+        VkBufferView AcquireUnboundTexelBufferView(VkFormat declaredFormat, SamplerNumericDomain numericDomain,
+                                                   Bool storage);
+        // A 1x1 (x1 layer, or 6 faces for a cube) texture of `format`, shaped for `target` so the
+        // view the descriptor gets has the view type the shader's image declaration demands.
+        // Null for a target with no single-sampled placeholder shape - multisample images, whose
+        // descriptor needs a multisample view that this cannot stand in for.
+        SharedPtr<MG_State::GLState::ITextureObject> GetUnboundStorageImageTexture(TextureTarget target,
+                                                                                    VkFormat format) const;
+        // The (target, format) pair a storage-image binding's placeholder is keyed by, resolved
+        // from reflection alone. False when the binding has no placeholder shape.
+        Bool ResolveUnboundStorageImagePlaceholder(const ProgramFactory::VkProgramObject& programObj, Uint32 binding,
+                                                   TextureTarget& outTarget, VkFormat& outFormat) const;
         // `element` indexes a sampler ARRAY inside one binding; each element carries its own
         // independently assigned GL texture unit, so it selects the texture, the sampler
         // override and the fallback separately from its neighbours.
@@ -251,6 +280,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                                       VkDescriptorSet& outDescriptorSet);
 
         VkDevice m_device = VK_NULL_HANDLE;
+        VkPhysicalDevice m_physicalDevice = VK_NULL_HANDLE;
         VkBufferManager* m_bufferManager = nullptr;
         ProgramFactory* m_programFactory = nullptr;
         Vector<FrameResources> m_frames;
@@ -263,6 +293,15 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         VkTextureManager* m_textureManager = nullptr;
         VkSamplerManager* m_samplerManager = nullptr;
         mutable SharedPtr<MG_State::GLState::ITextureObject> m_fallbackTexture2D;
+        // See AcquireUnboundTexelBufferView / GetUnboundStorageImageTexture. Both are lazily
+        // populated, never evicted (a program's declared formats are a fixed, tiny set) and torn
+        // down with the manager. The texel views are keyed by format AND by storage-vs-sampled
+        // because the two descriptor kinds demand different format FEATURES of the device, so one
+        // format can be usable for one and not the other. Deliberately NOT the per-frame
+        // texelBufferViews list: those are destroyed at every frame boundary, and these must
+        // outlive it or the placeholder would be rebuilt for every unbound binding every frame.
+        UnorderedMap<Uint64, VkBufferView> m_unboundTexelBufferViews;
+        mutable UnorderedMap<Uint64, SharedPtr<MG_State::GLState::ITextureObject>> m_unboundStorageImageTextures;
 
         // Per-draw scratch buffers for BindProgramUniformBuffers: reused (clear keeps
         // capacity) so the descriptor-write path stops allocating on every draw.

MobileGL/MG_Backend/DirectVulkan/Renderer/VkBufferManager.cpp

@@ -16,6 +16,15 @@ namespace MobileGL::MG_Backend::DirectVulkan {
             VMA_ALLOCATION_CREATE_HOST_ACCESS_SEQUENTIAL_WRITE_BIT;
         constexpr SizeT kLiveResourcePruneThreshold = 256;
 
+        // See VkBufferManager::AcquireUnboundStorageDescriptor. 256 bytes: comfortably past
+        // every minStorageBufferOffsetAlignment in the wild, and free.
+        constexpr VkDeviceSize kUnboundStorageDescriptorBytes = 256;
+        // See VkBufferManager::AcquireUnboundTexelBufferDescriptor. The same 256 bytes, for the
+        // same reason plus one: a texel buffer view's range must be a whole number of texels of
+        // whatever format the placeholder is asked for, and 256 divides by every texel size in
+        // the GL image-format table (1, 2, 4, 8 and 16 bytes).
+        constexpr VkDeviceSize kUnboundTexelBufferDescriptorBytes = 256;
+
         // A zero-copy persistent buffer is created once and never recreated (the app holds
         // its mapped pointer), and may be bound to any role, so it carries every usage.
         // TRANSFER_DST is added by CreateResidentStorage.
@@ -130,6 +139,8 @@ namespace MobileGL::MG_Backend::DirectVulkan {
             }
         }
         m_transientUploadArena.Shutdown();
+        m_unboundStorageBuffer.Destroy();
+        m_unboundTexelBuffer.Destroy();
         DestroyAllDeferredReleases();
         ReleaseAllLiveResources();
         m_copyProvider = nullptr;
@@ -706,6 +717,70 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         m_deferredResourceReleases[frameIndex].clear();
     }
 
+    BufferSlice VkBufferManager::AcquireUnboundStorageDescriptor() {
+        if (!m_unboundStorageBuffer.IsValid()) {
+            if (m_initInfo.allocator == nullptr) {
+                return {};
+            }
+            // Host-visible so the zero fill needs no command buffer: this can be reached from
+            // descriptor resolution, which runs inside an already-open recording and must not
+            // start a copy of its own. The size is a whole minStorageBufferOffsetAlignment-safe
+            // block rather than 4 bytes so that a shader which does read the block gets a
+            // plausible unsized-array length instead of one that rounds to zero.
+            const Bool created = m_unboundStorageBuffer.Create({
+                .allocator = m_initInfo.allocator,
+                .size = kUnboundStorageDescriptorBytes,
+                .usage = VK_BUFFER_USAGE_STORAGE_BUFFER_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT,
+                .memoryUsage = VMA_MEMORY_USAGE_AUTO,
+                .allocationFlags = VMA_ALLOCATION_CREATE_HOST_ACCESS_SEQUENTIAL_WRITE_BIT |
+                                   VMA_ALLOCATION_CREATE_MAPPED_BIT,
+                .requiredFlags = VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT,
+            });
+            if (!created) {
+                MGLOG_E_ONCE("VkBufferManager::AcquireUnboundStorageDescriptor: placeholder creation failed");
+                m_unboundStorageBuffer.Destroy();
+                return {};
+            }
+            if (void* mapped = m_unboundStorageBuffer.GetMappedData()) {
+                Memset(mapped, 0, static_cast<SizeT>(kUnboundStorageDescriptorBytes));
+            }
+        }
+        return m_unboundStorageBuffer.GetSlice();
+    }
+
+    BufferSlice VkBufferManager::AcquireUnboundTexelBufferDescriptor() {
+        if (!m_unboundTexelBuffer.IsValid()) {
+            if (m_initInfo.allocator == nullptr) {
+                return {};
+            }
+            // A SECOND placeholder rather than more usage bits on the storage-block one. The two
+            // are independent failure domains: a device that refuses this allocation must not
+            // take the storage-block placeholder - and with it the fix this one is a sibling of -
+            // down with it. Host-visible and zero-filled for the same reason as that one: this is
+            // reached from descriptor resolution, inside an already-open recording, which must
+            // not start a copy of its own.
+            const Bool created = m_unboundTexelBuffer.Create({
+                .allocator = m_initInfo.allocator,
+                .size = kUnboundTexelBufferDescriptorBytes,
+                .usage = VK_BUFFER_USAGE_UNIFORM_TEXEL_BUFFER_BIT | VK_BUFFER_USAGE_STORAGE_TEXEL_BUFFER_BIT |
+                         VK_BUFFER_USAGE_TRANSFER_DST_BIT,
+                .memoryUsage = VMA_MEMORY_USAGE_AUTO,
+                .allocationFlags = VMA_ALLOCATION_CREATE_HOST_ACCESS_SEQUENTIAL_WRITE_BIT |
+                                   VMA_ALLOCATION_CREATE_MAPPED_BIT,
+                .requiredFlags = VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT,
+            });
+            if (!created) {
+                MGLOG_E_ONCE("VkBufferManager::AcquireUnboundTexelBufferDescriptor: placeholder creation failed");
+                m_unboundTexelBuffer.Destroy();
+                return {};
+            }
+            if (void* mapped = m_unboundTexelBuffer.GetMappedData()) {
+                Memset(mapped, 0, static_cast<SizeT>(kUnboundTexelBufferDescriptorBytes));
+            }
+        }
+        return m_unboundTexelBuffer.GetSlice();
+    }
+
     VkBufferUsageFlags VkBufferManager::GetVkBufferUsage(BufferKind kind) {
         switch (kind) {
         case BufferKind::Vertex:

MobileGL/MG_Backend/DirectVulkan/Renderer/VkBufferManager.h

@@ -120,6 +120,27 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         Bool UploadTransient(BufferKind kind, Uint32 frameIndex, const void* data, VkDeviceSize size,
                              VkDeviceSize alignment, BufferSlice& outSlice);
 
+        // The descriptor a shader storage block gets when the program declares it and the
+        // application bound no buffer at its GL binding point. GL 4.6 core 7.8 makes that a
+        // legal state - the block simply has no store, so reads are undefined and writes go
+        // nowhere - whereas Vulkan has no such thing as an unwritten descriptor, so something
+        // real has to sit in the set or the whole draw/dispatch is lost. One zero-filled
+        // buffer, created once and shared by every unbound binding: bindings that are only
+        // declared (the case this exists for) never touch it, and one that is actually read
+        // sees zeros, which is inside GL's "undefined". robustBufferAccess bounds anything
+        // that indexes past it.
+        BufferSlice AcquireUnboundStorageDescriptor();
+
+        // The store a texel-buffer descriptor - `samplerBuffer` or `imageBuffer` - gets when the
+        // unit the program's uniform names has no buffer texture on it, or the buffer texture on
+        // it has no GL buffer attached. Both are legal GL states that make a fetch return
+        // undefined values (GL 4.6 core 8.9: a buffer texture with no attached buffer object is
+        // incomplete, and sampling an incomplete texture is undefined - not a lost draw), and both
+        // used to take the whole draw or dispatch with them. The VIEW over this - one per format,
+        // and the descriptor is a VkBufferView, not a buffer - is built by
+        // UniformManager::AcquireUnboundTexelBufferView.
+        BufferSlice AcquireUnboundTexelBufferDescriptor();
+
         // Draw-time acquire for resident (device-storage) buffers: ensures the
         // resource exists and is fully uploaded, marks it used this frame.
         Bool AcquireResidentSlice(BufferKind kind, const SharedPtr<MG_State::GLState::BufferObject>& bufferObject,
@@ -180,6 +201,11 @@ namespace MobileGL::MG_Backend::DirectVulkan {
 
         VkBufferManagerInitInfo m_initInfo{};
         BufferArena m_transientUploadArena;
+        // See AcquireUnboundStorageDescriptor. Lazily created, never re-created, torn down
+        // with the manager.
+        VkBufferObject m_unboundStorageBuffer;
+        // See AcquireUnboundTexelBufferDescriptor. Same lifetime rules.
+        VkBufferObject m_unboundTexelBuffer;
         IBufferCopyCommandProvider* m_copyProvider = nullptr;
         Vector<Vector<VkBufferObject>> m_deferredBufferReleases;
         Vector<Vector<SharedPtr<VkBufferResource>>> m_deferredResourceReleases;

MobileGL/MG_Backend/DirectVulkan/Renderer/VkClearManager.cpp

@@ -8,6 +8,10 @@
 
 #include "VkClearManager.h"
 
+// For the shared ResolveAttachmentLayerCount (and the ToVulkanLevelExtent it is built on): the
+// clear key's layer span has to be the same one the render pass builds its attachment view from.
+#include "VkTextureManager.h"
+
 #include "MG_State/GLState/Core.h"
 #include "MG_Util/Converters/MGToStr/FramebufferEnumConverter.h"
 #include "MG_Util/Converters/MGToStr/TextureEnumConverter.h"
@@ -100,13 +104,13 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         return ResolveAttachmentBaseArrayLayer(uploadTarget);
     }
 
-    static Uint32 ResolveAttachmentLayerCount(
-        const MG_State::GLState::FramebufferAttachmentObject& attachment) {
-        if (attachment.IsLayered()) {
-            return static_cast<Uint32>(std::max(attachment.GetSize().z(), 1));
-        }
-        return 1u;
-    }
+    // ResolveAttachmentLayerCount used to be duplicated here, reading attachment.GetSize().z()
+    // raw - no ToVulkanLevelExtent remap for a 1D array, no six-faces arm for a cube map. That is
+    // not a cosmetic difference: the count below is not key-only, it is written straight into
+    // VkImageSubresourceRange::layerCount by MaterializePendingClearForTexture, which then POPS
+    // the entry - so a layered cube map's glClear reached one face and the other five were lost
+    // for good, while the very same queued clear cleared all six through the render pass's
+    // LOAD_OP_CLEAR. The helper now lives once, in VkTextureManager.h beside ToVulkanLevelExtent.
 
     static const MG_State::GLState::FramebufferAttachmentObject* GetClearableAttachment(
         const MG_State::GLState::FramebufferObject& drawFbo, FramebufferAttachmentType attachmentType) {
@@ -122,8 +126,30 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         return &attachment;
     }
 
-    PendingClearKey VkClearManager::MakePendingClearKey(MG_State::GLState::ITextureObject* texture, Uint32 mipLevel,
+    // The texture a pending clear is actually ABOUT. A clear issued through a GL texture view
+    // (ARB_texture_view) targets the storage it views, so it must queue against - and be found
+    // by - the storage texture; keying it on the view instead left the clear invisible to every
+    // materialisation done through the parent's name (and vice versa), so the image stayed in
+    // VK_IMAGE_LAYOUT_UNDEFINED and the readback was dropped as unreadable.
+    static MG_State::GLState::ITextureObject* ClearStorageTextureOf(MG_State::GLState::ITextureObject* texture) {
+        if (texture == nullptr) {
+            return nullptr;
+        }
+        const auto& storageOwner = texture->GetViewStorageOwner();
+        return storageOwner ? storageOwner.get() : texture;
+    }
+
+    PendingClearKey VkClearManager::MakePendingClearKey(MG_State::GLState::ITextureObject* rawTexture, Uint32 mipLevel,
                                                         Uint32 baseArrayLayer, Uint32 layerCount) {
+        MG_State::GLState::ITextureObject* texture = ClearStorageTextureOf(rawTexture);
+        if (rawTexture != nullptr && texture != rawTexture) {
+            // The caller named a level and a layer of the VIEW; the key describes the STORAGE, so
+            // both have to be shifted into its numbering (GL 4.6 core 8.18). Without this a clear
+            // of a view's level 0 would collide with a clear of the storage's level 0 even when
+            // the view opened onto level 1.
+            mipLevel += static_cast<Uint32>(rawTexture->GetViewMinLevel());
+            baseArrayLayer += static_cast<Uint32>(rawTexture->GetViewMinLayer());
+        }
         return PendingClearKey {
             .texture = texture,
             .textureLifetimeId = texture ? texture->GetLifetimeId() : 0,
@@ -157,6 +183,15 @@ namespace MobileGL::MG_Backend::DirectVulkan {
     }
 
     TextureIdentity VkClearManager::MakeTextureIdentity(MG_State::GLState::ITextureObject* texture) {
+        // Same rule as VkTextureManager::MakeTextureIdentity: a GL texture view is identified by
+        // the storage it views. A clear posted against a view and one posted against its parent
+        // target the same image, so they have to coalesce rather than queue independently.
+        if (texture != nullptr) {
+            const auto& storageOwner = texture->GetViewStorageOwner();
+            if (storageOwner) {
+                texture = storageOwner.get();
+            }
+        }
         return TextureIdentity {
             .texture = texture,
             .lifetimeId = texture ? texture->GetLifetimeId() : 0,
@@ -286,9 +321,11 @@ namespace MobileGL::MG_Backend::DirectVulkan {
             return;
         }
 
-        const PendingClearKey key = MakePendingClearKey(texture.get());
+        const auto& storageOwner = texture->GetViewStorageOwner();
+        const SharedPtr<MG_State::GLState::ITextureObject>& storageTexture = storageOwner ? storageOwner : texture;
+        const PendingClearKey key = MakePendingClearKey(storageTexture.get());
         const std::lock_guard<std::mutex> lock(m_mutex);
-        m_aliveObjects[MakeTextureIdentity(texture.get())] = texture;
+        m_aliveObjects[MakeTextureIdentity(storageTexture.get())] = storageTexture;
         auto& pending = m_pendingClears[key];
         MergeClearPayload(pending, clearPayload);
         m_pendingCount.store(static_cast<Uint32>(m_pendingClears.size()), std::memory_order_relaxed);
@@ -305,8 +342,14 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         }
 
         const PendingClearKey key = MakePendingClearKey(attachment);
+        // The alive entry must hold the STORAGE object, because the key names it:
+        // LockTextureIdentityLocked cross-checks the two, and registering a view here under its
+        // storage's identity made every lookup of this clear fail that check and silently report
+        // "nothing pending" - which is how a clear issued through a view's framebuffer vanished.
+        const auto& storageOwner = texture->GetViewStorageOwner();
+        const SharedPtr<MG_State::GLState::ITextureObject>& storageTexture = storageOwner ? storageOwner : texture;
         const std::lock_guard<std::mutex> lock(m_mutex);
-        m_aliveObjects[MakeTextureIdentity(texture.get())] = texture;
+        m_aliveObjects[MakeTextureIdentity(storageTexture.get())] = storageTexture;
         auto& pending = m_pendingClears[key];
         MergeClearPayload(pending, clearPayload);
         m_pendingCount.store(static_cast<Uint32>(m_pendingClears.size()), std::memory_order_relaxed);
@@ -321,6 +364,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
             return false;  // per-draw hot path: nothing pending anywhere
         }
 
+        texture = ClearStorageTextureOf(texture);
         const Uint64 lifetimeId = texture->GetLifetimeId();
         const std::lock_guard<std::mutex> lock(m_mutex);
         for (auto it = m_pendingClears.begin(); it != m_pendingClears.end(); ++it) {
@@ -411,6 +455,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
             return false;  // per-draw hot path: nothing pending anywhere
         }
 
+        texture = ClearStorageTextureOf(texture);
         const Uint64 lifetimeId = texture->GetLifetimeId();
         const std::lock_guard<std::mutex> lock(m_mutex);
         SharedPtr<MG_State::GLState::ITextureObject> liveTexture;

MobileGL/MG_Backend/DirectVulkan/Renderer/VkClearManager.h

@@ -114,7 +114,8 @@ namespace MobileGL::MG_Backend::DirectVulkan {
     class VkClearManager {
     public:
         static PendingClearKey MakePendingClearKey(const MG_State::GLState::FramebufferAttachmentObject& attachment);
-        static PendingClearKey MakePendingClearKey(MG_State::GLState::ITextureObject* texture, Uint32 mipLevel = 0,
+        // Resolves a GL texture view to the storage it views before keying; see the definition.
+        static PendingClearKey MakePendingClearKey(MG_State::GLState::ITextureObject* rawTexture, Uint32 mipLevel = 0,
                                                    Uint32 baseArrayLayer = 0, Uint32 layerCount = 1);
 
         Bool Initialize();

MobileGL/MG_Backend/DirectVulkan/Renderer/VkRenderPassManager.cpp

@@ -67,28 +67,58 @@ namespace MobileGL::MG_Backend::DirectVulkan {
     }
 
     static Uint32 ResolveAttachmentBaseArrayLayer(const MG_State::GLState::FramebufferAttachmentObject& attachment) {
+        // Every branch has to go through ToStorageArrayLayer, including the two that name layer 0
+        // implicitly: a layered attachment of a texture VIEW starts at the view's first layer, not
+        // at the image's, and a cube FACE index is a layer index like any other. Leaving either
+        // unshifted made the render pass write layers [0, n) while the clear key, the blit, the
+        // copy and the readback for the same attachment all addressed [minLayer, minLayer + n) -
+        // they resolve the layer through their own copies of this helper, which do shift.
+        const auto* texture = attachment.GetTexture().get();
         if (attachment.IsLayered()) {
-            return 0;
+            return ToStorageArrayLayer(texture, 0);
         }
         const TextureUploadTarget uploadTarget = attachment.GetTextureUploadTarget();
         if (!IsCubeMapFaceUploadTarget(uploadTarget)) {
-            return static_cast<Uint32>(std::max(attachment.GetTextureLayer(), 0));
+            return ToStorageArrayLayer(texture, attachment.GetTextureLayer());
         }
-        return static_cast<Uint32>(uploadTarget) - static_cast<Uint32>(TextureUploadTarget::CubeMapPositiveX);
+        const Int face =
+            static_cast<Int>(uploadTarget) - static_cast<Int>(TextureUploadTarget::CubeMapPositiveX);
+        return ToStorageArrayLayer(texture, face);
     }
 
-    static Uint32 ResolveAttachmentLayerCount(const MG_State::GLState::FramebufferAttachmentObject& attachment) {
-        if (attachment.IsLayered()) {
-            return static_cast<Uint32>(std::max(attachment.GetSize().z(), 1));
-        }
-        return 1u;
-    }
+    // ResolveAttachmentLayerCount lives in VkTextureManager.h, beside ToVulkanLevelExtent, because
+    // VkClearManager needs the SAME answer: its pending-clear key's layerCount becomes a real
+    // VkImageSubresourceRange when a clear is materialised outside a render pass. See the header.
 
+    // VUID-VkFramebufferCreateInfo-flags-04113: every view handed to vkCreateFramebuffer must have
+    // been created as VK_IMAGE_VIEW_TYPE_2D or VK_IMAGE_VIEW_TYPE_2D_ARRAY. The image's OWN view
+    // type is not a legal answer for several of the targets GL can attach, and returning it
+    // unchanged is what took the process down on every layered 3D / cube-map-array attachment:
+    // a 3D view is refused outright by the layer-span guard in GetOrCreateAttachmentViewAtMipLevel
+    // (3D images have arrayLayers == 1) and a CUBE_ARRAY view is built happily and then rejected -
+    // or dereferenced - by the driver inside vkCreateFramebuffer.
+    //
+    // A 2D_ARRAY view is the legal spelling of all three: over a 2D-array-compatible 3D image its
+    // "layers" are the mip's z slices (VUID-VkImageViewCreateInfo-image-04970), and over a
+    // CUBE_COMPATIBLE 2D image - which is what both cube targets are - its layers are the faces.
+    //
+    // Knowingly NOT remapped: VK_IMAGE_VIEW_TYPE_1D / _1D_ARRAY, which 04113 also forbids. There is
+    // no legal alternative for them (a VK_IMAGE_TYPE_1D image admits no 2D-family view at all), so
+    // the only honest answer would be to decline the attachment - and every driver this has run on,
+    // lavapipe included, accepts them. Declining would turn working GL_TEXTURE_1D[_ARRAY] render
+    // targets into skipped draws to satisfy a VU nothing enforces. Left as-is, deliberately.
     static VkImageViewType ResolveAttachmentViewType(
         const MG_State::GLState::FramebufferAttachmentObject& attachment,
         const VkTextureManager::TextureResource& resource) {
         if (attachment.IsLayered()) {
-            return resource.viewType;
+            switch (resource.viewType) {
+            case VK_IMAGE_VIEW_TYPE_3D:
+            case VK_IMAGE_VIEW_TYPE_CUBE:
+            case VK_IMAGE_VIEW_TYPE_CUBE_ARRAY:
+                return VK_IMAGE_VIEW_TYPE_2D_ARRAY;
+            default:
+                return resource.viewType;
+            }
         }
         // A non-layered attachment names ONE layer, so the view over it is a plain 2D view whatever
         // the image's own view type is. The cube-face upload targets always meant this; a cube map
@@ -96,8 +126,15 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         // a single layer is not a legal attachment. The CUBE arm is inert today - no frontend path
         // produces a non-layered cube attachment without a face upload target - and is kept for
         // symmetry with CUBE_ARRAY.
+        //
+        // 3D belongs in the same list and was missing from it, which is why the "per-slice
+        // attachment view is a 2D view whose array layer is the slice" branch in
+        // GetOrCreateAttachmentViewAtMipLevel was unreachable: glFramebufferTextureLayer on a
+        // GL_TEXTURE_3D asked for a 3D view (illegal as an attachment) whose span was then checked
+        // against arrayLayers == 1, so every slice above z = 0 came back VK_NULL_HANDLE.
         if (IsCubeMapFaceUploadTarget(attachment.GetTextureUploadTarget()) ||
-            resource.viewType == VK_IMAGE_VIEW_TYPE_CUBE_ARRAY || resource.viewType == VK_IMAGE_VIEW_TYPE_CUBE) {
+            resource.viewType == VK_IMAGE_VIEW_TYPE_CUBE_ARRAY || resource.viewType == VK_IMAGE_VIEW_TYPE_CUBE ||
+            resource.viewType == VK_IMAGE_VIEW_TYPE_3D) {
             return VK_IMAGE_VIEW_TYPE_2D;
         }
         return resource.viewType;
@@ -318,47 +355,18 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         }
 
         const auto internalFormat = renderbuffer->GetInternalFormat();
-        // Three-channel color formats widen to their RGBA twin exactly like textures do
-        // (VkTextureManager::ResolveTextureFormatInfo): blits/resolves between a
-        // renderbuffer and a texture of the same GL format then see one VkFormat.
-        const VkFormat format = [&]() -> VkFormat {
-            switch (internalFormat) {
-            case TextureInternalFormat::RGB:
-            case TextureInternalFormat::RGB8:
-            case TextureInternalFormat::R3G3B2:
-            case TextureInternalFormat::RGB4:
-            case TextureInternalFormat::RGB5:
-                return VK_FORMAT_R8G8B8A8_UNORM;
-            case TextureInternalFormat::SRGB8:
-                return VK_FORMAT_R8G8B8A8_SRGB;
-            case TextureInternalFormat::RGB8Snorm:
-                return VK_FORMAT_R8G8B8A8_SNORM;
-            case TextureInternalFormat::RGB10:
-            case TextureInternalFormat::RGB12:
-            case TextureInternalFormat::RGB16:
-                return VK_FORMAT_R16G16B16A16_UNORM;
-            case TextureInternalFormat::RGB16Snorm:
-                return VK_FORMAT_R16G16B16A16_SNORM;
-            case TextureInternalFormat::RGB16F:
-                return VK_FORMAT_R16G16B16A16_SFLOAT;
-            case TextureInternalFormat::RGB32F:
-                return VK_FORMAT_R32G32B32A32_SFLOAT;
-            case TextureInternalFormat::RGB8I:
-                return VK_FORMAT_R8G8B8A8_SINT;
-            case TextureInternalFormat::RGB8UI:
-                return VK_FORMAT_R8G8B8A8_UINT;
-            case TextureInternalFormat::RGB16I:
-                return VK_FORMAT_R16G16B16A16_SINT;
-            case TextureInternalFormat::RGB16UI:
-                return VK_FORMAT_R16G16B16A16_UINT;
-            case TextureInternalFormat::RGB32I:
-                return VK_FORMAT_R32G32B32A32_SINT;
-            case TextureInternalFormat::RGB32UI:
-                return VK_FORMAT_R32G32B32A32_UINT;
-            default:
-                return MG_Util::ConvertTextureInternalFormatToVkEnum(internalFormat);
-            }
-        }();
+        // ONE resolver, shared with textures (VkTextureManager::ResolveTextureFormatInfo), so a
+        // renderbuffer and a texture of the same GL format cannot disagree about their VkFormat.
+        // `expandRgbToRgba` / `componentByteCount` / `alphaBytes` describe how to reshape a SHADOW
+        // UPLOAD, and a renderbuffer has none, so only `.format` is taken.
+        //
+        // This used to be a hand-maintained second copy of that table, and it was missing exactly
+        // four rows: RGBA2 and RGBA12 fell through to ConvertTextureInternalFormatToVkEnum's
+        // VK_FORMAT_UNDEFINED (no image at all - bound as a draw buffer the attachment became
+        // VK_ATTACHMENT_UNUSED and every draw into it was dropped), while RGBA4 and RGB5A1 fell
+        // through to the 16-bit packed formats and then faced 32-bit R8G8B8A8_UNORM textures across
+        // a size-incompatible vkCmdCopyImage.
+        const VkFormat format = ResolveTextureFormatInfo(internalFormat).format;
         const VkImageAspectFlags aspect = ResolveImageAspectMaskForFormat(format);
         // Renderbuffers are never sampled (GL has no way to bind one to a sampler), so the
         // usage set is attachment + transfer: transfer covers readback (vkCmdCopyImageToBuffer),
@@ -633,11 +641,13 @@ namespace MobileGL::MG_Backend::DirectVulkan {
             if (att.IsTexture()) {
                 const Uint64 textureLifetimeId = att.GetTexture()->GetLifetimeId();
                 XXHASH_VERIFY(XXH64_update(m_hashState, &textureLifetimeId, sizeof(textureLifetimeId)));
-                const Int textureLevel = att.GetTextureLevel();
+                const Int textureLevel = static_cast<Int>(ToStorageMipLevel(att.GetTexture().get(),
+                                                                             att.GetTextureLevel()));
                 XXHASH_VERIFY(XXH64_update(m_hashState, &textureLevel, sizeof(textureLevel)));
                 const TextureUploadTarget textureUploadTarget = att.GetTextureUploadTarget();
                 XXHASH_VERIFY(XXH64_update(m_hashState, &textureUploadTarget, sizeof(textureUploadTarget)));
-                const Int textureLayer = att.GetTextureLayer();
+                const Int textureLayer = static_cast<Int>(ToStorageArrayLayer(att.GetTexture().get(),
+                                                                              att.GetTextureLayer()));
                 XXHASH_VERIFY(XXH64_update(m_hashState, &textureLayer, sizeof(textureLayer)));
                 const Bool textureLayered = att.IsLayered();
                 XXHASH_VERIFY(XXH64_update(m_hashState, &textureLayered, sizeof(textureLayered)));
@@ -754,7 +764,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         return XXH64_digest(m_hashState);
     }
 
-    RenderPassEntry& VkRenderPassManager::GetOrCreateRenderPass(const MG_State::GLState::FramebufferObject& fbo,
+    RenderPassEntry* VkRenderPassManager::GetOrCreateRenderPass(const MG_State::GLState::FramebufferObject& fbo,
                                                                 Uint32 swapchainImageIndex,
                                                                 Bool drawUsesDepthStencil) {
         // Resolve the default-FBO depth flavor (see the header comment): keep the
@@ -840,7 +850,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
             auto activeIt = m_renderPasses.find(activeRenderPass->hash);
             if (activeIt != m_renderPasses.end()) {
                 activeIt->second.lastUsedFrame = m_frameCounter;
-                return activeIt->second;
+                return &activeIt->second;
             }
         }
 
@@ -864,13 +874,13 @@ namespace MobileGL::MG_Backend::DirectVulkan {
             m_rpFastRenderPassHash = activeRenderPass->hash;
             m_rpFastHadDepthStencil = activeIt->second.hasDepthStencilAttachment;
             activeIt->second.lastUsedFrame = m_frameCounter;
-            return activeIt->second;
+            return &activeIt->second;
         }
         auto hash = ComputeHash(fbo, swapchainImageIndex, true, includeDefaultFboDepthStencil);
         auto it = m_renderPasses.find(hash);
         if (it != m_renderPasses.end()) {
             it->second.lastUsedFrame = m_frameCounter;
-            return it->second;
+            return &it->second;
         }
 
         Bool isDefaultFbo = fbo.IsDefaultFramebuffer();
@@ -993,8 +1003,12 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                     textureResources.emplace_back(nullptr);
                     attachmentViews.emplace_back(rbAttachmentFormat != rbResource->format ? rbResource->unormTwinView
                                                                                           : rbResource->view);
-                    MOBILEGL_ASSERT(attachmentViews.back() != VK_NULL_HANDLE,
-                                    "GetOrCreateRenderPass: renderbuffer view missing at color attachment %d", i);
+                    if (attachmentViews.back() == VK_NULL_HANDLE) {
+                        MGLOG_E_ONCE("GetOrCreateRenderPass: renderbuffer %u has no usable view for color attachment "
+                                     "%u on FBO %u; declining the render pass",
+                                     renderbuffer->GetExternalIndex(), i, fbo.GetExternalIndex());
+                        return nullptr;
+                    }
 
                     colorAttachmentRefs[i].attachment = rbAttachmentIndex;
                     continue;
@@ -1006,7 +1020,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                 continue;
 
             auto& att = fbo.GetAttachment(drawbuf);
-            const Uint32 attachmentMipLevel = static_cast<Uint32>(std::max(att.GetTextureLevel(), 0));
+            const Uint32 attachmentMipLevel = ToStorageMipLevel(att.GetTexture().get(), att.GetTextureLevel());
             const auto textureTarget = texture->GetTarget();
             const Uint32 attachmentIndex = static_cast<Uint32>(attachmentDescriptions.size());
             attachmentDescriptions.emplace_back();
@@ -1049,8 +1063,12 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                             .key = VkClearManager::MakePendingClearKey(att)
                         });
                     }
-                    const IntVec2 attachmentExtent =
-                        ResolveRenderPassFramebufferExtent(isDefaultFbo, att.GetSize(), swapchainExtent);
+                    // Same remap as ResolveAttachmentLayerCount, for the same reason: a
+                    // 1D-array attachment's GL height is its layer count, and using it as the
+                    // framebuffer height asks for a framebuffer taller than the VK_IMAGE_TYPE_1D
+                    // image it is built over.
+                    const IntVec2 attachmentExtent = ResolveRenderPassFramebufferExtent(
+                        isDefaultFbo, ToVulkanLevelExtent(texture->GetTarget(), att.GetSize()), swapchainExtent);
                     if (width == 0)
                         width = attachmentExtent.x();
                     if (height == 0)
@@ -1078,8 +1096,15 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                         attachmentViews.emplace_back(swapchainViews[swapchainImageIndex]);
                     } else {
                         auto* textureResource = m_textureManager.SyncTextureAndGetDescriptor(*texture);
-                        MOBILEGL_ASSERT(textureResource,
-                                        "GetOrCreateRenderPass: SyncTextureAndGetDescriptor failed at color attachment %d", i);
+                        if (textureResource == nullptr) {
+                            // SyncTextureResource legitimately declines - an unsupported format,
+                            // sample count or image-flag combination, or a vkCreateImage the driver
+                            // refused. There is no image to attach, so there is no render pass.
+                            MGLOG_E_ONCE("GetOrCreateRenderPass: textureId=%d could not be backed for color "
+                                         "attachment %u on FBO %u; declining the render pass",
+                                         texture->GetExternalIndex(), i, fbo.GetExternalIndex());
+                            return nullptr;
+                        }
                         textureResources.emplace_back(textureResource);
                         desc.format = ResolveSrgbAttachmentWriteFormat(
                             textureResource->format,
@@ -1100,8 +1125,15 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                         attachmentViews.emplace_back(
                             m_textureManager.GetOrCreateAttachmentViewAtMipLevel(
                                 *texture, attachmentMipLevel, baseArrayLayer, layerCount, attachmentViewType));
-                        MOBILEGL_ASSERT(attachmentViews.back() != VK_NULL_HANDLE,
-                                        "GetOrCreateRenderPass: GetOrCreateAttachmentView failed at color attachment %d", i);
+                        if (attachmentViews.back() == VK_NULL_HANDLE) {
+                            MGLOG_E_ONCE("GetOrCreateRenderPass: no attachment view for textureId=%d mip=%u layers "
+                                         "[%u, %u) viewType=%d at color attachment %u on FBO %u; declining the "
+                                         "render pass",
+                                         texture->GetExternalIndex(), attachmentMipLevel, baseArrayLayer,
+                                         baseArrayLayer + layerCount, static_cast<Int>(attachmentViewType), i,
+                                         fbo.GetExternalIndex());
+                            return nullptr;
+                        }
                     }
                     desc.samples = attachmentSampleCount;
                     adoptRenderPassSampleCount(attachmentSampleCount, "color", texture->GetExternalIndex());
@@ -1144,7 +1176,8 @@ namespace MobileGL::MG_Backend::DirectVulkan {
             if (a.IsTexture() && b.IsTexture()) {
                 return a.GetTexture().get() == b.GetTexture().get() &&
                        a.GetTextureUploadTarget() == b.GetTextureUploadTarget() &&
-                       a.GetTextureLevel() == b.GetTextureLevel();
+                       ToStorageMipLevel(a.GetTexture().get(), a.GetTextureLevel()) ==
+                           ToStorageMipLevel(b.GetTexture().get(), b.GetTextureLevel());
             }
             if (a.IsRenderbuffer() && b.IsRenderbuffer()) {
                 return a.GetRenderbuffer().get() == b.GetRenderbuffer().get();
@@ -1193,20 +1226,29 @@ namespace MobileGL::MG_Backend::DirectVulkan {
             } else if (selectedDepthStencilAttachment->IsTexture()) {
                 auto& texture = *selectedDepthStencilAttachment->GetTexture();
                 depthTextureResource = m_textureManager.SyncTextureAndGetDescriptor(texture);
-                MOBILEGL_ASSERT(depthTextureResource,
-                                "GetOrCreateRenderPass: SyncTextureAndGetDescriptor failed at depth attachment");
+                if (depthTextureResource == nullptr) {
+                    MGLOG_E_ONCE("GetOrCreateRenderPass: textureId=%d could not be backed for the depth/stencil "
+                                 "attachment of FBO %u; declining the render pass",
+                                 texture.GetExternalIndex(), fbo.GetExternalIndex());
+                    return nullptr;
+                }
                 trackedDepthLayout = depthTextureResource->layout;
                 depthAttachmentDescription.format = depthTextureResource->format;
                 depthAttachmentSampleCount = depthTextureResource->sampleCount;
                 depthAttachmentId = static_cast<Int>(texture.GetExternalIndex());
-                attachmentExtent =
-                    ResolveRenderPassFramebufferExtent(isDefaultFbo, selectedDepthStencilAttachment->GetSize(),
-                                                       swapchainExtent);
+                attachmentExtent = ResolveRenderPassFramebufferExtent(
+                    isDefaultFbo,
+                    ToVulkanLevelExtent(texture.GetTarget(), selectedDepthStencilAttachment->GetSize()),
+                    swapchainExtent);
             } else {
                 const auto& renderbuffer = selectedDepthStencilAttachment->GetRenderbuffer();
                 depthRenderbufferResource = GetOrCreateRenderbufferResource(renderbuffer);
-                MOBILEGL_ASSERT(depthRenderbufferResource,
-                                "GetOrCreateRenderPass: GetOrCreateRenderbufferResource failed at depth attachment");
+                if (depthRenderbufferResource == nullptr) {
+                    MGLOG_E_ONCE("GetOrCreateRenderPass: renderbuffer %u could not be backed for the depth/stencil "
+                                 "attachment of FBO %u; declining the render pass",
+                                 renderbuffer->GetExternalIndex(), fbo.GetExternalIndex());
+                    return nullptr;
+                }
                 trackedDepthLayout = depthRenderbufferResource->layout;
                 depthAttachmentDescription.format = depthRenderbufferResource->format;
                 depthAttachmentSampleCount = depthRenderbufferResource->sampleCount;
@@ -1254,7 +1296,8 @@ namespace MobileGL::MG_Backend::DirectVulkan {
             } else if (selectedDepthStencilAttachment->IsTexture()) {
                 auto& texture = *selectedDepthStencilAttachment->GetTexture();
                 const Uint32 attachmentMipLevel =
-                    static_cast<Uint32>(std::max(selectedDepthStencilAttachment->GetTextureLevel(), 0));
+                    ToStorageMipLevel(selectedDepthStencilAttachment->GetTexture().get(),
+                                      selectedDepthStencilAttachment->GetTextureLevel());
                 MOBILEGL_ASSERT(depthTextureResource->layout != VK_IMAGE_LAYOUT_UNDEFINED ||
                                     depthAttachmentDescription.loadOp != VK_ATTACHMENT_LOAD_OP_LOAD,
                                 "GetOrCreateRenderPass: depth attachment textureId=%d has undefined tracked layout with LOAD_OP_LOAD",
@@ -1279,8 +1322,14 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                 attachmentViews.emplace_back(
                     m_textureManager.GetOrCreateAttachmentViewAtMipLevel(
                         texture, attachmentMipLevel, baseArrayLayer, layerCount, attachmentViewType));
-                MOBILEGL_ASSERT(attachmentViews.back() != VK_NULL_HANDLE,
-                                "GetOrCreateRenderPass: GetOrCreateAttachmentView failed at depth attachment");
+                if (attachmentViews.back() == VK_NULL_HANDLE) {
+                    MGLOG_E_ONCE("GetOrCreateRenderPass: no attachment view for textureId=%d mip=%u layers [%u, %u) "
+                                 "viewType=%d at the depth/stencil attachment of FBO %u; declining the render pass",
+                                 texture.GetExternalIndex(), attachmentMipLevel, baseArrayLayer,
+                                 baseArrayLayer + layerCount, static_cast<Int>(attachmentViewType),
+                                 fbo.GetExternalIndex());
+                    return nullptr;
+                }
                 if (width == 0 || height == 0) {
                     width = attachmentExtent.x();
                     height = attachmentExtent.y();
@@ -1302,6 +1351,12 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                 });
                 textureResources.emplace_back(nullptr);
                 attachmentViews.emplace_back(depthRenderbufferResource->view);
+                if (attachmentViews.back() == VK_NULL_HANDLE) {
+                    MGLOG_E_ONCE("GetOrCreateRenderPass: renderbuffer %u has no usable view for the depth/stencil "
+                                 "attachment of FBO %u; declining the render pass",
+                                 renderbuffer->GetExternalIndex(), fbo.GetExternalIndex());
+                    return nullptr;
+                }
                 if (width == 0 || height == 0) {
                     width = attachmentExtent.x();
                     height = attachmentExtent.y();
@@ -1399,8 +1454,25 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         renderPassCreateInfo.dependencyCount = 2;
         renderPassCreateInfo.pDependencies = subpassDependencies;
 
+        // NOT VK_VERIFY. VkIncludes.h states the rule this function now lives by: VK_VERIFY is the
+        // INVARIANT check - a should-never-happen state, fatal-logged unlatched and trapped in a
+        // DEBUG build - and "a soft, recoverable failure must therefore NOT be routed through
+        // VK_VERIFY. Check the VkResult directly and report it with MGLOG_E_ONCE". A decline here
+        // is recoverable by construction: the caller drops the draw. Routing it through VK_VERIFY
+        // would have made the recovery dead code in a DEBUG build (the TRAP fires inside the macro,
+        // before the handle is ever examined) and, in an INFO build, printed an UNLATCHED fatal
+        // line on every draw for the life of the process - a decline caches nothing, so every
+        // later draw to the same framebuffer re-enters this path and fails again.
         VkRenderPass renderPass = VK_NULL_HANDLE;
-        VK_VERIFY(vkCreateRenderPass(m_device, &renderPassCreateInfo, nullptr, &renderPass));
+        const VkResult renderPassResult =
+            vkCreateRenderPass(m_device, &renderPassCreateInfo, nullptr, &renderPass);
+        if (renderPassResult != VK_SUCCESS || renderPass == VK_NULL_HANDLE) {
+            MGLOG_E_ONCE("GetOrCreateRenderPass: vkCreateRenderPass failed (%s, %d) for FBO %u; declining the "
+                         "render pass",
+                         VkResultToString(renderPassResult), static_cast<Int>(renderPassResult),
+                         fbo.GetExternalIndex());
+            return nullptr;
+        }
 
         // Framebuffer
         VkFramebufferCreateInfo framebufferCreateInfo;
@@ -1413,8 +1485,21 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         framebufferCreateInfo.width = width;
         framebufferCreateInfo.height = height;
         framebufferCreateInfo.layers = framebufferLayers;
+        // Direct VkResult check, for the same reason as vkCreateRenderPass above.
         VkFramebuffer framebuffer = VK_NULL_HANDLE;
-        VK_VERIFY(vkCreateFramebuffer(m_device, &framebufferCreateInfo, nullptr, &framebuffer));
+        const VkResult framebufferResult =
+            vkCreateFramebuffer(m_device, &framebufferCreateInfo, nullptr, &framebuffer);
+        if (framebufferResult != VK_SUCCESS || framebuffer == VK_NULL_HANDLE) {
+            // The render pass has no entry to own it yet, so it is destroyed here rather than
+            // leaked - RenderPassEntry's destructor is the only other thing that would.
+            MGLOG_E_ONCE("GetOrCreateRenderPass: vkCreateFramebuffer failed (%s, %d) for FBO %u (%dx%d, "
+                         "%u attachments, %u layers); declining the render pass",
+                         VkResultToString(framebufferResult), static_cast<Int>(framebufferResult),
+                         fbo.GetExternalIndex(), width, height,
+                         static_cast<Uint32>(attachmentViews.size()), framebufferLayers);
+            vkDestroyRenderPass(m_device, renderPass, nullptr);
+            return nullptr;
+        }
         IntVec2 extent = {width, height};
         RenderPassEntry renderPassEntry {
             hash,
@@ -1439,7 +1524,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                 extent.y());
         auto [insertedIt, _] = m_renderPasses.emplace(hash, Move(renderPassEntry));
         insertedIt->second.lastUsedFrame = m_frameCounter;
-        return insertedIt->second;
+        return &insertedIt->second;
     }
 
     void VkRenderPassManager::OnPresent() {

MobileGL/MG_Backend/DirectVulkan/Renderer/VkRenderPassManager.h

@@ -243,9 +243,24 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         // draw against a depth-less active pass resolves to a new (incompatible)
         // entry, which the caller's compatibility check turns into a pass split;
         // the new pass's depth loads DONT_CARE (content was undefined all along).
-        RenderPassEntry& GetOrCreateRenderPass(const MG_State::GLState::FramebufferObject& fbo,
-                                               Uint32 swapchainImageIndex,
-                                               Bool drawUsesDepthStencil = true);
+        //
+        // Returns NULLPTR when this framebuffer cannot be represented as a Vulkan render pass at
+        // all - a texture the texture manager declined to back (an unsupported format or sample
+        // count), or an attachment view it cannot construct (a layer span the image has no room
+        // for, a 3D image whose format was refused 2D-array compatibility). This used to be
+        // unrepresentable: the function returned a reference, so the only thing the two fallible
+        // calls it builds on could do was trip a MOBILEGL_ASSERT - which is compiled out of every
+        // INFO build - and then dereference the null resource, or hand VK_NULL_HANDLE to
+        // vkCreateFramebuffer. That took the whole process down (51 lost CTS records over 21
+        // bodies, one runner restart each) where a declined draw is merely a wrong picture.
+        //
+        // EVERY caller must handle nullptr by dropping the operation, exactly as the draw path
+        // already drops a draw whose sampler descriptor could not be resolved
+        // (UniformManager::BindProgramUniformBuffers). The failure paths log MGLOG_E_ONCE
+        // themselves, so a caller needs no message of its own.
+        [[nodiscard]] RenderPassEntry* GetOrCreateRenderPass(const MG_State::GLState::FramebufferObject& fbo,
+                                                             Uint32 swapchainImageIndex,
+                                                             Bool drawUsesDepthStencil = true);
         void QueueRenderbufferClear(GLbitfield mask, const ClearFramebufferPayload& clearPayload,
                                     const MG_State::GLState::FramebufferObject& drawFbo);
         void QueueRenderbufferClear(const ClearAttachmentPayload& clearPayload,

MobileGL/MG_Backend/DirectVulkan/Renderer/VkSamplerManager.cpp

@@ -21,6 +21,156 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                    sampler.GetWrapR() == SamplerWrapMode::ClampToBorder;
         }
 
+        // The numeric domain the texture is SAMPLED in. Vulkan splits VkBorderColor into a float
+        // family and an integer family and requires the sampler's choice to match the image view's
+        // format (a float border on an integer view, or the reverse, is undefined) - so the domain
+        // comes from the TEXTURE, while the value comes from whichever GL entry point wrote it.
+        enum class BorderColorDomain {
+            Float,
+            SignedInteger,
+            UnsignedInteger
+        };
+
+        BorderColorDomain ResolveBorderColorDomain(TextureInternalFormat format) {
+            switch (format) {
+            case TextureInternalFormat::R8I:
+            case TextureInternalFormat::R16I:
+            case TextureInternalFormat::R32I:
+            case TextureInternalFormat::RG8I:
+            case TextureInternalFormat::RG16I:
+            case TextureInternalFormat::RG32I:
+            case TextureInternalFormat::RGB8I:
+            case TextureInternalFormat::RGB16I:
+            case TextureInternalFormat::RGB32I:
+            case TextureInternalFormat::RGBA8I:
+            case TextureInternalFormat::RGBA16I:
+            case TextureInternalFormat::RGBA32I:
+                return BorderColorDomain::SignedInteger;
+            case TextureInternalFormat::R8UI:
+            case TextureInternalFormat::R16UI:
+            case TextureInternalFormat::R32UI:
+            case TextureInternalFormat::RG8UI:
+            case TextureInternalFormat::RG16UI:
+            case TextureInternalFormat::RG32UI:
+            case TextureInternalFormat::RGB8UI:
+            case TextureInternalFormat::RGB16UI:
+            case TextureInternalFormat::RGB32UI:
+            case TextureInternalFormat::RGBA8UI:
+            case TextureInternalFormat::RGBA16UI:
+            case TextureInternalFormat::RGBA32UI:
+            case TextureInternalFormat::RGB10A2UI:
+                return BorderColorDomain::UnsignedInteger;
+            default:
+                return BorderColorDomain::Float;
+            }
+        }
+
+        Bool IsSignedNormalizedFormat(TextureInternalFormat format) {
+            switch (format) {
+            case TextureInternalFormat::R8Snorm:
+            case TextureInternalFormat::R16Snorm:
+            case TextureInternalFormat::RG8Snorm:
+            case TextureInternalFormat::RG16Snorm:
+            case TextureInternalFormat::RGB8Snorm:
+            case TextureInternalFormat::RGB16Snorm:
+            case TextureInternalFormat::RGBA8Snorm:
+            case TextureInternalFormat::RGBA16Snorm:
+                return true;
+            default:
+                return false;
+            }
+        }
+
+        // GL 4.6 core 8.14.2: "The border values are clamped before they are used, according to the
+        // format in which texture components are stored. For signed and unsigned normalized
+        // fixed-point formats, border values are clamped to [-1,1] and [0,1] respectively. For
+        // floating-point and integer formats, border values are clamped to the representable range of
+        // the format." Every clause of that sentence is a real case here - the clamp is not just the
+        // normalized one.
+        //
+        // Only the 32-bit float formats are genuinely unclamped: every finite float is representable
+        // in them. Half-float has a finite maximum, and the two packed "float" formats are UNSIGNED,
+        // so a negative border on them must come back as 0 rather than as a negative number the
+        // driver delivers verbatim through VK_BORDER_COLOR_FLOAT_CUSTOM_EXT.
+        struct FloatBorderRange {
+            Bool clamped = true;
+            Float minValue = 0.0f;
+            Float maxValue = 1.0f;
+        };
+
+        FloatBorderRange ResolveFloatBorderRange(TextureInternalFormat format, Bool isSignedNormalized) {
+            switch (format) {
+            case TextureInternalFormat::R32F:
+            case TextureInternalFormat::RG32F:
+            case TextureInternalFormat::RGB32F:
+            case TextureInternalFormat::RGBA32F:
+                return {false, 0.0f, 0.0f};
+            case TextureInternalFormat::R16F:
+            case TextureInternalFormat::RG16F:
+            case TextureInternalFormat::RGB16F:
+            case TextureInternalFormat::RGBA16F:
+                return {true, -65504.0f, 65504.0f};
+            // Unsigned packed floats: no sign bit at all. 65024 is the largest 11-bit float; the
+            // 10-bit blue channel tops out lower (64512) and RGB9E5 higher (65408), but the bound
+            // that matters for correctness is the lower one, and a single conservative upper bound
+            // costs nothing a real border colour will ever notice.
+            case TextureInternalFormat::R11FG11FB10F:
+                return {true, 0.0f, 64512.0f};
+            case TextureInternalFormat::RGB9E5:
+                return {true, 0.0f, 65408.0f};
+            default:
+                return {true, isSignedNormalized ? -1.0f : 0.0f, 1.0f};
+            }
+        }
+
+        // Per-component representable range of an integer texture format, as Int64 so that the whole
+        // signed and unsigned 32-bit ranges are expressible in one type and the clamp can be written
+        // once for both domains. Alpha is carried separately because RGB10_A2UI is the one format
+        // whose alpha is narrower than its colour channels.
+        struct IntegerBorderRange {
+            Int64 rgbMin = 0;
+            Int64 rgbMax = 0;
+            Int64 alphaMin = 0;
+            Int64 alphaMax = 0;
+        };
+
+        IntegerBorderRange ResolveIntegerBorderRange(TextureInternalFormat format) {
+            const auto uniform = [](Int64 low, Int64 high) { return IntegerBorderRange{low, high, low, high}; };
+            switch (format) {
+            case TextureInternalFormat::R8I:
+            case TextureInternalFormat::RG8I:
+            case TextureInternalFormat::RGB8I:
+            case TextureInternalFormat::RGBA8I:
+                return uniform(-128, 127);
+            case TextureInternalFormat::R16I:
+            case TextureInternalFormat::RG16I:
+            case TextureInternalFormat::RGB16I:
+            case TextureInternalFormat::RGBA16I:
+                return uniform(-32768, 32767);
+            case TextureInternalFormat::R8UI:
+            case TextureInternalFormat::RG8UI:
+            case TextureInternalFormat::RGB8UI:
+            case TextureInternalFormat::RGBA8UI:
+                return uniform(0, 255);
+            case TextureInternalFormat::R16UI:
+            case TextureInternalFormat::RG16UI:
+            case TextureInternalFormat::RGB16UI:
+            case TextureInternalFormat::RGBA16UI:
+                return uniform(0, 65535);
+            case TextureInternalFormat::R32UI:
+            case TextureInternalFormat::RG32UI:
+            case TextureInternalFormat::RGB32UI:
+            case TextureInternalFormat::RGBA32UI:
+                return uniform(0, 4294967295LL);
+            case TextureInternalFormat::RGB10A2UI:
+                return {0, 1023, 0, 3};
+            default:
+                // The signed 32-bit formats, and anything unexpected: the full int32 range, i.e. a
+                // clamp that cannot alter a value the GL entry points could have carried.
+                return uniform(-2147483648LL, 2147483647LL);
+            }
+        }
+
         Bool IsDepthTextureFormat(TextureInternalFormat format) {
             switch (format) {
             case TextureInternalFormat::DepthComponent:
@@ -72,6 +222,9 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         m_config = initInfo.config;
         m_samplerAnisotropySupported = initInfo.samplerAnisotropySupported;
         m_maxSamplerAnisotropy = std::max(initInfo.maxSamplerAnisotropy, 1.0f);
+        m_customBorderColorSupported = initInfo.customBorderColorSupported;
+        m_maxCustomBorderColorSamplers = initInfo.maxCustomBorderColorSamplers;
+        m_customBorderColorSamplerCount = 0;
         MOBILEGL_ASSERT(m_device != VK_NULL_HANDLE && m_config != nullptr,
                         "VkSamplerManager::Initialize failed: invalid initialization info");
         return true;
@@ -102,6 +255,9 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         m_device = VK_NULL_HANDLE;
         m_config = nullptr;
         m_frameBoundaryCounter = 0;
+        m_customBorderColorSupported = false;
+        m_maxCustomBorderColorSamplers = 0;
+        m_customBorderColorSamplerCount = 0;
     }
 
     void VkSamplerManager::OnFrameBoundary() {
@@ -123,6 +279,9 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                 if (m_device != VK_NULL_HANDLE && entry.handle != VK_NULL_HANDLE) {
                     vkDestroySampler(m_device, entry.handle, nullptr);
                 }
+                if (entry.usesCustomBorderColor && m_customBorderColorSamplerCount > 0) {
+                    --m_customBorderColorSamplerCount;
+                }
                 it = m_samplers.erase(it);
             } else {
                 ++it;
@@ -131,8 +290,8 @@ namespace MobileGL::MG_Backend::DirectVulkan {
     }
 
     Uint64 VkSamplerManager::BuildSamplerKey(const MG_State::GLState::SamplerObject& sampler,
-                                             const MG_State::GLState::ITextureObject& texture,
-                                             Bool forceNearestFiltering, Bool singleLevelView) const {
+                                             Bool forceNearestFiltering, Bool singleLevelView,
+                                             const ResolvedBorderColor& borderColor) const {
         MOBILEGL_ASSERT(m_config != nullptr, "VkSamplerManager::BuildSamplerKey: m_config is null");
         XXHASH_VERIFY(XXH64_reset(m_hashState, m_config->CacheVersion));
 
@@ -166,8 +325,13 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         XXHASH_VERIFY(XXH64_update(m_hashState, &compareMode, sizeof(compareMode)));
         const auto compareFunc = sampler.GetSamplerCompareFunc();
         XXHASH_VERIFY(XXH64_update(m_hashState, &compareFunc, sizeof(compareFunc)));
-        const auto borderColor = ResolveVkBorderColor(sampler, texture);
-        XXHASH_VERIFY(XXH64_update(m_hashState, &borderColor, sizeof(borderColor)));
+        // The resolved enum AND, when it is one of the *_CUSTOM_EXT values, the sixteen bytes of the
+        // colour itself: two samplers that differ only in a custom border colour carry the same enum
+        // and would otherwise collide onto whichever one was created first.
+        XXHASH_VERIFY(XXH64_update(m_hashState, &borderColor.color, sizeof(borderColor.color)));
+        if (borderColor.isCustom) {
+            XXHASH_VERIFY(XXH64_update(m_hashState, &borderColor.customValue, sizeof(borderColor.customValue)));
+        }
         return XXH64_digest(m_hashState);
     }
 
@@ -183,7 +347,9 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         // allocation for a genuinely single-level image) and faults the GPU - the same failure
         // the default-framebuffer blit shader had to work around with an explicit-LOD sample.
         const Bool singleLevelView = viewLevelCount == 1;
-        const Uint64 key = BuildSamplerKey(sampler, texture, forceNearestFiltering, singleLevelView);
+        // Resolved once and used for both the key and the create-info; see ResolvedBorderColor.
+        const ResolvedBorderColor borderColor = ResolveBorderColor(sampler, texture);
+        const Uint64 key = BuildSamplerKey(sampler, forceNearestFiltering, singleLevelView, borderColor);
         auto it = m_samplers.find(key);
         if (it != m_samplers.end()) {
             it->second.lastUsedFrameBoundary = m_frameBoundaryCounter;
@@ -211,9 +377,21 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         // Must match BuildSamplerKey's resolution exactly.
         samplerInfo.maxLod = ResolveSingleLevelMaxLod(sampler, singleLevelView);
         samplerInfo.minLod = ResolveEffectiveMinLod(sampler, samplerInfo.maxLod);
-        samplerInfo.borderColor = ResolveVkBorderColor(sampler, texture);
+        samplerInfo.borderColor = borderColor.color;
         samplerInfo.unnormalizedCoordinates = VK_FALSE;
 
+        // VK_EXT_custom_border_color. `format` stays UNDEFINED, which is legal only because
+        // customBorderColorWithoutFormat was required alongside customBorderColors at device
+        // creation - a GL sampler object has no idea which texture it will be paired with.
+        VkSamplerCustomBorderColorCreateInfoEXT customBorderColorInfo{};
+        if (borderColor.isCustom) {
+            customBorderColorInfo.sType = VK_STRUCTURE_TYPE_SAMPLER_CUSTOM_BORDER_COLOR_CREATE_INFO_EXT;
+            customBorderColorInfo.customBorderColor = borderColor.customValue;
+            customBorderColorInfo.format = VK_FORMAT_UNDEFINED;
+            customBorderColorInfo.pNext = samplerInfo.pNext;
+            samplerInfo.pNext = &customBorderColorInfo;
+        }
+
         VkSampler vkSampler = VK_NULL_HANDLE;
         VK_VERIFY(vkCreateSampler(m_device, &samplerInfo, nullptr, &vkSampler), "vkCreateSampler(texture)");
 
@@ -222,6 +400,10 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         entry.externalIndex = sampler.GetExternalIndex();
         entry.version = sampler.GetVersion();
         entry.lastUsedFrameBoundary = m_frameBoundaryCounter;
+        entry.usesCustomBorderColor = borderColor.isCustom;
+        if (entry.usesCustomBorderColor) {
+            ++m_customBorderColorSamplerCount;
+        }
         m_samplers[key] = entry;
         return vkSampler;
     }
@@ -281,39 +463,148 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         }
     }
 
-    VkBorderColor VkSamplerManager::ResolveVkBorderColor(const MG_State::GLState::SamplerObject& sampler,
-                                                         const MG_State::GLState::ITextureObject& texture) {
+    VkSamplerManager::ResolvedBorderColor VkSamplerManager::ResolveBorderColor(
+        const MG_State::GLState::SamplerObject& sampler, const MG_State::GLState::ITextureObject& texture) const {
+        ResolvedBorderColor resolved{};
         if (!UsesBorderColor(sampler)) {
-            return VK_BORDER_COLOR_FLOAT_TRANSPARENT_BLACK;
+            return resolved; // FLOAT_TRANSPARENT_BLACK, never sampled
         }
 
         // Border colour is sampler state: a bound sampler object supplies its own, and a texture
         // with none reaches the very same value through the sampler object it owns.
-        const auto& borderColor = sampler.GetBorderColor();
-        const Bool isDepthTexture = IsDepthTextureFormat(texture.GetFormat());
+        const auto format = texture.GetFormat();
+        const auto domain = ResolveBorderColorDomain(format);
+        const Bool canUseCustom = m_customBorderColorSupported && m_maxCustomBorderColorSamplers > 0 &&
+                                  m_customBorderColorSamplerCount < m_maxCustomBorderColorSamplers;
 
-        if (isDepthTexture) {
+        if (domain != BorderColorDomain::Float) {
+            // An integer image view REQUIRES an integer border colour, whatever the value is - even
+            // (0,0,0,1). The value itself is whichever integer form the application wrote; a float
+            // border on an integer texture is nonsense GL leaves undefined, so the derived integer
+            // representation (a plain cast) is as good an answer as any.
+            //
+            // Clamped to the format's representable range FIRST, per GL 4.6 core 8.14.2, and read
+            // through Int64 so the whole signed and unsigned 32-bit ranges are expressible at once.
+            //
+            // Which representation to start from is the TEXTURE's domain, not the entry-point form
+            // the application used. GL 4.6 core 8.10 stores an "I"-form border colour unmodified with
+            // an integer internal data type and does not define a sign conversion between the two
+            // integer forms, so the stored bits are reinterpreted in the sampled format's own
+            // signedness. Measured, not assumed: a border of -1 written with glTexParameterIiv
+            // against a GL_R8UI texture samples as 255 on the ES driver, i.e. as 0xFFFFFFFF clamped
+            // to the format's maximum - see the IntegerBorderColorScenario case that pins it. Picking
+            // the representation by the FORM instead would answer 0 here, which is a defensible
+            // reading of the same spec text but puts DirectVulkan at odds with DirectGLES - and
+            // DirectGLES cannot deviate, it forwards the value to the driver verbatim. Cross-backend
+            // agreement decides it.
+            const auto range = ResolveIntegerBorderRange(format);
+            const auto& borderColorI = sampler.GetBorderColorI();
+            const auto& borderColorUI = sampler.GetBorderColorUI();
+            const Bool startFromUnsigned = domain == BorderColorDomain::UnsignedInteger;
+            Int64 clamped[4];
+            for (SizeT channel = 0; channel < 4; ++channel) {
+                const Int64 raw = startFromUnsigned ? static_cast<Int64>(borderColorUI[channel])
+                                                    : static_cast<Int64>(borderColorI[channel]);
+                const Int64 low = channel == 3 ? range.alphaMin : range.rgbMin;
+                const Int64 high = channel == 3 ? range.alphaMax : range.rgbMax;
+                clamped[channel] = std::clamp(raw, low, high);
+            }
+
+            // Matched against the CLAMPED value, so a border the format cannot hold still lands on
+            // the palette entry it clamps to rather than missing every one of them.
+            const Bool allZeroRgb = clamped[0] == 0 && clamped[1] == 0 && clamped[2] == 0;
+            if (allZeroRgb && clamped[3] == 0) {
+                resolved.color = VK_BORDER_COLOR_INT_TRANSPARENT_BLACK;
+                return resolved;
+            }
+            if (allZeroRgb && clamped[3] == 1) {
+                resolved.color = VK_BORDER_COLOR_INT_OPAQUE_BLACK;
+                return resolved;
+            }
+            if (clamped[0] == 1 && clamped[1] == 1 && clamped[2] == 1 && clamped[3] == 1) {
+                resolved.color = VK_BORDER_COLOR_INT_OPAQUE_WHITE;
+                return resolved;
+            }
+            if (canUseCustom) {
+                resolved.color = VK_BORDER_COLOR_INT_CUSTOM_EXT;
+                resolved.isCustom = true;
+                for (SizeT channel = 0; channel < 4; ++channel) {
+                    if (domain == BorderColorDomain::UnsignedInteger) {
+                        resolved.customValue.uint32[channel] = static_cast<Uint32>(clamped[channel]);
+                    } else {
+                        resolved.customValue.int32[channel] = static_cast<Int32>(clamped[channel]);
+                    }
+                }
+                return resolved;
+            }
+            // No custom colour available: pick the nearest of the three integer palette entries
+            // rather than always answering transparent black, which is what turned an integer border
+            // of (-1,-1,-1,-1) into 0 and broke the CTS's clamped-texel detection outright.
+            const Bool opaque = clamped[3] != 0;
+            const Bool bright = clamped[0] != 0 || clamped[1] != 0 || clamped[2] != 0;
+            resolved.color = !opaque ? VK_BORDER_COLOR_INT_TRANSPARENT_BLACK
+                                     : (bright ? VK_BORDER_COLOR_INT_OPAQUE_WHITE : VK_BORDER_COLOR_INT_OPAQUE_BLACK);
+            return resolved;
+        }
+
+        // Float domain. GL 4.6 core 8.14.2/8.23: the border colour is interpreted in the texture's
+        // format, so it is clamped to that format's representable range first. Without the clamp the
+        // CTS's border of (255,255,255,255) on a GL_RGBA8 texture matched none of the palette entries
+        // and fell through to transparent black - every border texel sampled 0 where the test wanted
+        // 255. The range is per format class, not just the normalized [0,1] / [-1,1] pair: only the
+        // 32-bit float formats are unclamped.
+        FloatVec4 borderColor = sampler.GetBorderColor();
+        if (const auto range = ResolveFloatBorderRange(format, IsSignedNormalizedFormat(format)); range.clamped) {
+            borderColor = FloatVec4(std::clamp(borderColor.x(), range.minValue, range.maxValue),
+                                    std::clamp(borderColor.y(), range.minValue, range.maxValue),
+                                    std::clamp(borderColor.z(), range.minValue, range.maxValue),
+                                    std::clamp(borderColor.w(), range.minValue, range.maxValue));
+        }
+
+        // A depth texture samples one component, so only x decides - and its alpha reads as 1.
+        if (IsDepthTextureFormat(format)) {
             if (NearlyEqual(borderColor.x(), 1.0f)) {
-                return VK_BORDER_COLOR_FLOAT_OPAQUE_WHITE;
+                resolved.color = VK_BORDER_COLOR_FLOAT_OPAQUE_WHITE;
+                return resolved;
             }
             if (NearlyEqual(borderColor.x(), 0.0f)) {
-                return VK_BORDER_COLOR_FLOAT_OPAQUE_BLACK;
+                resolved.color = VK_BORDER_COLOR_FLOAT_OPAQUE_BLACK;
+                return resolved;
             }
         }
 
         const Bool rgbZero = NearlyEqual(borderColor.x(), 0.0f) && NearlyEqual(borderColor.y(), 0.0f) &&
                              NearlyEqual(borderColor.z(), 0.0f);
         if (rgbZero && NearlyEqual(borderColor.w(), 0.0f)) {
-            return VK_BORDER_COLOR_FLOAT_TRANSPARENT_BLACK;
+            resolved.color = VK_BORDER_COLOR_FLOAT_TRANSPARENT_BLACK;
+            return resolved;
         }
         if (rgbZero && NearlyEqual(borderColor.w(), 1.0f)) {
-            return VK_BORDER_COLOR_FLOAT_OPAQUE_BLACK;
+            resolved.color = VK_BORDER_COLOR_FLOAT_OPAQUE_BLACK;
+            return resolved;
         }
         if (NearlyEqual(borderColor.x(), 1.0f) && NearlyEqual(borderColor.y(), 1.0f) &&
             NearlyEqual(borderColor.z(), 1.0f) && NearlyEqual(borderColor.w(), 1.0f)) {
-            return VK_BORDER_COLOR_FLOAT_OPAQUE_WHITE;
+            resolved.color = VK_BORDER_COLOR_FLOAT_OPAQUE_WHITE;
+            return resolved;
         }
 
-        return VK_BORDER_COLOR_FLOAT_TRANSPARENT_BLACK;
+        if (canUseCustom) {
+            resolved.color = VK_BORDER_COLOR_FLOAT_CUSTOM_EXT;
+            resolved.isCustom = true;
+            resolved.customValue.float32[0] = borderColor.x();
+            resolved.customValue.float32[1] = borderColor.y();
+            resolved.customValue.float32[2] = borderColor.z();
+            resolved.customValue.float32[3] = borderColor.w();
+            return resolved;
+        }
+
+        // Nearest of the three float palette entries. Transparent black stays the answer for a
+        // transparent border, which is what the old unconditional fallback got right by accident.
+        const Bool opaque = borderColor.w() >= 0.5f;
+        const Bool bright = (borderColor.x() + borderColor.y() + borderColor.z()) >= 1.5f;
+        resolved.color = !opaque ? VK_BORDER_COLOR_FLOAT_TRANSPARENT_BLACK
+                                 : (bright ? VK_BORDER_COLOR_FLOAT_OPAQUE_WHITE : VK_BORDER_COLOR_FLOAT_OPAQUE_BLACK);
+        return resolved;
     }
 } // namespace MobileGL::MG_Backend::DirectVulkan

MobileGL/MG_Backend/DirectVulkan/Renderer/VkSamplerManager.h

@@ -28,6 +28,13 @@ public:
         Bool samplerAnisotropySupported = false;
         // VkPhysicalDeviceLimits::maxSamplerAnisotropy.
         Float maxSamplerAnisotropy = 1.0f;
+        // VK_EXT_custom_border_color was enabled with BOTH customBorderColors and
+        // customBorderColorWithoutFormat; see VulkanRenderer::m_customBorderColorFeatureEnabled.
+        Bool customBorderColorSupported = false;
+        // VkPhysicalDeviceCustomBorderColorPropertiesEXT::maxCustomBorderColorSamplers. A hard device
+        // limit on how many LIVE samplers may carry a custom border colour, so the cache counts them
+        // and falls back to the snapped predefined value once it is reached.
+        Uint32 maxCustomBorderColorSamplers = 0;
     };
 
     Bool Initialize(const InitInfo& initInfo);
@@ -52,6 +59,21 @@ public:
     // boundaries.
     void OnFrameBoundary();
 
+    // What GL_TEXTURE_BORDER_COLOR resolves to for one (sampler, texture) pair. `color` is always a
+    // legal VkBorderColor; when `isCustom` it is one of the *_CUSTOM_EXT values and `customValue`
+    // carries the actual components in a VkSamplerCustomBorderColorCreateInfoEXT.
+    //
+    // Resolved ONCE per GetOrCreateSampler call and threaded into both the cache key and the
+    // create-info, so the two cannot disagree - the same discipline the resolved anisotropy needs,
+    // and here it also makes the maxCustomBorderColorSamplers fallback deterministic: whether a
+    // custom colour was affordable is decided before the key is built, not twice with a budget
+    // change in between.
+    struct ResolvedBorderColor {
+        VkBorderColor color = VK_BORDER_COLOR_FLOAT_TRANSPARENT_BLACK;
+        VkClearColorValue customValue{};
+        Bool isCustom = false;
+    };
+
 private:
     struct SamplerCacheEntry {
         VkSampler handle = VK_NULL_HANDLE;
@@ -60,17 +82,18 @@ private:
         // Frame boundary of the last cache hit; entries idle past the
         // OnFrameBoundary retirement age have their VkSampler destroyed.
         Uint64 lastUsedFrameBoundary = 0;
+        // Counted against maxCustomBorderColorSamplers for as long as this entry lives.
+        Bool usesCustomBorderColor = false;
     };
 
-    Uint64 BuildSamplerKey(const MG_State::GLState::SamplerObject& sampler,
-                           const MG_State::GLState::ITextureObject& texture,
-                           Bool forceNearestFiltering, Bool singleLevelView) const;
+    Uint64 BuildSamplerKey(const MG_State::GLState::SamplerObject& sampler, Bool forceNearestFiltering,
+                           Bool singleLevelView, const ResolvedBorderColor& borderColor) const;
     static VkFilter ToVkFilter(SamplerFilterMode mode);
     static VkSamplerMipmapMode ToVkMipmapMode(SamplerMipmapMode mode);
     static VkSamplerAddressMode ToVkAddressMode(SamplerWrapMode mode);
     static VkCompareOp ToVkCompareOp(SamplerCompareFunc func);
-    static VkBorderColor ResolveVkBorderColor(const MG_State::GLState::SamplerObject& sampler,
-                                              const MG_State::GLState::ITextureObject& texture);
+    ResolvedBorderColor ResolveBorderColor(const MG_State::GLState::SamplerObject& sampler,
+                                           const MG_State::GLState::ITextureObject& texture) const;
     // The anisotropy Vulkan will actually apply: 1.0 (i.e. disabled) unless the feature is on and
     // the sampler filters linearly both ways, otherwise the GL request clamped to the device limit.
     // GL happily carries GL_TEXTURE_MAX_ANISOTROPY on a NEAREST sampler (Blaze3D's blocks do exactly
@@ -82,6 +105,12 @@ private:
     const VulkanRendererConfig* m_config = nullptr;
     Bool m_samplerAnisotropySupported = false;
     Float m_maxSamplerAnisotropy = 1.0f;
+    Bool m_customBorderColorSupported = false;
+    Uint32 m_maxCustomBorderColorSamplers = 0;
+    // Live cache entries carrying a custom border colour. Kept in step with the entries themselves
+    // in exactly the three places one can appear or disappear: creation, the OnFrameBoundary sweep,
+    // and Shutdown.
+    Uint32 m_customBorderColorSamplerCount = 0;
     UnorderedMap<Uint64, SamplerCacheEntry> m_samplers;
     // Monotonic frame-boundary counter (bumped in OnFrameBoundary) for cache aging.
     Uint64 m_frameBoundaryCounter = 0;

MobileGL/MG_Backend/DirectVulkan/Renderer/VkTextureManager.cpp

@@ -46,13 +46,6 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         return mipLevelCount;
     }
 
-    struct TextureFormatInfo {
-        VkFormat format = VK_FORMAT_UNDEFINED;
-        Bool expandRgbToRgba = false;
-        Uint32 componentByteCount = 0;
-        Array<Uint8, 4> alphaBytes = {0, 0, 0, 0};
-    };
-
     struct TextureShapeInfo {
         VkImageType imageType = VK_IMAGE_TYPE_2D;
         VkImageViewType viewType = VK_IMAGE_VIEW_TYPE_2D;
@@ -220,6 +213,22 @@ namespace MobileGL::MG_Backend::DirectVulkan {
 
     VkTextureManager::TextureIdentity VkTextureManager::MakeTextureIdentity(
         MG_State::GLState::ITextureObject* texture) {
+        // A GL texture view (ARB_texture_view) is identified by the texture whose STORAGE it
+        // views, not by itself. Everything this identity keys - the TextureResource, the tracked
+        // image layout, the alive-object weak reference, the storage-usage marks, the per-draw
+        // sync memos - is a property of the IMAGE, and a view shares that image exactly. Doing
+        // the resolution here rather than at each call site is what makes it impossible to miss
+        // one: a layout update posted against a view's own identity would have found no resource
+        // at all, which is precisely how an attached view came back blank.
+        //
+        // One hop suffices and cannot recurse: glTextureView composes a view-of-a-view onto the
+        // root at creation, so a storage owner is never itself a view.
+        if (texture != nullptr) {
+            const auto& storageOwner = texture->GetViewStorageOwner();
+            if (storageOwner) {
+                texture = storageOwner.get();
+            }
+        }
         return TextureIdentity{
             .texture = texture,
             .lifetimeId = texture ? texture->GetLifetimeId() : 0,
@@ -364,7 +373,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         return true;
     }
 
-    static TextureFormatInfo ResolveTextureFormatInfo(TextureInternalFormat format) {
+    TextureFormatInfo ResolveTextureFormatInfo(TextureInternalFormat format) {
         switch (format) {
         case TextureInternalFormat::RGB:
         case TextureInternalFormat::RGB8:
@@ -694,6 +703,8 @@ namespace MobileGL::MG_Backend::DirectVulkan {
     }
 
     void VkTextureManager::EraseTrackedTexture(const TextureIdentity& identity) {
+        m_viewRequestedImageFlags.erase(identity);
+        m_viewRequestedFormats.erase(identity);
         auto resourceIt = m_textureResources.find(identity);
         if (resourceIt != m_textureResources.end()) {
             DeferResourceRelease(Move(resourceIt->second));
@@ -737,9 +748,19 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         m_drawSyncedThisDraw.clear();
     }
 
-    VkTextureManager::TextureResource* VkTextureManager::SyncTextureAndGetDescriptor(MG_State::GLState::ITextureObject& texture) {
+    VkTextureManager::TextureResource* VkTextureManager::SyncTextureAndGetDescriptor(MG_State::GLState::ITextureObject& textureOrView) {
         MOBILEGL_ASSERT(m_device != VK_NULL_HANDLE, "SyncTextureAndGetDescriptor: m_device == VK_NULL_HANDLE");
 
+        // A GL texture view has no image of its own; it resolves to - and shares - the resource
+        // of the texture whose storage it views, so that there is exactly one VkImage, one
+        // tracked layout and one upload path per storage. Everything that makes the view a
+        // different texture (format, level/layer window, sampled aspect) is applied where the
+        // VkImageViews are built, keyed in alternateSampledViews / attachmentViews.
+        MG_State::GLState::ITextureObject& texture = StorageTextureOf(textureOrView);
+        if (&texture != &textureOrView) {
+            NoteTextureViewImageRequirements(textureOrView, texture);
+        }
+
         const TextureIdentity identity = MakeTextureIdentity(&texture);
 
         // Per-draw memo fast path (see BeginDrawSyncScope): a texture already fully
@@ -838,7 +859,19 @@ namespace MobileGL::MG_Backend::DirectVulkan {
 
     VkImageView VkTextureManager::GetOrCreateViewAtMipLevel(MG_State::GLState::ITextureObject& texture, Uint32 mipLevel) {
         TextureResource* resource = SyncTextureAndGetDescriptor(texture);
-        if (resource == nullptr || resource->image == VK_NULL_HANDLE || mipLevel >= resource->mipLevels) {
+        if (resource == nullptr || resource->image == VK_NULL_HANDLE) {
+            return VK_NULL_HANDLE;
+        }
+        // A GL texture view shares this resource with the texture it views, so it must not touch
+        // perMipViews: that vector is indexed by mip level alone and holds views built with the
+        // STORAGE texture's format and full layer range. Route it through the keyed attachment
+        // cache instead, where its own window is part of the key.
+        if (texture.IsTextureView()) {
+            const TextureViewWindow window = ResolveTextureViewWindow(texture, *resource);
+            return GetOrCreateAttachmentViewAtMipLevel(texture, mipLevel, window.baseArrayLayer, window.layerCount,
+                                                       window.viewType);
+        }
+        if (mipLevel >= resource->mipLevels) {
             return VK_NULL_HANDLE;
         }
 
@@ -866,21 +899,43 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                                                                       Uint32 layerCount,
                                                                       VkImageViewType viewType) {
         TextureResource* resource = SyncTextureAndGetDescriptor(texture);
-        if (resource == nullptr || resource->image == VK_NULL_HANDLE || mipLevel >= resource->mipLevels) {
+        if (resource == nullptr || resource->image == VK_NULL_HANDLE) {
             return VK_NULL_HANDLE;
         }
-        // A 3D image has arrayLayers == 1 and keeps its GL layers on the z axis, so a per-slice
-        // attachment view is a 2D view whose "array layer" is the slice - legal only on a
-        // 2D-array-compatible image (VUID-VkImageViewCreateInfo-image-04970), which
-        // SyncTextureResource asks for and may have had refused per format.
-        if (resource->viewType == VK_IMAGE_VIEW_TYPE_3D && viewType == VK_IMAGE_VIEW_TYPE_2D) {
+        // mipLevel and baseArrayLayer arrive in STORAGE space - every caller runs them through
+        // ToStorageMipLevel / ToStorageArrayLayer at the GL attachment boundary. What a GL texture
+        // view still contributes here is its own internal format, which may reinterpret the
+        // storage's (GL 4.6 core table 8.21) and is what the attachment must actually be written
+        // through.
+        VkFormat viewFormatOverride = VK_FORMAT_UNDEFINED;
+        if (texture.IsTextureView()) {
+            viewFormatOverride = ResolveTextureViewWindow(texture, *resource).format;
+        }
+        if (mipLevel >= resource->mipLevels) {
+            return VK_NULL_HANDLE;
+        }
+        // A 3D image has arrayLayers == 1 and keeps its GL layers on the z axis, so an attachment
+        // view over it addresses SLICES through baseArrayLayer/layerCount: one slice for a
+        // non-layered attachment (a 2D view) and the whole span for a layered one (a 2D_ARRAY view,
+        // which is what a layered GL_TEXTURE_3D attachment plus a gl_Layer-writing geometry shader
+        // means). BOTH spellings are legal only on a 2D-array-compatible image
+        // (VUID-VkImageViewCreateInfo-image-04970 / -06723), which SyncTextureResource asks for and
+        // may have had refused per format.
+        //
+        // The span is validated against the MIP's slice count, never against arrayLayers: a 3D
+        // image's arrayLayers is 1 by construction, so measuring a layered span against it rejected
+        // every layered 3D attachment - the null view that used to reach vkCreateFramebuffer.
+        if (resource->viewType == VK_IMAGE_VIEW_TYPE_3D &&
+            (viewType == VK_IMAGE_VIEW_TYPE_2D || viewType == VK_IMAGE_VIEW_TYPE_2D_ARRAY)) {
             const Uint32 sliceCount = std::max(resource->depth >> mipLevel, 1u);
             if ((resource->imageCreateFlags & VK_IMAGE_CREATE_2D_ARRAY_COMPATIBLE_BIT) == 0 ||
                 layerCount == 0 || baseArrayLayer >= sliceCount || baseArrayLayer + layerCount > sliceCount) {
-                MGLOG_D("%s: cannot name slice span [%u, %u) of 3D textureId=%d (mip %u has %u slices, "
-                        "2D-array-compatible=%d)",
+                // Not an error line: the render-pass builder turns the null view into one
+                // MGLOG_E_ONCE and a skipped draw, which is the level this belongs at.
+                MGLOG_D("%s: cannot name slice span [%u, %u) of 3D textureId=%d as viewType=%d (mip %u has %u "
+                        "slices, 2D-array-compatible=%d)",
                         __func__, baseArrayLayer, baseArrayLayer + layerCount, texture.GetExternalIndex(),
-                        mipLevel, sliceCount,
+                        static_cast<Int>(viewType), mipLevel, sliceCount,
                         (int)((resource->imageCreateFlags & VK_IMAGE_CREATE_2D_ARRAY_COMPATIBLE_BIT) != 0));
                 return VK_NULL_HANDLE;
             }
@@ -894,10 +949,15 @@ namespace MobileGL::MG_Backend::DirectVulkan {
 
         const Bool framebufferSrgbEnabled =
             MG_State::pGLContext->IsCapabilityEnabled(MobileGL::CapabilityInput::FramebufferSrgb);
-        const VkFormat attachmentFormat = ResolveSrgbAttachmentWriteFormat(resource->format, framebufferSrgbEnabled);
+        const VkFormat baseAttachmentFormat =
+            viewFormatOverride != VK_FORMAT_UNDEFINED ? viewFormatOverride : resource->format;
+        const VkFormat attachmentFormat =
+            ResolveSrgbAttachmentWriteFormat(baseAttachmentFormat, framebufferSrgbEnabled);
 
-        if (attachmentFormat == resource->format && baseArrayLayer == 0 && layerCount == resource->arrayLayers &&
-            viewType == resource->viewType) {
+        // The shortcut back to the per-mip vector is only sound for the storage texture itself;
+        // for a view every field below is part of what distinguishes it from its parent.
+        if (viewFormatOverride == VK_FORMAT_UNDEFINED && attachmentFormat == resource->format &&
+            baseArrayLayer == 0 && layerCount == resource->arrayLayers && viewType == resource->viewType) {
             return GetOrCreateViewAtMipLevel(texture, mipLevel);
         }
 
@@ -932,7 +992,22 @@ namespace MobileGL::MG_Backend::DirectVulkan {
     VkImageView VkTextureManager::GetOrCreateSampledViewAtMipLevel(MG_State::GLState::ITextureObject& texture,
                                                                    Uint32 mipLevel) {
         TextureResource* resource = SyncTextureAndGetDescriptor(texture);
-        if (resource == nullptr || resource->image == VK_NULL_HANDLE || mipLevel >= resource->mipLevels) {
+        if (resource == nullptr || resource->image == VK_NULL_HANDLE) {
+            return VK_NULL_HANDLE;
+        }
+        // As in GetOrCreateViewAtMipLevel: perMipSampledViews belongs to the storage texture's
+        // own format and aspect, so a GL view has to go to the keyed cache.
+        if (texture.IsTextureView()) {
+            if (mipLevel >= resource->mipLevels) {
+                return VK_NULL_HANDLE;
+            }
+            TextureViewWindow window = ResolveTextureViewWindow(texture, *resource);
+            // Storage space already (see ToStorageMipLevel); only the level COUNT narrows.
+            window.baseMipLevel = mipLevel;
+            window.levelCount = 1;
+            return GetOrCreateWindowedSampledView(texture, *resource, window);
+        }
+        if (mipLevel >= resource->mipLevels) {
             return VK_NULL_HANDLE;
         }
 
@@ -960,14 +1035,76 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         return perMipSampledView;
     }
 
-    VkImageView VkTextureManager::GetOrCreateSampledImageView(MG_State::GLState::ITextureObject& texture,
-                                                               VkFormat format) {
-        TextureResource* resource = SyncTextureAndGetDescriptor(texture);
-        if (resource == nullptr || resource->image == VK_NULL_HANDLE ||
-            resource->sampledView == VK_NULL_HANDLE) {
+    // Builds (and caches) one sampled VkImageView over `resource`'s image for an arbitrary
+    // window - the shared back end of every GL-texture-view sampled path. Keyed by the whole
+    // window, which is what keeps a D24S8's depth-aspect view and its stencil-aspect view apart
+    // in the same cache while both name the same image, the same levels and the same layers.
+    VkImageView VkTextureManager::GetOrCreateWindowedSampledView(MG_State::GLState::ITextureObject& texture,
+                                                                 TextureResource& resource,
+                                                                 const TextureViewWindow& window) {
+        const TextureResource::SampledImageViewKey key{
+            .baseMipLevel = window.baseMipLevel,
+            .levelCount = window.levelCount,
+            .baseArrayLayer = window.baseArrayLayer,
+            .layerCount = window.layerCount,
+            .viewType = window.viewType,
+            .format = window.format,
+            .aspect = window.sampledAspect,
+            .componentSwizzle = PackComponentSwizzle(window.components),
+        };
+        const auto existing = resource.alternateSampledViews.find(key);
+        if (existing != resource.alternateSampledViews.end()) {
+            return existing->second;
+        }
+
+        if (window.format != resource.format &&
+            (resource.imageCreateFlags & VK_IMAGE_CREATE_MUTABLE_FORMAT_BIT) == 0) {
+            MGLOG_E_ONCE("%s: textureId=%d needs a mutable-format image to be viewed as format=%d "
+                         "(image format=%d)",
+                         __func__, texture.GetExternalIndex(), static_cast<Int>(window.format),
+                         static_cast<Int>(resource.format));
             return VK_NULL_HANDLE;
         }
 
+        const VkImageView view =
+            CreateImageView(resource.image, window.format, window.sampledAspect, window.viewType,
+                            window.baseMipLevel, window.levelCount, window.baseArrayLayer, window.layerCount,
+                            &window.components);
+        if (view == VK_NULL_HANDLE) {
+            MGLOG_E_ONCE("%s: failed to create sampled view for textureId=%d format=%d aspect=0x%x "
+                         "mips=[%u,%u) layers=[%u,%u)",
+                         __func__, texture.GetExternalIndex(), static_cast<Int>(window.format),
+                         static_cast<Uint32>(window.sampledAspect), window.baseMipLevel,
+                         window.baseMipLevel + window.levelCount, window.baseArrayLayer,
+                         window.baseArrayLayer + window.layerCount);
+            return VK_NULL_HANDLE;
+        }
+        resource.alternateSampledViews.emplace(key, view);
+        return view;
+    }
+
+    VkImageView VkTextureManager::GetOrCreateSampledImageView(MG_State::GLState::ITextureObject& texture,
+                                                               VkFormat format) {
+        TextureResource* resource = SyncTextureAndGetDescriptor(texture);
+        if (resource == nullptr || resource->image == VK_NULL_HANDLE) {
+            return VK_NULL_HANDLE;
+        }
+
+        // A GL texture view never has a sampledView of its own on this resource - that one
+        // belongs to the storage texture, with the storage texture's format, level range and
+        // depth/stencil aspect. The window is the view's whole identity, so it always goes to the
+        // keyed cache, even when the requested format happens to match the image's.
+        if (texture.IsTextureView()) {
+            TextureViewWindow window = ResolveTextureViewWindow(texture, *resource);
+            if (format != VK_FORMAT_UNDEFINED) {
+                window.format = format;
+            }
+            return GetOrCreateWindowedSampledView(texture, *resource, window);
+        }
+
+        if (resource->sampledView == VK_NULL_HANDLE) {
+            return VK_NULL_HANDLE;
+        }
         if (format == VK_FORMAT_UNDEFINED || format == resource->format) {
             return resource->sampledView;
         }
@@ -987,8 +1124,13 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         const TextureResource::SampledImageViewKey key{
             .baseMipLevel = resource->sampledBaseMipLevel,
             .levelCount = resource->sampledLevelCount,
+            .baseArrayLayer = 0,
+            .layerCount = resource->arrayLayers,
             .viewType = resource->viewType,
             .format = format,
+            .aspect = VK_IMAGE_ASPECT_COLOR_BIT,
+            .componentSwizzle = PackComponentSwizzle(
+                ResolveSampledViewComponents(texture, ResolveTextureFormatInfo(texture.GetFormat()))),
         };
         const auto existing = resource->alternateSampledViews.find(key);
         if (existing != resource->alternateSampledViews.end()) {
@@ -1029,6 +1171,8 @@ namespace MobileGL::MG_Backend::DirectVulkan {
     VkImageView VkTextureManager::GetOrCreateStorageImageView(MG_State::GLState::ITextureObject& texture,
                                                                Uint32 mipLevel, VkFormat format,
                                                                Bool layered, Int32 layer) {
+        // mipLevel and layer arrive in STORAGE space; ResolveStorageImageDescriptor converts
+        // the glBindImageTexture values with ToStorageMipLevel / ToStorageArrayLayer.
         TextureResource* resource = SyncTextureAndGetDescriptor(texture);
         if (resource == nullptr || resource->image == VK_NULL_HANDLE || mipLevel >= resource->mipLevels ||
             resource->sampleCount != VK_SAMPLE_COUNT_1_BIT ||
@@ -1053,8 +1197,13 @@ namespace MobileGL::MG_Backend::DirectVulkan {
             return VK_NULL_HANDLE;
         }
 
-        Uint32 baseArrayLayer = 0;
-        Uint32 layerCount = resource->arrayLayers;
+        // A GL texture view opens onto a WINDOW of the storage's layers; a layered image
+        // binding of it must not reach past that window into the parent's other layers.
+        Uint32 baseArrayLayer = ToStorageArrayLayer(&texture, 0);
+        Uint32 layerCount = texture.IsTextureView()
+                                ? std::min(static_cast<Uint32>(texture.GetViewNumLayers()),
+                                           resource->arrayLayers - baseArrayLayer)
+                                : resource->arrayLayers;
         VkImageViewType viewType = resource->viewType;
         if (!layered) {
             switch (resource->viewType) {
@@ -1087,7 +1236,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
 
         const Bool isFullResourceView = baseArrayLayer == 0 && layerCount == resource->arrayLayers &&
                                         viewType == resource->viewType;
-        if (format == resource->format && isFullResourceView) {
+        if (format == resource->format && isFullResourceView && !texture.IsTextureView()) {
             return GetOrCreateViewAtMipLevel(texture, mipLevel);
         }
 
@@ -1613,7 +1762,20 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         const Bool storageUpgradePending =
             !outResource.storageUsageResolved &&
             m_storageImageTextures.find(MakeTextureIdentity(&texture)) != m_storageImageTextures.end();
-        if (outResource.image != VK_NULL_HANDLE && !storageUpgradePending &&
+        // Same shape for a GL texture view's demands on the image (MUTABLE_FORMAT for a
+        // format-reinterpreting view, CUBE_COMPATIBLE for a cube view of an array texture):
+        // nothing about the texture itself changed, but the live image cannot carry the view.
+        // Masked by what this format can actually be given: MUTABLE_FORMAT is deliberately
+        // withheld from formats the driver already refused it for (see SyncTextureResource), and
+        // without this mask the "upgrade still pending" test below could never come true again -
+        // costing every later sync of that texture the whole slow path, forever.
+        VkImageCreateFlags requestedViewFlags = GetViewRequestedImageFlags(texture);
+        if (m_mutableFormatUnsupported.find(outResource.format) != m_mutableFormatUnsupported.end()) {
+            requestedViewFlags &= ~VK_IMAGE_CREATE_MUTABLE_FORMAT_BIT;
+        }
+        const Bool viewFlagUpgradePending =
+            (outResource.imageCreateFlags & requestedViewFlags) != requestedViewFlags;
+        if (outResource.image != VK_NULL_HANDLE && !storageUpgradePending && !viewFlagUpgradePending &&
             outResource.syncedContentVersion == syncingContentVersion &&
             outResource.syncedShapeVersion == syncingShapeVersion &&
             outResource.syncedTextureParamsVersion == texture.GetTextureParamsVersion() &&
@@ -1807,6 +1969,16 @@ namespace MobileGL::MG_Backend::DirectVulkan {
             m_mutableFormatUnsupported.find(format) == m_mutableFormatUnsupported.end()) {
             imageCreateFlags |= VK_IMAGE_CREATE_MUTABLE_FORMAT_BIT;
         }
+        // Flags a GL texture view over this storage asked for (see NoteTextureViewImageRequirements).
+        // MUTABLE_FORMAT is still withheld from formats the driver has already refused it for, so a
+        // reinterpreting view degrades to no view rather than to no texture.
+        const VkImageCreateFlags requestedViewFlags = GetViewRequestedImageFlags(texture);
+        if (requestedViewFlags != 0) {
+            imageCreateFlags |= requestedViewFlags;
+            if (m_mutableFormatUnsupported.find(format) != m_mutableFormatUnsupported.end()) {
+                imageCreateFlags &= ~VK_IMAGE_CREATE_MUTABLE_FORMAT_BIT;
+            }
+        }
         // sRGB color images attach through their UNORM twin while GL_FRAMEBUFFER_SRGB is
         // disabled (see ResolveSrgbAttachmentWriteFormat), which needs format-reinterpreting
         // views - multisample sRGB render targets included.
@@ -1967,6 +2139,11 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                     viewFormats.push_back(viewFormat);
                 }
             }
+            // ...plus every format a glTextureView over this storage reinterprets it as. Those
+            // are NOT enumerable from ResolveSampledImageViewFormat - an application may name any
+            // member of the format's view class (GL 4.6 core table 8.21) - so without this the
+            // list would forbid the very view the MUTABLE_FORMAT bit was requested for.
+            AppendViewRequestedFormats(texture, viewFormats);
             formatListInfo.sType = VK_STRUCTURE_TYPE_IMAGE_FORMAT_LIST_CREATE_INFO;
             formatListInfo.viewFormatCount = static_cast<Uint32>(viewFormats.size());
             formatListInfo.pViewFormats = viewFormats.data();
@@ -1999,12 +2176,16 @@ namespace MobileGL::MG_Backend::DirectVulkan {
             }
             if (imageFormatResult != VK_SUCCESS && !isMultisampleTexture &&
                 (imageInfo.flags & VK_IMAGE_CREATE_2D_ARRAY_COMPATIBLE_BIT) != 0) {
-                // Losing 2D-array compatibility only costs per-slice framebuffer attachment for this
-                // format; failing creation would lose the texture entirely. Remembered so later syncs
-                // neither reprobe nor flag-mismatch against this image and recreate it.
+                // Losing 2D-array compatibility only costs framebuffer attachment of this format's
+                // 3D images - per-slice AND layered, since both are spelled as a 2D-family view over
+                // the z axis; failing creation would lose the texture entirely. Recorded here (the
+                // per-format set below) so later syncs neither reprobe nor flag-mismatch against this
+                // image and recreate it, and so GetOrCreateAttachmentViewAtMipLevel declines rather
+                // than handing back a view that cannot exist - the render-pass builder then turns
+                // that decline into a skipped draw instead of a null VkImageView in pAttachments.
                 MGLOG_W_ONCE("%s: VK_IMAGE_CREATE_2D_ARRAY_COMPATIBLE_BIT is unsupported for format=%d "
-                        "textureId=%d; creating without it (per-slice framebuffer attachment will be "
-                        "unavailable for it)",
+                        "textureId=%d; creating without it (per-slice and layered framebuffer "
+                        "attachment of 3D textures in this format will be unavailable)",
                         __func__, static_cast<Int>(format), texture.GetExternalIndex());
                 m_2dArrayCompatibleUnsupported.insert(format);
                 imageInfo.flags &= ~VK_IMAGE_CREATE_2D_ARRAY_COMPATIBLE_BIT;
@@ -2392,6 +2573,164 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         m_deferredViewReleases[m_currentFrameIndex].push_back(view);
     }
 
+    MG_State::GLState::ITextureObject& VkTextureManager::StorageTextureOf(
+        MG_State::GLState::ITextureObject& texture) {
+        const auto& storageOwner = texture.GetViewStorageOwner();
+        return storageOwner ? *storageOwner : texture;
+    }
+
+    // The VkImageViewType a GL texture view's own target asks for. Deliberately derived from the
+    // GL target rather than inherited from the storage image: a 2D view of a 2D-array texture is
+    // a VK_IMAGE_VIEW_TYPE_2D over one layer, and a cube view of the same image is a
+    // VK_IMAGE_VIEW_TYPE_CUBE over six - which is the whole reason table 8.20 lists those pairs.
+    static VkImageViewType ResolveTextureViewImageViewType(TextureTarget target,
+                                                           VkImageViewType storageViewType) {
+        switch (target) {
+        case TextureTarget::Texture1D:
+            return VK_IMAGE_VIEW_TYPE_1D;
+        case TextureTarget::Texture1DArray:
+            return VK_IMAGE_VIEW_TYPE_1D_ARRAY;
+        case TextureTarget::Texture2D:
+        case TextureTarget::TextureRectangle:
+        case TextureTarget::Texture2DMultisample:
+            return VK_IMAGE_VIEW_TYPE_2D;
+        case TextureTarget::Texture2DArray:
+        case TextureTarget::Texture2DMultisampleArray:
+            return VK_IMAGE_VIEW_TYPE_2D_ARRAY;
+        case TextureTarget::TextureCubeMap:
+            return VK_IMAGE_VIEW_TYPE_CUBE;
+        case TextureTarget::TextureCubeMapArray:
+            return VK_IMAGE_VIEW_TYPE_CUBE_ARRAY;
+        default:
+            return storageViewType;
+        }
+    }
+
+    VkTextureManager::TextureViewWindow VkTextureManager::ResolveTextureViewWindow(
+        MG_State::GLState::ITextureObject& texture, const TextureResource& resource) const {
+        TextureViewWindow window{};
+        window.format = resource.format;
+        window.viewType = resource.viewType;
+        window.baseArrayLayer = 0;
+        window.layerCount = resource.arrayLayers;
+        window.sampledAspect =
+            ResolveSampledImageViewAspectMask(resource.aspect, texture.GetDepthStencilTextureMode());
+        window.components = ResolveSampledViewComponents(texture, ResolveTextureFormatInfo(texture.GetFormat()));
+        ResolveViewMipRange(texture, resource.mipLevels, window.baseMipLevel, window.levelCount);
+        if (!texture.IsTextureView()) {
+            return window;
+        }
+
+        window.isTextureView = true;
+        // GL 4.6 core 8.18: the view's TEXTURE_BASE_LEVEL / TEXTURE_MAX_LEVEL are relative to the
+        // view, so ResolveViewMipRange above already clamped them against the view's own level
+        // count (TextureObjectView reports it); shifting by TEXTURE_VIEW_MIN_LEVEL puts them back
+        // into the storage image's numbering.
+        window.baseMipLevel += static_cast<Uint32>(texture.GetViewMinLevel());
+        window.baseArrayLayer = static_cast<Uint32>(texture.GetViewMinLayer());
+        window.layerCount = static_cast<Uint32>(texture.GetViewNumLayers());
+        window.viewType = ResolveTextureViewImageViewType(texture.GetTarget(), resource.viewType);
+        // The view's OWN internalformat, which may reinterpret the storage's (table 8.21).
+        const VkFormat viewFormat = ResolveTextureFormatInfo(texture.GetFormat()).format;
+        if (viewFormat != VK_FORMAT_UNDEFINED) {
+            window.format = viewFormat;
+        }
+        // Recomputed against the view's own format: a depth/stencil storage viewed as
+        // depth/stencil still has to honour the VIEW's DEPTH_STENCIL_TEXTURE_MODE, which is the
+        // one parameter Better Clouds deliberately sets differently on the two names.
+        window.sampledAspect =
+            ResolveSampledImageViewAspectMask(GetAspectMaskForFormat(window.format) != VK_IMAGE_ASPECT_NONE
+                                                  ? GetAspectMaskForFormat(window.format)
+                                                  : resource.aspect,
+                                              texture.GetDepthStencilTextureMode());
+
+        // Clamp to what the image actually has; a malformed view must degrade to an empty range
+        // rather than reach vkCreateImageView with an out-of-bounds subresource.
+        if (window.baseMipLevel >= resource.mipLevels) {
+            window.baseMipLevel = resource.mipLevels - 1;
+            window.levelCount = 1;
+        } else {
+            window.levelCount = std::min(window.levelCount, resource.mipLevels - window.baseMipLevel);
+        }
+        if (window.levelCount == 0) window.levelCount = 1;
+        if (window.baseArrayLayer >= resource.arrayLayers) {
+            window.baseArrayLayer = resource.arrayLayers - 1;
+            window.layerCount = 1;
+        } else {
+            window.layerCount = std::min(window.layerCount, resource.arrayLayers - window.baseArrayLayer);
+        }
+        if (window.layerCount == 0) window.layerCount = 1;
+        return window;
+    }
+
+    // The extra VkImageCreateFlags a GL texture view needs on the image it views. Recorded
+    // BEFORE the storage texture is synced (see SyncTextureAndGetDescriptor) so the very first
+    // resolve of a view already creates - or recreates and copies forward - an image the view can
+    // legally be built over, instead of handing back VK_NULL_HANDLE for a frame.
+    void VkTextureManager::NoteTextureViewImageRequirements(MG_State::GLState::ITextureObject& viewTexture,
+                                                            MG_State::GLState::ITextureObject& storageTexture) {
+        const TextureIdentity storageIdentity = MakeTextureIdentity(&storageTexture);
+        VkImageCreateFlags required = 0;
+        const VkFormat viewFormat = ResolveTextureFormatInfo(viewTexture.GetFormat()).format;
+        const VkFormat storageFormat = ResolveTextureFormatInfo(storageTexture.GetFormat()).format;
+        if (viewFormat != VK_FORMAT_UNDEFINED && storageFormat != VK_FORMAT_UNDEFINED &&
+            viewFormat != storageFormat) {
+            required |= VK_IMAGE_CREATE_MUTABLE_FORMAT_BIT;
+            // The image may be created with a NARROWED format list (see SyncTextureResource), and
+            // that list is a promise about every format the image will ever be viewed as. Record
+            // this one so the promise stays true.
+            m_viewRequestedFormats[storageIdentity].insert(viewFormat);
+        }
+        const TextureTarget viewTarget = viewTexture.GetTarget();
+        if (viewTarget == TextureTarget::TextureCubeMap || viewTarget == TextureTarget::TextureCubeMapArray) {
+            // Only when the storage could legally carry the bit. VK_IMAGE_CREATE_CUBE_COMPATIBLE
+            // demands a 2D image with square levels and at least six array layers
+            // (VUID-VkImageCreateInfo-flags-00954), and asking for it on a storage that has fewer
+            // would fail vkCreateImage - which, because SyncTextureResource has already released
+            // the old resource by then, would leave the PARENT texture with no image at all. A
+            // degenerate view must not be able to destroy the texture it views; let its own view
+            // creation fail instead.
+            const IntVec3 storageSize = storageTexture.GetBaseSize();
+            const Bool storageCanBeCube = storageSize.x() == storageSize.y() &&
+                                          storageTexture.GetViewNumLayers() >= 6 &&
+                                          storageTexture.GetTarget() != TextureTarget::Texture3D;
+            if (storageCanBeCube) {
+                required |= VK_IMAGE_CREATE_CUBE_COMPATIBLE_BIT;
+            } else {
+                MGLOG_W_ONCE("Texture view %d wants a cube view of texture %d, whose storage is %dx%d with %u "
+                             "layers and cannot be cube-compatible; the view will have no image view.",
+                             viewTexture.GetExternalIndex(), storageTexture.GetExternalIndex(), storageSize.x(),
+                             storageSize.y(), storageTexture.GetViewNumLayers());
+            }
+        }
+        if (required == 0) {
+            return;
+        }
+        VkImageCreateFlags& stored = m_viewRequestedImageFlags[storageIdentity];
+        stored |= required;
+    }
+
+    VkImageCreateFlags VkTextureManager::GetViewRequestedImageFlags(
+        const MG_State::GLState::ITextureObject& storageTexture) const {
+        const auto it = m_viewRequestedImageFlags.find(
+            MakeTextureIdentity(const_cast<MG_State::GLState::ITextureObject*>(&storageTexture)));
+        return it == m_viewRequestedImageFlags.end() ? 0 : it->second;
+    }
+
+    void VkTextureManager::AppendViewRequestedFormats(const MG_State::GLState::ITextureObject& storageTexture,
+                                                      Vector<VkFormat>& outFormats) const {
+        const auto it = m_viewRequestedFormats.find(
+            MakeTextureIdentity(const_cast<MG_State::GLState::ITextureObject*>(&storageTexture)));
+        if (it == m_viewRequestedFormats.end()) {
+            return;
+        }
+        for (const VkFormat viewFormat : it->second) {
+            if (std::find(outFormats.begin(), outFormats.end(), viewFormat) == outFormats.end()) {
+                outFormats.push_back(viewFormat);
+            }
+        }
+    }
+
     Bool VkTextureManager::SyncTextureViews(const MG_State::GLState::ITextureObject& texture, TextureResource& resource) {
         MOBILEGL_ASSERT(resource.image != VK_NULL_HANDLE, "SyncTextureViews: image == VK_NULL_HANDLE");
 

MobileGL/MG_Backend/DirectVulkan/Renderer/VkTextureManager.h

@@ -10,8 +10,10 @@
 
 #include "../VkIncludes.h"
 #include <Includes.h>
+#include <MG_State/GLState/FramebufferState/FramebufferObject.h>
 #include <MG_State/GLState/TextureState/TextureObject.h>
 #include <vk_mem_alloc.h>
+#include <algorithm>
 #include <unordered_map>
 #include <unordered_set>
 
@@ -22,6 +24,31 @@ class ITextureObject;
 namespace MobileGL::MG_Backend::DirectVulkan {
 enum class SamplerNumericDomain : Uint8;
 
+// What VkFormat a GL internal format is BACKED with, and how a shadow upload has to be reshaped to
+// fit it. This is not the same question as "is there an exact VkFormat for this GL format", which is
+// what ConvertTextureInternalFormatToVkEnum answers: several GL formats have no Vulkan twin at all
+// (RGBA2, RGBA12) and several three-channel ones are deliberately widened to their four-channel twin
+// because Vulkan devices rarely support the 3-channel layouts.
+//
+// SHARED, and it must stay the only answer to that question. A renderbuffer and a texture of the
+// same GL format have to resolve to the SAME VkFormat or every blit, resolve and glCopyImageSubData
+// between them crosses a size-incompatible pair, which vkCmdCopyImage leaves undefined
+// (VUID-vkCmdCopyImage-srcImage-01548). The renderbuffer path used to carry a hand-maintained second
+// copy of this table that was missing four rows - RGBA2, RGBA4, RGB5A1 and RGBA12 - so those four
+// renderbuffer formats either got no image at all or a 16-bit-packed one facing a 32-bit texture.
+struct TextureFormatInfo {
+    VkFormat format = VK_FORMAT_UNDEFINED;
+    // The GL format has three channels and is carried in a four-channel image; a shadow upload has
+    // to be expanded, inserting `alphaBytes` after every `componentByteCount * 3` source bytes.
+    Bool expandRgbToRgba = false;
+    Uint32 componentByteCount = 0;
+    Array<Uint8, 4> alphaBytes = {0, 0, 0, 0};
+};
+
+// Callers that only need the backing VkFormat (a renderbuffer has no shadow upload to reshape) take
+// `.format` and ignore the rest.
+TextureFormatInfo ResolveTextureFormatInfo(TextureInternalFormat format);
+
 // A GL 1D-ARRAY level keeps its LAYER COUNT in the state-side HEIGHT: that is what
 // glTexImage2D(GL_TEXTURE_1D_ARRAY, width, layers) means, and the frontend records the level
 // as {width, layers, 1} (see GL_Texture.cpp's AllocateStorage and the completeness walk in
@@ -41,6 +68,58 @@ inline IntVec3 ToVulkanLevelExtent(TextureTarget stateTarget, const IntVec3& glT
     return glTexelSize;
 }
 
+// How many Vulkan array layers (or, for a 3D image, z slices) a GL framebuffer attachment spans.
+//
+// THE ONE COPY, deliberately. This used to exist twice - privately in VkRenderPassManager.cpp and
+// again in VkClearManager.cpp - and the two are not independent: the render pass builds the
+// attachment view and VkFramebufferCreateInfo::layers from one, while the CLEAR key built from the
+// other is written verbatim into VkImageSubresourceRange::layerCount when a queued glClear is
+// materialised outside a render pass (MaterializePendingClearForTexture). They are two consumers
+// of the same GL clear, so any disagreement means the same glClear produces two different pictures
+// depending only on which path happens to consume it first - and the materialise path then POPS
+// the entry, so the other one never runs. Fixing one copy and leaving the other is exactly how
+// that split gets introduced; keep them the same function.
+//
+// Two shapes make this more than `size.z()`:
+//   * GL_TEXTURE_1D_ARRAY keeps its layer count in the state-side HEIGHT (see ToVulkanLevelExtent
+//     just above), so z reads 1 and every layer above the first was silently dropped.
+//   * GL_TEXTURE_CUBE_MAP is attached layered as its REPRESENTATIVE upload target, the +X face
+//     (ResolveRepresentableFramebufferTextureUploadTarget), and one face's level size has z = 1 -
+//     but a layered cube attachment names all six faces (GL 4.6 core 9.2.8), which are the image's
+//     six array layers. A cube ARRAY needs no such arm: its representative target carries 6n in z.
+inline Uint32 ResolveAttachmentLayerCount(const MG_State::GLState::FramebufferAttachmentObject& attachment) {
+    if (!attachment.IsLayered()) {
+        return 1u;
+    }
+    const auto& texture = attachment.GetTexture();
+    const TextureTarget target = texture != nullptr ? texture->GetTarget() : TextureTarget::Unknown;
+    if (target == TextureTarget::TextureCubeMap) {
+        return 6u;
+    }
+    return static_cast<Uint32>(std::max(ToVulkanLevelExtent(target, attachment.GetSize()).z(), 1));
+}
+
+// A GL framebuffer attachment's level/layer, and a GL image unit's, are relative to the texture
+// the application NAMED. When that texture was created by glTextureView (ARB_texture_view) they
+// are relative to the VIEW, and have to be shifted into the storage image's numbering before they
+// can index a Vulkan subresource - DirectVulkan gives a view no image of its own, it shares the
+// storage texture's (VkTextureManager::StorageTextureOf).
+//
+// Apply EXACTLY ONCE, at the boundary where a GL level/layer becomes a subresource index. Every
+// GetOrCreate*View entry point below expects values that have already been through here, and so
+// does everything that reads or copies an attachment directly. Both are identity on a plain
+// texture (TEXTURE_VIEW_MIN_LEVEL / MIN_LAYER are 0 there), so the conversion is unconditional
+// and there is no second, view-only code path to keep in step.
+inline Uint32 ToStorageMipLevel(const MG_State::GLState::ITextureObject* texture, Int glLevel) {
+    const Uint32 level = static_cast<Uint32>(glLevel > 0 ? glLevel : 0);
+    return texture != nullptr ? level + static_cast<Uint32>(texture->GetViewMinLevel()) : level;
+}
+
+inline Uint32 ToStorageArrayLayer(const MG_State::GLState::ITextureObject* texture, Int glLayer) {
+    const Uint32 layer = static_cast<Uint32>(glLayer > 0 ? glLayer : 0);
+    return texture != nullptr ? layer + static_cast<Uint32>(texture->GetViewMinLayer()) : layer;
+}
+
 class VkTextureManager {
 public:
     // Monotonic epoch bumped whenever a texture VkImage is (re)created. The render-pass
@@ -139,17 +218,35 @@ public:
             }
         };
 
+        // Layer range and aspect join the key because a GL texture view (ARB_texture_view) can
+        // differ from its storage on either: the Better Clouds shape samples ONE D24S8 image
+        // through two GL names in one draw, the parent with the stencil aspect and the view with
+        // the depth aspect, and a layer-sliced view of an array texture names a sub-range of the
+        // same image. Without these two fields those views would alias each other in the cache.
         struct SampledImageViewKey {
             Uint32 baseMipLevel = 0;
             Uint32 levelCount = 1;
+            Uint32 baseArrayLayer = 0;
+            Uint32 layerCount = 1;
             VkImageViewType viewType = VK_IMAGE_VIEW_TYPE_2D;
             VkFormat format = VK_FORMAT_UNDEFINED;
+            VkImageAspectFlags aspect = VK_IMAGE_ASPECT_COLOR_BIT;
+            // GL_TEXTURE_SWIZZLE_* is per-texture state, so two views over one storage with the
+            // same window but different swizzles are different views. Baked into the key because
+            // a GL texture view's ONLY sampled view lives in this cache: unlike the storage
+            // texture's own sampledView, which SyncTextureViews rebuilds whenever the params
+            // version moves, nothing else would ever notice a swizzle change on a view.
+            Uint32 componentSwizzle = 0;
 
             Bool operator==(const SampledImageViewKey& other) const {
                 return baseMipLevel == other.baseMipLevel &&
                        levelCount == other.levelCount &&
+                       baseArrayLayer == other.baseArrayLayer &&
+                       layerCount == other.layerCount &&
                        viewType == other.viewType &&
-                       format == other.format;
+                       format == other.format &&
+                       aspect == other.aspect &&
+                       componentSwizzle == other.componentSwizzle;
             }
         };
 
@@ -157,10 +254,15 @@ public:
             SizeT operator()(const SampledImageViewKey& key) const {
                 SizeT hash = std::hash<Uint32>{}(key.baseMipLevel);
                 hash ^= std::hash<Uint32>{}(key.levelCount) + 0x9e3779b9u + (hash << 6) + (hash >> 2);
+                hash ^= std::hash<Uint32>{}(key.baseArrayLayer) + 0x9e3779b9u + (hash << 6) + (hash >> 2);
+                hash ^= std::hash<Uint32>{}(key.layerCount) + 0x9e3779b9u + (hash << 6) + (hash >> 2);
                 hash ^= std::hash<Uint32>{}(static_cast<Uint32>(key.viewType)) +
                         0x9e3779b9u + (hash << 6) + (hash >> 2);
                 hash ^= std::hash<Uint32>{}(static_cast<Uint32>(key.format)) +
                         0x9e3779b9u + (hash << 6) + (hash >> 2);
+                hash ^= std::hash<Uint32>{}(static_cast<Uint32>(key.aspect)) +
+                        0x9e3779b9u + (hash << 6) + (hash >> 2);
+                hash ^= std::hash<Uint32>{}(key.componentSwizzle) + 0x9e3779b9u + (hash << 6) + (hash >> 2);
                 return hash;
             }
         };
@@ -358,6 +460,58 @@ public:
     // present-less frame-boundary drain.
     void CollectAllDeferredReleases();
 
+    // ---- GL texture views (ARB_texture_view / GL 4.6 core 8.18) ----
+    // The GL texture whose STORAGE backs the given one: itself, or - for a texture created by
+    // glTextureView - the texture it views. Every image-scoped question (which VkImage, its
+    // LAYOUT, its uploads, its extent, its usage) must be asked of this object, because a view
+    // has none of its own; only the VkImageViews differ per GL texture object. Sharing one
+    // TextureResource is not an optimisation, it is the only correct arrangement: layout is a
+    // property of the image, and VulkanRenderer caches raw pointers straight to the resource's
+    // layout field, so a second resource aliasing the same image would desynchronise the moment
+    // either of them transitioned it.
+    static MG_State::GLState::ITextureObject& StorageTextureOf(MG_State::GLState::ITextureObject& texture);
+
+    // The window a GL texture object opens onto its storage image. For a plain texture this is
+    // the resource's own full extent; for a view it is the sub-range, format and aspect
+    // glTextureView gave it. Views built from a non-default window must live in the KEYED caches
+    // (attachmentViews / alternateSampledViews), never in the per-mip vectors, which belong to
+    // the storage texture's own defaults.
+    struct TextureViewWindow {
+        Uint32 baseMipLevel = 0;
+        Uint32 levelCount = 1;
+        Uint32 baseArrayLayer = 0;
+        Uint32 layerCount = 1;
+        VkFormat format = VK_FORMAT_UNDEFINED;
+        VkImageViewType viewType = VK_IMAGE_VIEW_TYPE_2D;
+        VkImageAspectFlags sampledAspect = VK_IMAGE_ASPECT_COLOR_BIT;
+        VkComponentMapping components{VK_COMPONENT_SWIZZLE_R, VK_COMPONENT_SWIZZLE_G, VK_COMPONENT_SWIZZLE_B,
+                                      VK_COMPONENT_SWIZZLE_A};
+        Bool isTextureView = false;
+    };
+
+    // The four component swizzles packed into one value, for the sampled-view cache key.
+    static Uint32 PackComponentSwizzle(const VkComponentMapping& components) {
+        return (static_cast<Uint32>(components.r) & 0xFFu) | ((static_cast<Uint32>(components.g) & 0xFFu) << 8) |
+               ((static_cast<Uint32>(components.b) & 0xFFu) << 16) |
+               ((static_cast<Uint32>(components.a) & 0xFFu) << 24);
+    }
+    TextureViewWindow ResolveTextureViewWindow(MG_State::GLState::ITextureObject& texture,
+                                               const TextureResource& resource) const;
+    // Records what a GL texture view needs of the image it views, so the next sync of the
+    // STORAGE texture creates (or recreates and copies forward) an image the view can be built
+    // over. See m_viewRequestedImageFlags for why this is lazy rather than unconditional.
+    void NoteTextureViewImageRequirements(MG_State::GLState::ITextureObject& viewTexture,
+                                          MG_State::GLState::ITextureObject& storageTexture);
+    VkImageCreateFlags GetViewRequestedImageFlags(const MG_State::GLState::ITextureObject& storageTexture) const;
+    // Appends every format a GL texture view reinterprets this storage as, for the narrowed
+    // VkImageFormatListCreateInfo the image is created with.
+    void AppendViewRequestedFormats(const MG_State::GLState::ITextureObject& storageTexture,
+                                    Vector<VkFormat>& outFormats) const;
+    // Builds (and caches, keyed by the whole window) one sampled VkImageView over a storage
+    // image. Shared back end of every GL-texture-view sampled path.
+    VkImageView GetOrCreateWindowedSampledView(MG_State::GLState::ITextureObject& texture,
+                                               TextureResource& resource, const TextureViewWindow& window);
+
     TextureResource* SyncTextureAndGetDescriptor(
         MG_State::GLState::ITextureObject& texture);
     VkImageView GetOrCreateViewAtMipLevel(MG_State::GLState::ITextureObject& texture, Uint32 mipLevel);
@@ -572,6 +726,19 @@ private:
     std::unordered_map<TextureIdentity, TextureResource, TextureIdentityHash> m_textureResources;
     // Textures that have been bound to a GL image unit (see MarkStorageImageTexture).
     std::unordered_set<TextureIdentity, TextureIdentityHash> m_storageImageTextures;
+    // Extra VkImageCreateFlags a GL texture view needs on the storage image it views, keyed by
+    // the STORAGE texture's identity. Requested lazily, exactly like STORAGE usage above and for
+    // the same reason: VK_IMAGE_CREATE_MUTABLE_FORMAT_BIT costs bandwidth compression on tilers
+    // (it is what VK_KHR_image_format_list exists to claw back), so setting it on every
+    // immutable-storage texture would tax every glTexStorage2D render target in a game for a
+    // feature almost none of them use. A SAME-format view - which is the common case, and the
+    // Better Clouds case - needs no flag at all and therefore costs nothing.
+    std::unordered_map<TextureIdentity, VkImageCreateFlags, TextureIdentityHash> m_viewRequestedImageFlags;
+    // Every VkFormat a GL texture view has asked to reinterpret this storage as. The narrowed
+    // VkImageFormatListCreateInfo the image is created with must name them: the list is a promise
+    // that NO other format will ever be viewed, and building a view outside it is
+    // VUID-VkImageViewCreateInfo-pNext-01585. Keyed, like the flags above, by the STORAGE texture.
+    std::unordered_map<TextureIdentity, std::unordered_set<VkFormat>, TextureIdentityHash> m_viewRequestedFormats;
     // Supported multisample counts per format, so repeat texture syncs do not
     // re-query vkGetPhysicalDeviceImageFormatProperties.
     std::unordered_map<VkFormat, VkSampleCountFlags> m_multisampleCountsByFormat;

MobileGL/MG_Backend/DirectVulkan/Renderer/VulkanRenderer.cpp

@@ -19,6 +19,7 @@
 #include "MG_State/GLState/SamplerState/SamplerObject.h"
 #include "MG_State/GLState/TextureState/TextureObject.h"
 #include "MG_Impl/GLImpl/Framebuffer/GL_Framebuffer.h"
+#include "MG_Impl/GLImpl/Texture/GL_Texture.h"
 #include "MG_Util/Converters/GLToMG/TextureEnumConverter.h"
 // Only reached from an MGLOG_W, which the shipping INFO log level compiles out - so the
 // missing include never broke a default build and did break every WARN/DEBUG-level one.
@@ -30,6 +31,7 @@
 #include "MG_Util/Texture/PixelStoreProcessor.h"
 #include <Config.h>
 #include <algorithm>
+#include <bit>
 #include <cstdlib>
 #include <cstring>
 #include <vulkan/utility/vk_format_utils.h>
@@ -1258,7 +1260,8 @@ namespace MobileGL::MG_Backend::DirectVulkan {
 
         return depthAttachment.GetTexture().get() != stencilAttachment.GetTexture().get() ||
                depthAttachment.GetTextureUploadTarget() != stencilAttachment.GetTextureUploadTarget() ||
-               depthAttachment.GetTextureLevel() != stencilAttachment.GetTextureLevel();
+               ToStorageMipLevel(depthAttachment.GetTexture().get(), depthAttachment.GetTextureLevel()) !=
+                   ToStorageMipLevel(stencilAttachment.GetTexture().get(), stencilAttachment.GetTextureLevel());
     }
 
     static Bool IsColorAttachment(FramebufferAttachmentType attachmentType) {
@@ -1416,6 +1419,48 @@ void main() {
             return {width, height, depth};
         }
 
+        // How many components of a GL-space texel size actually halve down the mip chain. An array
+        // texture's LAYER count is not a dimension of the image (GL 4.6 core 8.14.3): it stays put
+        // all the way down, and GetMipmapTexelSize parks it in the slot after the image's own
+        // dimensions. This is the same split IsMipmapCompleteForFilter applies, and the two have to
+        // agree - allocating a chain whose layer count shrinks builds levels the completeness rule
+        // then rejects. Vulkan-space extents need none of this: layers live in arrayLayers there,
+        // so resource->depth is already 1 for every array target.
+        static Int MipShrinkingComponentCount(TextureTarget target) {
+            switch (target) {
+            case TextureTarget::Texture1DArray:
+                return 1;
+            case TextureTarget::Texture2DArray:
+            case TextureTarget::TextureCubeMapArray:
+                return 2;
+            default:
+                return 3;
+            }
+        }
+
+        static IntVec3 ComputeMipTexelSizeWithFixedComponents(const IntVec3& baseTexelSize, Uint32 relativeMipLevel,
+                                                              Int shrinkingComponents) {
+            IntVec3 size = baseTexelSize;
+            for (Int component = 0; component < shrinkingComponents && component < 3; ++component) {
+                size[component] = std::max<Int>(size[component] >> static_cast<Int>(relativeMipLevel), 1);
+            }
+            return size;
+        }
+
+        static Uint32 ComputeFullMipLevelCountWithFixedComponents(const IntVec3& baseTexelSize,
+                                                                  Int shrinkingComponents) {
+            Int maxDimension = 1;
+            for (Int component = 0; component < shrinkingComponents && component < 3; ++component) {
+                maxDimension = std::max<Int>(maxDimension, baseTexelSize[component]);
+            }
+            Uint32 mipLevelCount = 1;
+            while (maxDimension > 1) {
+                maxDimension = std::max<Int>(maxDimension / 2, 1);
+                ++mipLevelCount;
+            }
+            return mipLevelCount;
+        }
+
         static Bool EnsureGenerateMipmapStorageAllocated(::MobileGL::MG_State::GLState::TextureObjectMipmap& texture,
                                                          Uint32 baseMipLevel) {
             const Uint32 existingMipLevelCount = static_cast<Uint32>(texture.GetMipmapLevelCount());
@@ -1428,6 +1473,8 @@ void main() {
                 return false;
             }
 
+            const Int shrinkingComponents = MipShrinkingComponentCount(texture.GetTarget());
+
             for (const auto uploadTarget : uploadTargets) {
                 const IntVec3 baseTexelSize = texture.GetMipmapTexelSize(uploadTarget, baseMipLevel);
                 const SizeT baseByteSize = texture.GetMipmapByteSize(uploadTarget, baseMipLevel);
@@ -1444,13 +1491,15 @@ void main() {
                 }
 
                 const SizeT bytesPerTexel = baseByteSize / baseTexelCount;
-                const Uint32 requiredMipLevelCount = baseMipLevel + ComputeFullMipLevelCount(baseTexelSize);
+                const Uint32 requiredMipLevelCount =
+                    baseMipLevel + ComputeFullMipLevelCountWithFixedComponents(baseTexelSize, shrinkingComponents);
                 if (existingMipLevelCount >= requiredMipLevelCount) {
                     continue;
                 }
 
                 for (Uint32 level = existingMipLevelCount; level < requiredMipLevelCount; ++level) {
-                    const IntVec3 levelTexelSize = ComputeMipTexelSize(baseTexelSize, level - baseMipLevel);
+                    const IntVec3 levelTexelSize = ComputeMipTexelSizeWithFixedComponents(
+                        baseTexelSize, level - baseMipLevel, shrinkingComponents);
                     const SizeT levelByteSize = bytesPerTexel * static_cast<SizeT>(levelTexelSize.x()) *
                                                 static_cast<SizeT>(levelTexelSize.y()) *
                                                 static_cast<SizeT>(levelTexelSize.z());
@@ -1475,11 +1524,15 @@ void main() {
         static Uint32 ResolveAttachmentBaseArrayLayer(const MG_State::GLState::FramebufferAttachmentObject& attachment) {
             const TextureUploadTarget uploadTarget = attachment.GetTextureUploadTarget();
             if (IsCubeMapFaceUploadTarget(uploadTarget)) {
-                return static_cast<Uint32>(uploadTarget) - static_cast<Uint32>(TextureUploadTarget::CubeMapPositiveX);
+                // The face index IS the layer index, so it takes the same view shift as one that
+                // arrived through GetTextureLayer (see ToStorageArrayLayer).
+                const Int face = static_cast<Int>(uploadTarget) -
+                                 static_cast<Int>(TextureUploadTarget::CubeMapPositiveX);
+                return ToStorageArrayLayer(attachment.GetTexture().get(), face);
             }
             // Every other layered attachment names its layer directly. Returning 0 regardless made
             // every blit, copy and ReadPixels against such an attachment read layer zero.
-            return static_cast<Uint32>(std::max(attachment.GetTextureLayer(), 0));
+            return ToStorageArrayLayer(attachment.GetTexture().get(), attachment.GetTextureLayer());
         }
 
         // A 3D image has arrayLayers == 1: its "layer" is a z slice, which has to travel as an
@@ -1755,10 +1808,10 @@ void main() {
             outBinding.sampleCount = resource->sampleCount;
             const auto attachmentExtent = attachment.GetSize();
             outBinding.extent = {attachmentExtent.x(), attachmentExtent.y()};
-            outBinding.mipLevel = static_cast<Uint32>(std::max(attachment.GetTextureLevel(), 0));
+            outBinding.mipLevel = ToStorageMipLevel(attachment.GetTexture().get(), attachment.GetTextureLevel());
             outBinding.mipLevelCount = resource->mipLevels;
             if (AttachmentIsDepthSlice(attachment)) {
-                outBinding.depthOffset = static_cast<Uint32>(std::max(attachment.GetTextureLayer(), 0));
+                outBinding.depthOffset = ToStorageArrayLayer(attachment.GetTexture().get(), attachment.GetTextureLayer());
                 outBinding.baseArrayLayer = 0;
             } else {
                 outBinding.baseArrayLayer = ResolveAttachmentBaseArrayLayer(attachment);
@@ -1871,10 +1924,10 @@ void main() {
             outBinding.sampleCount = resource->sampleCount;
             const auto attachmentExtent = attachment.GetSize();
             outBinding.extent = {attachmentExtent.x(), attachmentExtent.y()};
-            outBinding.mipLevel = static_cast<Uint32>(std::max(attachment.GetTextureLevel(), 0));
+            outBinding.mipLevel = ToStorageMipLevel(attachment.GetTexture().get(), attachment.GetTextureLevel());
             outBinding.mipLevelCount = resource->mipLevels;
             if (AttachmentIsDepthSlice(attachment)) {
-                outBinding.depthOffset = static_cast<Uint32>(std::max(attachment.GetTextureLayer(), 0));
+                outBinding.depthOffset = ToStorageArrayLayer(attachment.GetTexture().get(), attachment.GetTextureLayer());
                 outBinding.baseArrayLayer = 0;
             } else {
                 outBinding.baseArrayLayer = ResolveAttachmentBaseArrayLayer(attachment);
@@ -2020,10 +2073,10 @@ void main() {
             outBinding.sampleCount = resource->sampleCount;
             const auto attachmentExtent = attachment.GetSize();
             outBinding.extent = {attachmentExtent.x(), attachmentExtent.y()};
-            outBinding.mipLevel = static_cast<Uint32>(std::max(attachment.GetTextureLevel(), 0));
+            outBinding.mipLevel = ToStorageMipLevel(attachment.GetTexture().get(), attachment.GetTextureLevel());
             outBinding.mipLevelCount = 1;
             if (AttachmentIsDepthSlice(attachment)) {
-                outBinding.depthOffset = static_cast<Uint32>(std::max(attachment.GetTextureLayer(), 0));
+                outBinding.depthOffset = ToStorageArrayLayer(attachment.GetTexture().get(), attachment.GetTextureLayer());
                 outBinding.baseArrayLayer = 0;
             } else {
                 outBinding.baseArrayLayer = ResolveAttachmentBaseArrayLayer(attachment);
@@ -3121,7 +3174,9 @@ void main() {
         m_samplerManager = MakeUnique<VkSamplerManager>();
         MOBILEGL_ASSERT(m_samplerManager != nullptr, "VkSamplerManager creation failed.");
         succeeded = m_samplerManager->Initialize({m_device, &m_config, m_samplerAnisotropyFeatureEnabled,
-                                                  m_physicalDevice.properties.limits.maxSamplerAnisotropy});
+                                                  m_physicalDevice.properties.limits.maxSamplerAnisotropy,
+                                                  m_customBorderColorFeatureEnabled,
+                                                  m_maxCustomBorderColorSamplers});
         MOBILEGL_ASSERT(succeeded, "VkSamplerManager initialization failed.");
         succeeded = InitializeBlitResources();
         MOBILEGL_ASSERT(succeeded, "Blit pipeline resource initialization failed.");
@@ -3131,7 +3186,7 @@ void main() {
         m_uniformManager = MakeUnique<UniformManager>();
         MOBILEGL_ASSERT(m_uniformManager != nullptr, "UniformDescriptorBinder creation failed.");
         succeeded = m_uniformManager->Initialize(
-            m_device, &m_bufferManager, m_programFactory.get(),
+            m_device, m_physicalDevice.handle, &m_bufferManager, m_programFactory.get(),
             m_physicalDevice.properties.limits.minUniformBufferOffsetAlignment, m_config.MaxFramesInFlight,
             maxProgramBindings, kDescriptorSetsPerFrame, m_textureManager.get(), m_samplerManager.get());
         MOBILEGL_ASSERT(succeeded, "UniformDescriptorBinder initialization failed.");
@@ -3910,10 +3965,15 @@ void main() {
         // Copies index data, replacing every occurrence of the application's arbitrary restart
         // index with the fixed all-ones value of the index type - the only one Vulkan restarts
         // on. An index that already equals the fixed value would then be indistinguishable from
-        // a restart, so it is nudged to the next-lowest value: it can only be a real index (the
-        // application's restart index is a different number), and the vertex it selects is
-        // outside any well-defined draw anyway, whereas leaving it alone would tear the
-        // primitive in two.
+        // a restart, so it is nudged to the next-lowest value, which silently draws the wrong
+        // vertex. That is a real (if narrow) loss and it is reported once rather than left
+        // invisible; DirectGLES avoids it for 8- and 16-bit indices by widening the copy instead,
+        // and the same treatment here is follow-up work.
+        //
+        // The caller guarantees applicationRestartIndex fits the index type, so no truncating
+        // cast is needed - and none may be used: truncating turns glPrimitiveRestartIndex(0x100)
+        // over 8-bit indices into "restart on index 0", which shreds every primitive that
+        // references vertex 0.
         void RewriteRestartIndices(const void* source, SizeT sizeBytes, VkIndexType indexType,
                                    Uint32 applicationRestartIndex, Vector<Uint8>& output) {
             output.resize(sizeBytes);
@@ -3927,6 +3987,12 @@ void main() {
                     if (indices[i] == static_cast<decltype(fixedMax)>(applicationRestartIndex)) {
                         indices[i] = fixedMax;
                     } else if (indices[i] == fixedMax) {
+                        MGLOG_E_ONCE("GL_PRIMITIVE_RESTART with restart index %u over index data that also uses "
+                                     "the all-ones index %u: both cannot be spelled at this index width, so every "
+                                     "all-ones index is drawn one vertex lower. Use "
+                                     "GL_PRIMITIVE_RESTART_FIXED_INDEX, or keep the all-ones value out of the "
+                                     "index data.",
+                                     applicationRestartIndex, static_cast<Uint32>(fixedMax));
                         indices[i] = fixedMax - 1;
                     }
                 }
@@ -3979,14 +4045,13 @@ void main() {
         const RenderStateParameters& rsp = MG_State::pGLContext->GetRenderStateParameters();
         if (rsp.PrimitiveRestartEnabled && !rsp.PrimitiveRestartFixedIndexEnabled) {
             const Uint32 restartIndex = rsp.PrimitiveRestartIndex;
-            Uint32 fixedMax = 0;
-            switch (vkIndexType) {
-            case VK_INDEX_TYPE_UINT8: fixedMax = 0xFFu; break;
-            case VK_INDEX_TYPE_UINT16: fixedMax = 0xFFFFu; break;
-            case VK_INDEX_TYPE_UINT32: fixedMax = 0xFFFFFFFFu; break;
-            default: break;
-            }
-            substituteRestart = restartIndex != fixedMax;
+            const Uint32 fixedMax = MG_Util::FixedRestartIndexForGLType(pIndexBufferView->indexType);
+            // STRICTLY less, and never truncated. Equal needs no rewrite (the driver already
+            // restarts there); GREATER means the index type cannot hold the application's restart
+            // index, so GL 4.6 core 10.3.6 says nothing matches it and the draw restarts nowhere -
+            // which is exactly what ResolvePrimitiveRestartEnable told the pipeline, so rewriting
+            // here would put restarts into a stream the pipeline was built not to restart on.
+            substituteRestart = restartIndex < fixedMax;
             substituteRestartIndex = restartIndex;
         }
 
@@ -4702,8 +4767,9 @@ void main() {
     // build in GetOrCreatePipeline - any new GL-state read there must be added here:
     //   - capability bits: CullFace, DepthTest, PolygonOffsetFill (mode gating rides
     //     the memo's mode key), RasterizerDiscard, ColorLogicOp, StencilTest,
-    //     PrimitiveRestart(+FixedIndex), plus the depth write mask
-    //   - patch vertices, polygon mode, cull face mode, depth func, logic op
+    //     PrimitiveRestart(+FixedIndex), SampleShading, plus the depth write mask
+    //   - patch vertices, polygon mode, cull face mode, depth func, logic op,
+    //     min sample shading
     //   - front/back stencil ops + compare funcs (ref/mask are dynamic state)
     //   - per draw buffer up to the render pass's colour span: indexed blend enable,
     //     blend factors/equations, indexed colour write mask (broadcast from index 0
@@ -4728,8 +4794,31 @@ void main() {
         capabilityBits |= p.PrimitiveRestartEnabled ? 1ull << 6 : 0;
         capabilityBits |= p.PrimitiveRestartFixedIndexEnabled ? 1ull << 7 : 0;
         capabilityBits |= p.DepthMask ? 1ull << 8 : 0;
+        capabilityBits |= p.SampleShadingEnabled ? 1ull << 9 : 0;
         Uint64 hash = CombinePipelineStateWord(0x243F6A8885A308D3ull, capabilityBits);
+        // glMinSampleShading. Hashed by BITS, not by value: this memo compares hashes rather than
+        // versions, so an unhashed float would let a pipeline built at one rate be handed back
+        // after glMinSampleShading moved it - the memo would see identical state.
+        {
+            Uint32 minSampleShadingBits = 0;
+            std::memcpy(&minSampleShadingBits, &p.MinSampleShadingValue, sizeof(minSampleShadingBits));
+            hash = CombinePipelineStateWord(hash, static_cast<Uint64>(minSampleShadingBits));
+        }
         hash = CombinePipelineStateWord(hash, static_cast<Uint64>(p.PatchVertices));
+        // The default tessellation levels belong here for the same reason PatchVertices does:
+        // when a program has an evaluation stage and no control stage, both are compiled into the
+        // synthesized pass-through control stage, so two draws that differ only in a level need
+        // different pipelines. Hashed over the RAW BITS so a NaN level - which glPatchParameterfv
+        // accepts - keys to itself. Six extra words on a path that only recomputes when the
+        // pipeline-state version moved.
+        for (Uint32 i = 0; i < 4; ++i) {
+            hash = CombinePipelineStateWord(hash,
+                                            static_cast<Uint64>(std::bit_cast<Uint32>(p.PatchDefaultOuterLevel[i])));
+        }
+        for (Uint32 i = 0; i < 2; ++i) {
+            hash = CombinePipelineStateWord(hash,
+                                            static_cast<Uint64>(std::bit_cast<Uint32>(p.PatchDefaultInnerLevel[i])));
+        }
         hash = CombinePipelineStateWord(hash, static_cast<Uint64>(p.PolygonModeFront));
         hash = CombinePipelineStateWord(hash, static_cast<Uint64>(p.CullFaceModeSetting));
         hash = CombinePipelineStateWord(hash, static_cast<Uint64>(p.DepthFunc));
@@ -4774,13 +4863,42 @@ void main() {
         return program.HasLinkedShaderStage(ShaderStage::Geometry);
     }
 
+    // GL primitive restart is defined on the INDEX STREAM (GL 4.6 core 10.3.6): it splits
+    // primitives when a fetched index matches PRIMITIVE_RESTART_INDEX. Two consequences the
+    // capability bits alone cannot express, both resolved here because only the caller knows them:
+    //
+    //  - A non-indexed draw has no index stream, so restart is a no-op for it. Leaving the
+    //    pipeline's primitiveRestartEnable on for a glDrawArrays is what made the list-topology
+    //    guard below refuse those draws, so an application that enables GL_PRIMITIVE_RESTART once
+    //    at init lost every glDrawArrays on a device without the extension.
+    //  - The comparison is against the full 32-bit restart index with the fetched index
+    //    zero-extended, so a restart index the type cannot hold (0x100FF against UNSIGNED_BYTE
+    //    data) matches no index and that draw restarts NOWHERE. UploadAndBindIndexBuffer makes the
+    //    same call for the rewrite, and the two must agree or the pipeline says "restart" over
+    //    index data nothing rewrote.
+    Bool VulkanRenderer::ResolvePrimitiveRestartEnable(Flags<DrawSetupAspect> aspects,
+                                                       const IndexBufferView* pIndexBufferView) const {
+        if (!(aspects & DrawSetupAspect::IndexBuffer) || pIndexBufferView == nullptr) {
+            return false;
+        }
+        const RenderStateParameters& rsp = MG_State::pGLContext->GetRenderStateParameters();
+        if (rsp.PrimitiveRestartFixedIndexEnabled) {
+            return true;
+        }
+        if (!rsp.PrimitiveRestartEnabled) {
+            return false;
+        }
+        return rsp.PrimitiveRestartIndex <= MG_Util::FixedRestartIndexForGLType(pIndexBufferView->indexType);
+    }
+
     VkPipeline VulkanRenderer::GetOrCreatePipeline(
             GLenum mode,
             const MG_State::GLState::ProgramObject& program,
             const ProgramFactory::VkProgramObject& programObj,
             ProgramFactory::CompileOptionFlags transformFlags,
             const MG_State::GLState::VertexArrayObject& vao,
-            const RenderPassEntry& renderPassEntry) {
+            const RenderPassEntry& renderPassEntry,
+            Bool primitiveRestartEnable) {
         Bool invertClockwise = transformFlags & ProgramFactory::CompileOptionBit::PositionYFlip;
         if (programObj.stages.empty()) {
             MGLOG_D("GetOrCreatePipeline skipped: program has no shader stages");
@@ -4822,6 +4940,7 @@ void main() {
                 entry.programHash == programObj.hash && entry.vertexInputHash == vertexLayoutHash &&
                 entry.renderPassHash == renderPassHash &&
                 entry.pipelineStateHash == pipelineStateHash &&
+                entry.primitiveRestartEnable == primitiveRestartEnable &&
                 entry.transformFlags == transformFlags) {
                 return entry.pipeline;
             }
@@ -5020,22 +5139,50 @@ void main() {
                 : VK_POLYGON_MODE_FILL;
 
         const VkPrimitiveTopology vkTopology = MG_Util::ConvertPrimitiveModeToVkEnum(mode);
-        const Bool primitiveRestartEnabled =
-            MG_State::pGLContext->IsCapabilityEnabled(CapabilityInput::PrimitiveRestart) ||
-            MG_State::pGLContext->IsCapabilityEnabled(CapabilityInput::PrimitiveRestartFixedIndex);
+        // Resolved by the caller (ResolvePrimitiveRestartEnable), which knows whether the draw is
+        // indexed and with what index type; the capability bits alone answer neither.
+        Bool primitiveRestartEnabled = primitiveRestartEnable;
+
+        // GL applies restart to PATCHES only when PRIMITIVE_RESTART_FOR_PATCHES_SUPPORTED is true
+        // (GL 4.6 core 10.3.6). MobileGL supports no such thing - neither backend has a way to
+        // restart a patch stream - and GL_FALSE is a legal answer to that query, so a patch draw
+        // simply never restarts here. Doing this BEFORE the feature guard below is what keeps a
+        // perfectly ordinary GL_PATCHES draw from being refused on a device that lacks
+        // VK_EXT_primitive_topology_list_restart. (When GL_PRIMITIVE_RESTART_FOR_PATCHES_SUPPORTED
+        // is eventually added to glGetIntegerv it has to report GL_FALSE to stay consistent with
+        // this.)
+        if (vkTopology == VK_PRIMITIVE_TOPOLOGY_PATCH_LIST) {
+            primitiveRestartEnabled = false;
+        }
+
         // Primitive restart on a *list* topology requires the primitiveTopologyListRestart feature;
-        // strip/fan restart works without it. Silently dropping restarts would corrupt geometry, so
-        // hard-fail here (at the draw) with the reason when the device lacks the feature.
+        // strip/fan restart works without it. There is no fallback - silently dropping the restarts
+        // would weld the primitives on either side of each one together - so the draw is declined
+        // here with the reason.
+        //
+        // Declined, not thrown. This used to THROW_EXCEPTION, which unwinds a C++ exception through
+        // the C GL ABI and takes the process down (the hazard GL_Texture.cpp and RenderState.cpp
+        // already name); an application that merely enabled a legal desktop feature died instead of
+        // getting a draw that rendered nothing. VK_NULL_HANDLE is this function's established
+        // "skip this draw" answer, used by the no-stages case above.
+        //
+        // Reached only when this draw's index stream really does restart. Testing the raw
+        // capability bits here instead - which is what it did - refused every NON-INDEXED
+        // list-topology draw as well, so an application that enables GL_PRIMITIVE_RESTART once at
+        // init and then calls glDrawArrays(GL_TRIANGLES, ...) rendered nothing at all.
         const auto isListTopology = [](VkPrimitiveTopology t) {
             return t == VK_PRIMITIVE_TOPOLOGY_POINT_LIST || t == VK_PRIMITIVE_TOPOLOGY_LINE_LIST ||
                    t == VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST ||
                    t == VK_PRIMITIVE_TOPOLOGY_LINE_LIST_WITH_ADJACENCY ||
-                   t == VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST_WITH_ADJACENCY || t == VK_PRIMITIVE_TOPOLOGY_PATCH_LIST;
+                   t == VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST_WITH_ADJACENCY;
         };
         if (primitiveRestartEnabled && !m_primitiveTopologyListRestartFeatureEnabled && isListTopology(vkTopology)) {
-            THROW_EXCEPTION("Primitive restart on a list topology requires the primitiveTopologyListRestart device "
-                            "feature (VK_EXT_primitive_topology_list_restart), which this device does not support; use "
-                            "a strip/fan topology or a device that supports it.");
+            MGLOG_E_ONCE("Draw skipped: primitive restart on a list topology (0x%x) requires the "
+                         "primitiveTopologyListRestart device feature (VK_EXT_primitive_topology_list_restart), "
+                         "which this device does not support; use a strip/fan topology, or disable primitive "
+                         "restart for list-topology draws.",
+                         mode);
+            return VK_NULL_HANDLE;
         }
 
         PipelineFactory::PipelineCreatePayload payload {
@@ -5045,6 +5192,13 @@ void main() {
             .renderPass = renderPassEntry.renderPass,
             .colorAttachmentCount = renderPassEntry.colorAttachmentCount,
             .rasterizationSamples = renderPassEntry.sampleCount,
+            // ARB_sample_shading. Dropped on a device without sampleRateShading rather than
+            // hard-failing the draw: the rate is a hint, and the pipeline renders correctly at the
+            // driver's own rate. Both halves move the render state's PIPELINE version, so a cached
+            // pipeline built at the old rate cannot be handed back for the new one.
+            .sampleShadingEnable = m_sampleRateShadingFeatureEnabled &&
+                                   MG_State::pGLContext->IsCapabilityEnabled(CapabilityInput::SampleShading),
+            .minSampleShading = MG_State::pGLContext->GetMinSampleShadingValue(),
             .subpass = 0,
             .topology = vkTopology,
             .primitiveRestartEnable = primitiveRestartEnabled,
@@ -5101,10 +5255,21 @@ void main() {
         // program with a tessellation stage may only be drawn with GL_PATCHES), so nothing legal
         // loses its pass-through here; what it does lose is the pipeline, because the refusal
         // below then sees an evaluation stage with no control stage and declines.
+        //
+        // The default tessellation levels (glPatchParameterfv) are draw state for the same reason
+        // and are compiled into the same module, so they are read here too and their key is mixed
+        // into the pipeline hash - without that a pipeline memoised at one set of levels would be
+        // handed back after the application changed them.
         if (programObj.needsPassthroughTessControl && programObj.passthroughTessControlEmulatable &&
             vkTopology == VK_PRIMITIVE_TOPOLOGY_PATCH_LIST) {
-            payload.passthroughTessControlStage =
-                m_programFactory->GetOrCreatePassthroughTessControlStage(payload.patchControlPoints);
+            const FloatVec4& defaultOuterLevel = MG_State::pGLContext->GetPatchDefaultOuterLevel();
+            const FloatVec2& defaultInnerLevel = MG_State::pGLContext->GetPatchDefaultInnerLevel();
+            payload.passthroughTessControlKey = ProgramFactory::ComputePassthroughTessControlKey(
+                payload.patchControlPoints, defaultOuterLevel, defaultInnerLevel,
+                programObj.passthroughPerVertexMembers);
+            payload.passthroughTessControlStage = m_programFactory->GetOrCreatePassthroughTessControlStage(
+                payload.patchControlPoints, defaultOuterLevel, defaultInnerLevel,
+                programObj.passthroughPerVertexMembers);
         }
         if (!payload.stencilTestEnable) {
             payload.frontStencilFailOp = VK_STENCIL_OP_KEEP;
@@ -5304,8 +5469,21 @@ void main() {
                     colorAttachmentFormat = m_swapchainObject.GetSurfaceFormat().format;
                 } else if (colorAttachmentRenderbuffer != nullptr) {
                     textureExternalIndex = static_cast<Int>(colorAttachmentRenderbuffer->GetExternalIndex());
-                    colorAttachmentFormat = MG_Util::ConvertTextureInternalFormatToVkEnum(
-                        colorAttachmentRenderbuffer->GetInternalFormat());
+                    // The SAME resolver GetOrCreateRenderbufferResource backs the image with, so the
+                    // probe cannot ask about a format the attachment does not have. The strict 1:1
+                    // converter is the wrong question here and answered VK_FORMAT_UNDEFINED for
+                    // RGBA2/RGBA12/RGB10/RGB12/RGB16 and the packed 16-bit formats for RGBA4/RGB5_A1
+                    // - and VkFormatProperties for UNDEFINED are all zero, so blending was
+                    // force-disabled forever on attachments that blend perfectly well. Every
+                    // three-channel colour renderbuffer was in that set too (R8G8B8_UNORM is rarely
+                    // supported), which is the more ordinary shape.
+                    //
+                    // Resolved rather than looked up: GetOrCreateRenderbufferResource creates images
+                    // and bumps epochs, which a pipeline-state query must not do as a side effect.
+                    // A renderbuffer has no device-fallback step after the resolver (unlike the
+                    // texture path's D24 -> D32 substitution), so the resolver IS its live format.
+                    colorAttachmentFormat =
+                        ResolveTextureFormatInfo(colorAttachmentRenderbuffer->GetInternalFormat()).format;
                 } else {
                     auto* texture = colorAttachmentTexture;
                     MOBILEGL_ASSERT(texture != nullptr,
@@ -5386,6 +5564,7 @@ void main() {
             entry.vertexInputHash = vertexLayoutHash;
             entry.renderPassHash = renderPassHash;
             entry.pipelineStateHash = pipelineStateHash;
+            entry.primitiveRestartEnable = primitiveRestartEnable;
             entry.transformFlags = transformFlags;
             entry.pipeline = pipeline;
             m_pipelineMemoNext = (m_pipelineMemoNext + 1) % kPipelineMemoSize;
@@ -5770,7 +5949,11 @@ void main() {
             return false;
         }
         SetupDrawSnapshot& snap = *snapPtr;
-        if (snap.aspects != aspects.GetRaw() || snap.mode != mode) {
+        // Resolved once for the whole function: it guards the snapshot, keys the pipeline memo
+        // probe below, and is handed to GetOrCreatePipeline on a miss - all three must agree.
+        const Bool drawPrimitiveRestartEnable = ResolvePrimitiveRestartEnable(aspects, pIndexBufferView);
+        if (snap.aspects != aspects.GetRaw() || snap.mode != mode ||
+            snap.primitiveRestartEnable != drawPrimitiveRestartEnable) {
             return false;
         }
         if (m_clearManager->HasAnyPendingClears()) {
@@ -6040,6 +6223,7 @@ void main() {
                     entry.programHash == programObj.hash && entry.vertexInputHash == vaoLayoutHash &&
                     entry.renderPassHash == snap.renderPassHash &&
                     entry.pipelineStateHash == m_pipelineStateHash &&
+                    entry.primitiveRestartEnable == drawPrimitiveRestartEnable &&
                     entry.transformFlags == memoTransformFlags) {
                     pipeline = entry.pipeline;
                     break;
@@ -6050,14 +6234,17 @@ void main() {
                 // index, depth/stencil participation, image epochs, no pending clears)
                 // was verified unchanged above, so this is a pure cache hit on the same
                 // entry the snapshot's pipeline was built against.
-                const RenderPassEntry& renderPassEntry = m_renderPassManager->GetOrCreateRenderPass(
+                const RenderPassEntry* renderPassEntry = m_renderPassManager->GetOrCreateRenderPass(
                     *drawFbo, m_imageIndexAcquired, snap.drawUsesDepthStencil);
-                if (!activeRenderPass->CompatibleWith(renderPassEntry)) {
+                // A decline (nullptr) is an attachment DirectVulkan cannot represent; the builder
+                // has already logged it. Fall out of the fast path the same way an incompatible
+                // pass does - the full path re-resolves, declines again and drops the draw.
+                if (renderPassEntry == nullptr || !activeRenderPass->CompatibleWith(*renderPassEntry)) {
                     return false;
                 }
                 pipeline = GetOrCreatePipeline(mode, program, programObj,
                                                ProgramFactory::CompileOptionFlags(snap.resolvedTransformFlags),
-                                               vao, renderPassEntry);
+                                               vao, *renderPassEntry, drawPrimitiveRestartEnable);
                 if (pipeline == VK_NULL_HANDLE) {
                     return false;
                 }
@@ -6442,12 +6629,23 @@ void main() {
             MG_State::pGLContext->IsCapabilityEnabled(CapabilityInput::DepthTest) ||
             MG_State::pGLContext->IsCapabilityEnabled(CapabilityInput::StencilTest);
         auto* renderPassEntry =
-            &m_renderPassManager->GetOrCreateRenderPass(*drawFbo, m_imageIndexAcquired, drawUsesDepthStencil);
+            m_renderPassManager->GetOrCreateRenderPass(*drawFbo, m_imageIndexAcquired, drawUsesDepthStencil);
+        // nullptr: the framebuffer has an attachment DirectVulkan cannot represent (a texture the
+        // texture manager declined to back, or a view it could not build). The builder logged which
+        // one; drop the draw here, exactly as an unresolvable sampler descriptor drops one in
+        // BindProgramUniformBuffers. Before this existed the same condition dereferenced a null
+        // resource or handed VK_NULL_HANDLE to vkCreateFramebuffer and took the process down.
+        if (renderPassEntry == nullptr) {
+            return false;
+        }
         if (activeRenderPass && !activeRenderPass->CompatibleWith(*renderPassEntry)) {
             VkRenderPassManager::EndRenderPass(frame.commandBuffer);
             activeRenderPass = nullptr;
             renderPassEntry =
-                &m_renderPassManager->GetOrCreateRenderPass(*drawFbo, m_imageIndexAcquired, drawUsesDepthStencil);
+                m_renderPassManager->GetOrCreateRenderPass(*drawFbo, m_imageIndexAcquired, drawUsesDepthStencil);
+            if (renderPassEntry == nullptr) {
+                return false;
+            }
         }
         if (renderPassEntry->attachmentCount == 0 || renderPassEntry->extent.x() <= 0 || renderPassEntry->extent.y() <= 0) {
             MGLOG_D("SetupDraw skipped: drawFbo=%u resolved to an empty render pass (attachmentCount=%u extent=%dx%d)",
@@ -6494,7 +6692,8 @@ void main() {
             }
         }
 
-        auto pipeline = GetOrCreatePipeline(mode, program, programObj, transformFlags, vao, *renderPassEntry);
+        auto pipeline = GetOrCreatePipeline(mode, program, programObj, transformFlags, vao, *renderPassEntry,
+                                            ResolvePrimitiveRestartEnable(aspects, pIndexBufferView));
         // GetOrCreatePipeline documents a VK_NULL_HANDLE return (empty stages, or a driver that
         // rejected vkCreateGraphicsPipelines). Binding it dereferences null inside the driver -
         // 9 of the 15 CTS process deaths were exactly this vkCmdBindPipeline. A draw that has no
@@ -6557,6 +6756,7 @@ void main() {
             if (nowActiveRenderPass != nullptr && !programObj.hasStorageImages) {
                 snap.valid = true;
                 snap.aspects = aspects.GetRaw();
+                snap.primitiveRestartEnable = ResolvePrimitiveRestartEnable(aspects, pIndexBufferView);
                 snap.mode = mode;
                 snap.programLifetimeId = program.GetLifetimeId();
                 snap.programVersion = program.GetBackendStateVersion();
@@ -6791,8 +6991,10 @@ void main() {
         }
 
         auto* activeRenderPass = VkRenderPassManager::GetActiveRenderPass();
-        auto* renderPassEntry = &m_renderPassManager->GetOrCreateRenderPass(framebuffer, m_imageIndexAcquired);
-        if (renderPassEntry->attachmentCount == 0 ||
+        auto* renderPassEntry = m_renderPassManager->GetOrCreateRenderPass(framebuffer, m_imageIndexAcquired);
+        // A declined render pass is the same answer as an empty one for a clear: there is nothing
+        // attached that can be cleared inside a pass. The builder has already logged the reason.
+        if (renderPassEntry == nullptr || renderPassEntry->attachmentCount == 0 ||
             renderPassEntry->extent.x() <= 0 || renderPassEntry->extent.y() <= 0) {
             return ScissoredClearPrep::NoOp;
         }
@@ -6822,7 +7024,10 @@ void main() {
             activeRenderPass = nullptr;
             // Re-resolve: ending the pass updates tracked attachment layouts, which feed the
             // entry's load ops and initial layouts.
-            renderPassEntry = &m_renderPassManager->GetOrCreateRenderPass(framebuffer, m_imageIndexAcquired);
+            renderPassEntry = m_renderPassManager->GetOrCreateRenderPass(framebuffer, m_imageIndexAcquired);
+            if (renderPassEntry == nullptr) {
+                return ScissoredClearPrep::NoOp;
+            }
         }
         // A still-active pass is necessarily compatible here: the block above ended any
         // incompatible one and nothing since can change the active pass.
@@ -7987,8 +8192,14 @@ void main() {
 
         // A color-only blit never touches depth/stencil: let the default-FBO pass
         // it opens skip the depth attachment (depth-less flavor).
-        auto& renderPassEntry =
+        auto* renderPassEntryPtr =
             m_renderPassManager->GetOrCreateRenderPass(drawFbo, m_imageIndexAcquired, /*drawUsesDepthStencil=*/false);
+        if (renderPassEntryPtr == nullptr) {
+            // Declined (the builder logged which attachment). The caller's contract for `false` is
+            // "this blit was not serviced here", which is the honest answer.
+            return false;
+        }
+        auto& renderPassEntry = *renderPassEntryPtr;
         const Bool ok = VkRenderPassManager::BeginRenderPass(frame.commandBuffer, renderPassEntry);
         MOBILEGL_ASSERT(ok, "%s: BeginRenderPass failed", __func__);
 
@@ -8939,6 +9150,12 @@ void main() {
             VkExtent2D extent = {0, 0};
             Uint32 depth = 1;
             Uint32 arrayLayers = 1;
+            // Both resources carry a format; this copy used to decline to read it, which is why a
+            // four-row drift between the texture and renderbuffer format tables turned into
+            // corrupted texels with nothing in the log. vkCmdCopyImage requires size-compatible
+            // formats whenever they differ (VUID-vkCmdCopyImage-srcImage-01548) and there is no
+            // downstream check - a mismatched pair is a promise the driver takes at face value.
+            VkFormat format = VK_FORMAT_UNDEFINED;
         };
 
         Bool TryResolveCopyImageSliceMapping(TextureTarget target, const CopyImageVkImage& image, Uint32 mipLevel,
@@ -9020,7 +9237,15 @@ void main() {
         // and an overlap check). Refused outright, and refused for real rather than through an
         // assertion the release build drops: recording the pair anyway is a validation error and,
         // on a tiler, a copy whose source has already been overwritten.
-        if (srcEndpoint.Texture == dstEndpoint.Texture && srcEndpoint.Renderbuffer == dstEndpoint.Renderbuffer) {
+        // Compared by STORAGE, not by GL object: a texture view and the texture it views are two
+        // different objects over one VkImage (ARB_texture_view), and GL 4.6 core 8.18 explicitly
+        // permits copying between them - so an object-identity test would let exactly the case
+        // this guard exists for through.
+        const auto* srcStorageTexture =
+            srcEndpoint.Texture ? &VkTextureManager::StorageTextureOf(*srcEndpoint.Texture) : nullptr;
+        const auto* dstStorageTexture =
+            dstEndpoint.Texture ? &VkTextureManager::StorageTextureOf(*dstEndpoint.Texture) : nullptr;
+        if (srcStorageTexture == dstStorageTexture && srcEndpoint.Renderbuffer == dstEndpoint.Renderbuffer) {
             MGLOG_E_ONCE("%s: in-place copy on objectId=%u is not supported; declining the copy", __func__,
                          CopyImageEndpointName(srcEndpoint));
             return;
@@ -9050,6 +9275,7 @@ void main() {
                 out.extent = resource->extent;
                 out.depth = 1;
                 out.arrayLayers = 1;
+                out.format = resource->format;
                 return out.image != VK_NULL_HANDLE;
             }
             // An endpoint that named nothing is the frontend validator's INVALID_VALUE and never
@@ -9065,6 +9291,7 @@ void main() {
             out.extent = resource->extent;
             out.depth = resource->depth;
             out.arrayLayers = resource->arrayLayers;
+            out.format = resource->format;
             return true;
         };
         CopyImageVkImage srcImage{};
@@ -9090,12 +9317,57 @@ void main() {
             MGLOG_E_ONCE("%s: source or destination image failed to sync; declining the copy", __func__);
             return;
         }
+        // Storage space from here down. srcImage/dstImage are the STORAGE textures' resources
+        // (SyncTextureAndGetDescriptor resolves a view to the texture it views), while srcLevel /
+        // dstLevel and the z origins below arrived relative to whichever name the application
+        // passed - so a view's level 0 has to become the parent level it opened onto before it
+        // can index a subresource, exactly as at every other attachment boundary.
+        srcLevel = static_cast<GLint>(ToStorageMipLevel(srcEndpoint.Texture.get(), srcLevel));
+        dstLevel = static_cast<GLint>(ToStorageMipLevel(dstEndpoint.Texture.get(), dstLevel));
+        srcZ = static_cast<GLint>(ToStorageArrayLayer(srcEndpoint.Texture.get(), srcZ));
+        dstZ = static_cast<GLint>(ToStorageArrayLayer(dstEndpoint.Texture.get(), dstZ));
         if (srcLevel < 0 || dstLevel < 0 || static_cast<Uint32>(srcLevel) >= srcImage.mipLevels ||
             static_cast<Uint32>(dstLevel) >= dstImage.mipLevels) {
             MGLOG_E_ONCE("%s: mip level out of range (src %d of %u, dst %d of %u); declining the copy", __func__,
                          srcLevel, srcImage.mipLevels, dstLevel, dstImage.mipLevels);
             return;
         }
+        // Size compatibility, the guard whose absence let a table drift two files away reach the
+        // driver as a promise. glCopyImageSubData is a raw texel-block move (GL 4.6 core 18.3.2), and
+        // Vulkan says as much: when the two formats differ they must be size-compatible - the same
+        // texel block size - or vkCmdCopyImage is undefined (VUID-vkCmdCopyImage-srcImage-01548).
+        // Nothing else on this path asks: the three checks around it cover the mip range, the region
+        // bounds and the slice range, and none of them ever looked at a format.
+        //
+        // A decline rather than a MOBILEGL_ASSERT, for the reason the neighbouring guards spell out:
+        // assertions compile out of the release build that the CTS and shipping both run, which is
+        // exactly where the corruption was observed.
+        if (srcImage.format != dstImage.format) {
+            // Size-compatibility is the COLOUR rule. Vulkan makes each depth/stencil format compatible
+            // only with ITSELF, and the texel block sizes cannot tell them apart: X8_D24_UNORM_PACK32,
+            // D32_SFLOAT and D24_UNORM_S8_UINT are all 4 bytes and all in different compatibility
+            // classes, so a raw block-size test waves through exactly the pairs Vulkan forbids. The
+            // frontend cannot filter them either - its own texel-block resolver is byte-size only, so
+            // glCopyImageSubData between a GL_DEPTH_COMPONENT24 texture and a GL_DEPTH_COMPONENT32F
+            // one reaches here with two different depth formats and 4 == 4.
+            const Bool eitherIsDepthStencil =
+                ((srcImage.aspect | dstImage.aspect) & (VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT)) != 0;
+            if (eitherIsDepthStencil) {
+                MGLOG_E_ONCE("%s: depth/stencil formats are compatible only with themselves, and source format "
+                             "%d differs from destination format %d; declining the copy",
+                             __func__, static_cast<Int>(srcImage.format), static_cast<Int>(dstImage.format));
+                return;
+            }
+            const Uint32 srcBlockSize = vkuGetFormatInfo(srcImage.format).texel_block_size;
+            const Uint32 dstBlockSize = vkuGetFormatInfo(dstImage.format).texel_block_size;
+            if (srcBlockSize == 0 || dstBlockSize == 0 || srcBlockSize != dstBlockSize) {
+                MGLOG_E_ONCE("%s: source format %d and destination format %d are not size-compatible "
+                             "(%u vs %u bytes per texel block); declining the copy",
+                             __func__, static_cast<Int>(srcImage.format), static_cast<Int>(dstImage.format),
+                             srcBlockSize, dstBlockSize);
+                return;
+            }
+        }
         const VkImageAspectFlags copyAspectMask =
             srcImage.aspect & dstImage.aspect &
             (VK_IMAGE_ASPECT_COLOR_BIT | VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT);
@@ -9828,8 +10100,8 @@ void main() {
             vkFormat = resource->format;
             trackedLayout = &resource->layout;
             imageAspect = resource->aspect;
-            mipLevel = static_cast<Uint32>(std::max(attachment.GetTextureLevel(), 0));
-            baseArrayLayer = static_cast<Uint32>(std::max(attachment.GetTextureLayer(), 0));
+            mipLevel = ToStorageMipLevel(attachment.GetTexture().get(), attachment.GetTextureLevel());
+            baseArrayLayer = ToStorageArrayLayer(attachment.GetTexture().get(), attachment.GetTextureLayer());
         } else if (attachment.IsRenderbuffer() && attachment.GetRenderbuffer()) {
             const auto& renderbufferObject = attachment.GetRenderbuffer();
             const Bool clearReady = MaterializePendingClearForRenderbuffer(frame.commandBuffer, renderbufferObject);
@@ -10168,9 +10440,38 @@ void main() {
         }
 
         auto* resource = m_textureManager->SyncTextureAndGetDescriptor(*textureObject);
-        if (resource == nullptr || resource->image == VK_NULL_HANDLE) {
-            MGLOG_E_ONCE("DirectVulkan::GetTexImage skipped: failed to sync textureId=%u",
-                    textureObject->GetExternalIndex());
+        // Two shapes end up in the same place, and for the same reason: the GL level being read has
+        // no GPU storage, so UploadDirtyMipLevels never wrote it and the CPU shadow is the ONLY copy
+        // of its bytes - which makes the shadow both the safe answer and the correct one.
+        //
+        //   (a) No VkImage at all. A mutable texture whose GL level 0 was never defined -
+        //       glTexImage2D(GL_TEXTURE_2D, 5, ...) and nothing else, exactly what the
+        //       clear_tex_image conformance cases build. VkTextureManager takes storage mip 0 as the
+        //       physical image extent (CheckMipmapCompleteness), so it refuses to back the texture.
+        //   (b) A VkImage with FEWER mip levels than the GL level count. GetUploadMipLevelCount
+        //       breaks at the first level with a zero extent, so "level 0 defined, a gap, level 3
+        //       defined" produces a one-mip image while GL_TEXTURE_MAX_LEVEL-style state still
+        //       reports four levels. The same clamp also fires on a base level small enough that the
+        //       full chain is shorter than the levels the application defined.
+        //
+        // (b) is the dangerous one and is why the level is bounded against the RESOURCE and not only
+        // against the GL-side count above: writing that level into imageSubresource.mipLevel is an
+        // out-of-range subresource, which is the promise the driver takes at face value. The
+        // glCopyImageSubData path two functions up carries the same guard for the same reason, added
+        // after it SIGSEGV'd inside the Adreno driver; the readback never had one.
+        const Bool hasImage = resource != nullptr && resource->image != VK_NULL_HANDLE;
+        const Bool levelIsBacked =
+            hasImage && ToStorageMipLevel(textureObject.get(), level) < resource->mipLevels;
+        if (!levelIsBacked) {
+            // Never gated on "syncing was inconvenient": a blanket shadow answer would silently
+            // return stale bytes for every render-to-texture result.
+            MGLOG_D("DirectVulkan::GetTexImage: textureId=%u level %d has no GPU storage (%s); answering "
+                    "from the CPU shadow",
+                    textureObject->GetExternalIndex(), level,
+                    hasImage ? "the image has fewer mip levels" : "the texture has no VkImage");
+            MG_Impl::GLImpl::CopyTextureImageToClientOrPBO_State(textureObject, textureUploadTarget, level, format,
+                                                                 type, bufSize, pixels,
+                                                                 "DirectVulkan::GetTextureImage");
             return;
         }
 
@@ -10192,10 +10493,14 @@ void main() {
                     textureMipmapObject->GetMipmapTexelSize(textureUploadTarget, static_cast<Uint>(level));
                 const Bool isCubeFace = textureUploadTarget >= TextureUploadTarget::CubeMapPositiveX &&
                     textureUploadTarget <= TextureUploadTarget::CubeMapNegativeZ;
-                const Uint32 arrayLayer = isCubeFace
-                    ? static_cast<Uint32>(textureUploadTarget) -
-                        static_cast<Uint32>(TextureUploadTarget::CubeMapPositiveX)
+                // Storage space: `resource` is the storage texture's, so a view's level and
+                // layer have to be shifted into its numbering (see ToStorageMipLevel).
+                const Int glArrayLayer = isCubeFace
+                    ? static_cast<Int>(textureUploadTarget) -
+                        static_cast<Int>(TextureUploadTarget::CubeMapPositiveX)
                     : 0;
+                const Uint32 arrayLayer = ToStorageArrayLayer(textureObject.get(), glArrayLayer);
+                const Uint32 storageLevel = ToStorageMipLevel(textureObject.get(), level);
                 // A 1D array's levelSize.y() is its LAYER count, and those layers are the rows
                 // GL wants back - but in Vulkan they are array layers of a one-row image, not
                 // rows of layer 0, so the read has to be told which of the two it is looking at.
@@ -10204,7 +10509,7 @@ void main() {
                         ? static_cast<Uint32>(std::max<Int>(levelSize.y(), 1))
                         : 1u;
                 ReadDepthStencilImageToClient(resource->image, resource->format, &resource->layout, resource->aspect,
-                                              static_cast<Uint32>(level), arrayLayer, 0, 0, levelSize.x(),
+                                              storageLevel, arrayLayer, 0, 0, levelSize.x(),
                                               levelSize.y(), format, type, pixels,
                                               /*defaultFramebufferOrientation=*/false, sourceLayers);
             } else {
@@ -10281,13 +10586,15 @@ void main() {
             frame.commandBuffer, resource->image, resource->layout, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL,
             srcStageMask, VK_PIPELINE_STAGE_TRANSFER_BIT,
             srcAccessMask, VK_ACCESS_TRANSFER_READ_BIT, resource->aspect,
-            static_cast<Uint32>(level), 1);
+            ToStorageMipLevel(textureObject.get(), level), 1);
         MOBILEGL_ASSERT(ok, "%s: failed to transition texture image", __func__);
 
         VkBufferImageCopy copyRegion{};
         copyRegion.imageSubresource.aspectMask = resource->aspect;
-        copyRegion.imageSubresource.mipLevel = static_cast<Uint32>(level);
-        copyRegion.imageSubresource.baseArrayLayer = 0;
+        // Storage space, as above: a texture view reads its own level 0 out of whichever level
+        // and layer of the parent it opened onto.
+        copyRegion.imageSubresource.mipLevel = ToStorageMipLevel(textureObject.get(), level);
+        copyRegion.imageSubresource.baseArrayLayer = ToStorageArrayLayer(textureObject.get(), 0);
         copyRegion.imageSubresource.layerCount = static_cast<Uint32>(arrayLayers);
         copyRegion.imageExtent = {static_cast<Uint32>(width),
                                   is1dArrayImage ? 1u : static_cast<Uint32>(height),
@@ -10302,7 +10609,7 @@ void main() {
             frame.commandBuffer, resource->image, resource->layout, originalLayout,
             VK_PIPELINE_STAGE_TRANSFER_BIT, restoreStageMask,
             VK_ACCESS_TRANSFER_READ_BIT, restoreAccessMask, resource->aspect,
-            static_cast<Uint32>(level), 1);
+            ToStorageMipLevel(textureObject.get(), level), 1);
         MOBILEGL_ASSERT(ok, "%s: failed to restore texture image layout", __func__);
 
         if (!SubmitReadbackCommandsAndWait(frame)) {
@@ -10323,15 +10630,23 @@ void main() {
 
     void VulkanRenderer::GenerateMipmap(GLenum target) {
         const auto textureTarget = MG_Util::ConvertGLEnumToTextureTarget(target);
-        // The other mipmappable targets - 1D, 1D array, cube map array - are legal GL and the front
-        // end lets them through, so reaching one here is a coverage gap in this backend, not a
-        // broken invariant. Declining leaves the mip chain unwritten; asserting took the process
-        // down with it.
+        // Whatever is left here is a coverage gap in this backend, not a broken invariant, so it
+        // declines (leaving the mip chain unwritten) rather than asserting the process down. What
+        // remains is the multisample targets, which GL 4.6 core 8.14.4 forbids to glGenerateMipmap
+        // outright.
+        //
+        // Every ARRAY target - 1D array, 2D array, cube map array - needs no blit code of its own:
+        // its layers live in the VkImage's arrayLayers, so resource->extent/depth already describe
+        // one layer's image and the loop below already copies every layer per level via
+        // srcSubresource.layerCount = resource->arrayLayers. The one thing they DO need is that the
+        // GL-space storage allocation not shrink the layer count down the chain, which
+        // MipShrinkingComponentCount handles.
         if (textureTarget != TextureTarget::Texture2D && textureTarget != TextureTarget::Texture2DArray &&
             textureTarget != TextureTarget::Texture3D && textureTarget != TextureTarget::TextureCubeMap &&
+            textureTarget != TextureTarget::TextureCubeMapArray &&
             // A 1D texture needs nothing special: its storage extent is {width, 1, 1}, so the blit
             // loop below already emits the y and z offsets of 0 and 1 that a 1D image requires.
-            textureTarget != TextureTarget::Texture1D) {
+            textureTarget != TextureTarget::Texture1D && textureTarget != TextureTarget::Texture1DArray) {
             MGLOG_W_ONCE("GenerateMipmap: unsupported target %s", MG_Util::ConvertTextureTargetToString(textureTarget).c_str());
             return;
         }
@@ -12681,6 +12996,11 @@ void main() {
         m_fillModeNonSolidFeatureEnabled = deviceFeatures.fillModeNonSolid == VK_TRUE;
         deviceFeatures.dualSrcBlend = supportedDeviceFeatures.dualSrcBlend;
         m_dualSrcBlendFeatureEnabled = deviceFeatures.dualSrcBlend == VK_TRUE;
+        // ARB_sample_shading. Without this feature a pipeline may not set sampleShadingEnable
+        // (VUID-VkPipelineMultisampleStateCreateInfo-sampleShadingEnable-00784), so the GL enable
+        // has to be dropped rather than forwarded - which is what the flag below records.
+        deviceFeatures.sampleRateShading = supportedDeviceFeatures.sampleRateShading;
+        m_sampleRateShadingFeatureEnabled = deviceFeatures.sampleRateShading == VK_TRUE;
         // ARB_viewport_array rasterization. Without multiViewport a pipeline may declare exactly
         // one viewport (VUID-VkPipelineViewportStateCreateInfo-viewportCount-01216), so a shader's
         // gl_ViewportIndex can only ever select viewport 0 and the other fifteen rectangles are
@@ -12948,6 +13268,55 @@ void main() {
             }
         }
 
+        // VK_EXT_custom_border_color: an arbitrary GL_TEXTURE_BORDER_COLOR, in float or integer form,
+        // instead of the four predefined VkBorderColor values. Without it a border outside
+        // transparent black / opaque black / opaque white has to be snapped, which is what made every
+        // border texel of a GL_RGBA8 texture with border (255,255,255,255) sample as 0 and what made
+        // an integer border of -1 come back as 0.
+        //
+        // customBorderColorWithoutFormat is required alongside customBorderColors, not merely
+        // preferred: a GL sampler object carries a border colour with no idea which texture it will
+        // be paired with, so the VkSamplerCustomBorderColorCreateInfoEXT this backend builds has to
+        // leave `format` VK_FORMAT_UNDEFINED.
+        m_customBorderColorFeatureEnabled = false;
+        m_maxCustomBorderColorSamplers = 0;
+        VkPhysicalDeviceCustomBorderColorFeaturesEXT customBorderColorFeatures{};
+        customBorderColorFeatures.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_CUSTOM_BORDER_COLOR_FEATURES_EXT;
+        if (IsExtensionSupported(availableExtensions, VK_EXT_CUSTOM_BORDER_COLOR_EXTENSION_NAME) &&
+            getPhysicalDeviceFeatures2 != nullptr) {
+            VkPhysicalDeviceFeatures2 featureQuery{};
+            featureQuery.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FEATURES_2;
+            featureQuery.pNext = &customBorderColorFeatures;
+            getPhysicalDeviceFeatures2(m_physicalDevice.handle, &featureQuery);
+            if (customBorderColorFeatures.customBorderColors == VK_TRUE &&
+                customBorderColorFeatures.customBorderColorWithoutFormat == VK_TRUE) {
+                if (!IsExtensionAlreadyEnabled(enabledDeviceExtensions,
+                                               VK_EXT_CUSTOM_BORDER_COLOR_EXTENSION_NAME)) {
+                    enabledDeviceExtensions.push_back(VK_EXT_CUSTOM_BORDER_COLOR_EXTENSION_NAME);
+                }
+                customBorderColorFeatures.pNext = const_cast<void*>(deviceCreateInfo.pNext);
+                deviceCreateInfo.pNext = &customBorderColorFeatures;
+                m_customBorderColorFeatureEnabled = true;
+
+                if (getPhysicalDeviceProperties2 != nullptr) {
+                    VkPhysicalDeviceCustomBorderColorPropertiesEXT customBorderColorProperties{};
+                    customBorderColorProperties.sType =
+                        VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_CUSTOM_BORDER_COLOR_PROPERTIES_EXT;
+                    VkPhysicalDeviceProperties2 propertyQuery{};
+                    propertyQuery.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PROPERTIES_2;
+                    propertyQuery.pNext = &customBorderColorProperties;
+                    getPhysicalDeviceProperties2(m_physicalDevice.handle, &propertyQuery);
+                    m_maxCustomBorderColorSamplers = customBorderColorProperties.maxCustomBorderColorSamplers;
+                }
+                MGLOG_I("Enabled optional device extension: %s (maxCustomBorderColorSamplers=%u)",
+                        VK_EXT_CUSTOM_BORDER_COLOR_EXTENSION_NAME, m_maxCustomBorderColorSamplers);
+            }
+        }
+        if (!m_customBorderColorFeatureEnabled) {
+            MGLOG_I("%s unavailable; GL_TEXTURE_BORDER_COLOR snaps to the nearest predefined VkBorderColor",
+                    VK_EXT_CUSTOM_BORDER_COLOR_EXTENSION_NAME);
+        }
+
         // Native subgroup topology, and VK_EXT_subgroup_size_control's
         // computeFullSubgroups feature. REQUIRE_FULL_SUBGROUPS on a compute stage is what
         // turns the derived gl_NumSubgroups (DeriveNumSubgroupsPass) from

MobileGL/MG_Backend/DirectVulkan/Renderer/VulkanRenderer.h

@@ -584,6 +584,22 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         // needs no feature). Both cached at device creation and drive a hard-fail-at-draw when absent.
         Bool m_dualSrcBlendFeatureEnabled = false;
         Bool m_primitiveTopologyListRestartFeatureEnabled = false;
+        // VK_EXT_custom_border_color. Vulkan's four predefined VkBorderColor values cover only
+        // transparent/opaque black and opaque white; GL_TEXTURE_BORDER_COLOR is an arbitrary vec4 (or
+        // an arbitrary ivec4/uvec4 through the "I" entry points). Without this extension a border
+        // colour outside the palette has to be snapped to the nearest predefined one. Both features
+        // are required together: customBorderColorWithoutFormat is what lets a sampler carry a custom
+        // colour without naming the image format it will be paired with, which GL's sampler objects
+        // cannot know. maxCustomBorderColorSamplers is a real device limit, so the sampler cache has
+        // to be able to fall back to the snapped value once it is reached.
+        Bool m_customBorderColorFeatureEnabled = false;
+        Uint32 m_maxCustomBorderColorSamplers = 0;
+        // sampleRateShading gates VkPipelineMultisampleStateCreateInfo::sampleShadingEnable, i.e.
+        // glEnable(GL_SAMPLE_SHADING) + glMinSampleShading. Unlike dualSrcBlend this does NOT
+        // hard-fail the draw when absent: sample shading is a rate hint, and every sample-rate
+        // pipeline is still correct (just not per-sample) at the default rate - so the enable is
+        // dropped and the draw proceeds, which is what a GL implementation with SAMPLES=1 does too.
+        Bool m_sampleRateShadingFeatureEnabled = false;
         // multiViewport gates rasterizing into more than one of ARB_viewport_array's 16 viewports
         // (gl_ViewportIndex). m_maxRasterizableViewports is min(MAX_VIEWPORTS, device limit), or 1
         // when the feature is off, and is the viewportCount a gl_ViewportIndex-writing pipeline
@@ -733,6 +749,12 @@ namespace MobileGL::MG_Backend::DirectVulkan {
             // values the memo already holds.
             Uint64 pipelineStateHash = 0;
             ProgramFactory::CompileOptionFlags transformFlags = {};
+            // Baked into the pipeline (PipelineFactory::ComputeHash mixes it), and NOT derivable
+            // from anything else in this key: it depends on whether the draw is indexed and on the
+            // index type, neither of which the mode/program/state hashes carry. Without it an
+            // indexed and a non-indexed draw over the same program and state collide on one entry
+            // and the second one gets the first one's restart setting.
+            Bool primitiveRestartEnable = false;
             VkPipeline pipeline = VK_NULL_HANDLE;
         };
         static constexpr Uint32 kPipelineMemoSize = 8;
@@ -865,6 +887,12 @@ namespace MobileGL::MG_Backend::DirectVulkan {
             Uint64 bindGeneration = 0;
             Uint32 baseTransformFlags = 0;
             Uint32 resolvedTransformFlags = 0;
+            // What ResolvePrimitiveRestartEnable answered for the draw this snapshot was taken
+            // from, i.e. what its pipeline's primitiveRestartEnable was built with. `aspects`
+            // already separates indexed from non-indexed draws, but not one index TYPE from
+            // another, and a restart index that fits GL_UNSIGNED_INT but not GL_UNSIGNED_SHORT
+            // makes those two draws want different pipelines.
+            Bool primitiveRestartEnable = false;
             Uint64 renderPassHash = 0;
             Uint32 imageIndex = 0;
             Uint64 textureEraseEpoch = 0;
@@ -1168,13 +1196,23 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         void CreateSwapchain();
         void CreateCommandPool();
 
+        // Whether THIS draw's primitive stream restarts, and therefore what
+        // VkPipelineInputAssemblyStateCreateInfo::primitiveRestartEnable must be. Resolved by the
+        // caller because it needs two facts a pipeline cannot see: whether the draw is indexed at
+        // all (GL primitive restart acts on the index stream, so it is a no-op for glDrawArrays),
+        // and the index TYPE (an application restart index that does not fit the type matches no
+        // index, so that draw restarts nowhere - see UploadAndBindIndexBuffer).
+        Bool ResolvePrimitiveRestartEnable(Flags<DrawSetupAspect> aspects,
+                                           const IndexBufferView* pIndexBufferView) const;
+
         VkPipeline GetOrCreatePipeline(
             GLenum mode,
             const MG_State::GLState::ProgramObject& program,
             const ProgramFactory::VkProgramObject& programObj,
             ProgramFactory::CompileOptionFlags transformFlags,
             const MG_State::GLState::VertexArrayObject& vao,
-            const RenderPassEntry& renderPassEntry);
+            const RenderPassEntry& renderPassEntry,
+            Bool primitiveRestartEnable);
         VkPipeline GetOrCreateComputePipeline(const ProgramFactory::VkProgramObject& programObj);
         void DestroyComputePipelines();
         // Takes the frame rather than a command buffer: a first-time storage-usage upgrade has to

MobileGL/MG_Impl/GLImpl/Debug/GL_Debug.cpp

@@ -0,0 +1,271 @@
+// MobileGL - MobileGL/MG_Impl/GLImpl/Debug/GL_Debug.cpp
+// Copyright (c) 2025-2026 MobileGL-Dev
+// Licensed under the GNU Lesser General Public License v3.0:
+//   https://www.gnu.org/licenses/gpl-3.0.txt
+//   https://www.gnu.org/licenses/lgpl-3.0.txt
+// SPDX-License-Identifier: LGPL-3.0-only
+// End of Source File Header
+
+#include "GL_Debug.h"
+
+#include <cstring>
+
+#include <MG_State/GLState/Core.h>
+#include <MG_State/GLState/ErrorState/Error.h>
+#include <MG_Impl/GLImpl/Query/GL_Query.h>
+#include <MG_Util/Converters/GLToStr/GLEnumConverter.h>
+
+namespace MobileGL::MG_Impl::GLImpl {
+    namespace {
+        // Must agree with what GL_Getter answers for GL_MAX_DEBUG_GROUP_STACK_DEPTH and
+        // GL_MAX_DEBUG_MESSAGE_LENGTH / GL_MAX_LABEL_LENGTH; an application that sizes a buffer
+        // off the query and then trips a different limit here would have no way to explain it.
+        constexpr SizeT kMaxDebugGroupStackDepth = 64;
+        constexpr GLsizei kMaxDebugMessageLength = 1024;
+        constexpr GLsizei kMaxLabelLength = 256;
+
+        // The debug state KHR_debug makes per-context. Held here rather than on GLContext because
+        // nothing else in MobileGL reads it, and it is keyed on the context id so a
+        // destroyed-and-recreated context starts with an empty stack and no labels - which the
+        // unit tests, which recreate the context between cases, depend on.
+        struct DebugState {
+            Uint64 contextId = 0;
+            // The messages pushed with glPushDebugGroup, innermost last. The base group GL creates
+            // the context with is implicit and is what makes the reported depth start at 1.
+            Vector<String> groupStack;
+            // Keyed by (identifier, name); see MakeObjectLabelKey.
+            UnorderedMap<Uint64, String> objectLabels;
+        };
+
+        DebugState& State() {
+            static DebugState state;
+            const Uint64 contextId = MG_State::pGLContext ? MG_State::pGLContext->GetTextureContextId() : 0;
+            if (state.contextId != contextId) {
+                state.contextId = contextId;
+                state.groupStack.clear();
+                state.objectLabels.clear();
+            }
+            return state;
+        }
+
+        Uint64 MakeObjectLabelKey(GLenum identifier, GLuint name) {
+            return (static_cast<Uint64>(identifier) << 32) | static_cast<Uint64>(name);
+        }
+
+        void RecordDebugError(ErrorCode code, const char* caller, const String& message) {
+            MG_State::pGLContext->RecordError(code, MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", caller, message));
+        }
+
+        // GL 4.6 core 20.2: only an APPLICATION or THIRD_PARTY source may be injected; the rest
+        // are reserved for the implementation itself.
+        Bool ValidateInjectedSource(GLenum source, const char* caller) {
+            if (source == GL_DEBUG_SOURCE_APPLICATION || source == GL_DEBUG_SOURCE_THIRD_PARTY) {
+                return true;
+            }
+            RecordDebugError(ErrorCode::InvalidEnum, caller,
+                             std::format("source {} is not GL_DEBUG_SOURCE_APPLICATION or "
+                                         "GL_DEBUG_SOURCE_THIRD_PARTY.",
+                                         MG_Util::ConvertGLEnumToString(source)));
+            return false;
+        }
+
+        // A negative length means the string is NUL-terminated (GL 4.6 core 20.2), which is how
+        // every one of these entry points spells "just use the whole thing".
+        Bool ValidateDebugStringLength(GLsizei length, const GLchar* text, GLsizei limit, const char* caller,
+                                       const char* what) {
+            const GLsizei effective =
+                length < 0 ? static_cast<GLsizei>(text != nullptr ? std::strlen(text) : 0) : length;
+            if (effective < limit) {
+                return true;
+            }
+            RecordDebugError(ErrorCode::InvalidValue, caller,
+                             std::format("{} length {} is not less than the {} limit of {}.", what, effective, what,
+                                         limit));
+            return false;
+        }
+
+        String MakeDebugString(GLsizei length, const GLchar* text) {
+            if (text == nullptr) return {};
+            return length < 0 ? String(text) : String(text, static_cast<SizeT>(length));
+        }
+
+        // Whether `name` currently names an object of `identifier`'s type. GL 4.6 core 20.5 makes
+        // labelling something that does not exist INVALID_VALUE, and every type KHR_debug lists
+        // has a frontend name check - so this is answered exactly rather than waved through.
+        // GL_DISPLAY_LIST is deliberately absent: it exists only in the compatibility profile,
+        // which MobileGL does not expose, so it falls to the INVALID_ENUM path below.
+        Bool ValidateLabelledObject(GLenum identifier, GLuint name, Bool& outIdentifierKnown) {
+            outIdentifierKnown = true;
+            auto* context = MG_State::pGLContext.get();
+            switch (identifier) {
+            case GL_BUFFER:
+                return context->ValidateBufferName(name);
+            case GL_SHADER:
+                return context->ValidateShaderName(name);
+            case GL_PROGRAM:
+                return context->ValidateProgramName(name);
+            case GL_VERTEX_ARRAY:
+                return context->ValidateVertexArrayName(name);
+            case GL_QUERY:
+                return IsQuery(name) == GL_TRUE;
+            case GL_PROGRAM_PIPELINE:
+                return context->ValidateProgramPipelineName(name);
+            case GL_TRANSFORM_FEEDBACK:
+                return context->ValidateTransformFeedbackName(name);
+            case GL_SAMPLER:
+                return context->ValidateSamplerName(name);
+            case GL_TEXTURE:
+                return context->ValidateTextureName(name);
+            case GL_RENDERBUFFER:
+                return context->ValidateRenderbufferName(name);
+            case GL_FRAMEBUFFER:
+                // Name 0 is the default framebuffer, which is a real, labellable object.
+                return name == 0 || context->ValidateFramebufferName(name);
+            default:
+                outIdentifierKnown = false;
+                return false;
+            }
+        }
+    } // namespace
+
+    GLint GetDebugGroupStackDepth() {
+        // GL 4.6 core 20.6: the context is created with one group already on the stack, so the
+        // reported depth is one more than the number of pushes the application has made.
+        return static_cast<GLint>(State().groupStack.size()) + 1;
+    }
+
+    void PushDebugGroup(GLenum source, GLuint id, GLsizei length, const GLchar* message) {
+        static_cast<void>(id);
+        if (!ValidateInjectedSource(source, __func__)) return;
+        if (!ValidateDebugStringLength(length, message, kMaxDebugMessageLength, __func__, "message")) return;
+
+        auto& state = State();
+        if (state.groupStack.size() + 1 >= kMaxDebugGroupStackDepth) {
+            // Not INVALID_*: KHR_debug gives the group stack its own error code.
+            RecordDebugError(ErrorCode::StackOverflow, __func__,
+                             std::format("the debug group stack is already {} deep, which is its maximum.",
+                                         kMaxDebugGroupStackDepth));
+            return;
+        }
+        state.groupStack.push_back(MakeDebugString(length, message));
+        MGLOG_D("glPushDebugGroup(%s) -> depth %d", state.groupStack.back().c_str(), GetDebugGroupStackDepth());
+    }
+
+    void PopDebugGroup() {
+        auto& state = State();
+        if (state.groupStack.empty()) {
+            // The base group the context was created with may not be popped (GL 4.6 core 20.6).
+            RecordDebugError(ErrorCode::StackUnderflow, __func__,
+                             "the debug group stack holds only the group the context was created with.");
+            return;
+        }
+        MGLOG_D("glPopDebugGroup(%s)", state.groupStack.back().c_str());
+        state.groupStack.pop_back();
+    }
+
+    void DebugMessageInsert(GLenum source, GLenum type, GLuint id, GLenum severity, GLsizei length,
+                            const GLchar* buf) {
+        static_cast<void>(id);
+        if (!ValidateInjectedSource(source, __func__)) return;
+        switch (type) {
+        case GL_DEBUG_TYPE_ERROR:
+        case GL_DEBUG_TYPE_DEPRECATED_BEHAVIOR:
+        case GL_DEBUG_TYPE_UNDEFINED_BEHAVIOR:
+        case GL_DEBUG_TYPE_PORTABILITY:
+        case GL_DEBUG_TYPE_PERFORMANCE:
+        case GL_DEBUG_TYPE_MARKER:
+        case GL_DEBUG_TYPE_PUSH_GROUP:
+        case GL_DEBUG_TYPE_POP_GROUP:
+        case GL_DEBUG_TYPE_OTHER:
+            break;
+        default:
+            RecordDebugError(ErrorCode::InvalidEnum, __func__,
+                             std::format("type {} is not a debug message type.",
+                                         MG_Util::ConvertGLEnumToString(type)));
+            return;
+        }
+        switch (severity) {
+        case GL_DEBUG_SEVERITY_HIGH:
+        case GL_DEBUG_SEVERITY_MEDIUM:
+        case GL_DEBUG_SEVERITY_LOW:
+        case GL_DEBUG_SEVERITY_NOTIFICATION:
+            break;
+        default:
+            RecordDebugError(ErrorCode::InvalidEnum, __func__,
+                             std::format("severity {} is not a debug message severity.",
+                                         MG_Util::ConvertGLEnumToString(severity)));
+            return;
+        }
+        if (!ValidateDebugStringLength(length, buf, kMaxDebugMessageLength, __func__, "message")) return;
+
+        // No callback is ever invoked and the message log is empty by construction
+        // (GL_MAX_DEBUG_LOGGED_MESSAGES is 1 and glGetDebugMessageLog returns nothing), so the
+        // application-visible effect is exactly the error checking above. The text still reaches
+        // MobileGL's own log, where it is worth having next to the calls it annotates - at debug
+        // level, so an application that inserts a message per draw costs nothing in a release build.
+        MGLOG_D("glDebugMessageInsert: %s", MakeDebugString(length, buf).c_str());
+    }
+
+    void ObjectLabel(GLenum identifier, GLuint name, GLsizei length, const GLchar* label) {
+        Bool identifierKnown = false;
+        const Bool objectExists = ValidateLabelledObject(identifier, name, identifierKnown);
+        if (!identifierKnown) {
+            RecordDebugError(ErrorCode::InvalidEnum, __func__,
+                             std::format("identifier {} is not a labellable object type.",
+                                         MG_Util::ConvertGLEnumToString(identifier)));
+            return;
+        }
+        if (!objectExists) {
+            RecordDebugError(ErrorCode::InvalidValue, __func__,
+                             std::format("{} {} is not the name of an existing object.",
+                                         MG_Util::ConvertGLEnumToString(identifier), name));
+            return;
+        }
+        if (!ValidateDebugStringLength(length, label, kMaxLabelLength, __func__, "label")) return;
+
+        auto& labels = State().objectLabels;
+        const Uint64 key = MakeObjectLabelKey(identifier, name);
+        if (label == nullptr) {
+            // GL 4.6 core 20.5: a NULL label removes any label the object had.
+            labels.erase(key);
+            return;
+        }
+        labels[key] = MakeDebugString(length, label);
+    }
+
+    void GetObjectLabel(GLenum identifier, GLuint name, GLsizei bufSize, GLsizei* length, GLchar* label) {
+        if (bufSize < 0) {
+            RecordDebugError(ErrorCode::InvalidValue, __func__, "bufSize must not be negative.");
+            return;
+        }
+        Bool identifierKnown = false;
+        const Bool objectExists = ValidateLabelledObject(identifier, name, identifierKnown);
+        if (!identifierKnown) {
+            RecordDebugError(ErrorCode::InvalidEnum, __func__,
+                             std::format("identifier {} is not a labellable object type.",
+                                         MG_Util::ConvertGLEnumToString(identifier)));
+            return;
+        }
+        if (!objectExists) {
+            RecordDebugError(ErrorCode::InvalidValue, __func__,
+                             std::format("{} {} is not the name of an existing object.",
+                                         MG_Util::ConvertGLEnumToString(identifier), name));
+            return;
+        }
+
+        const auto& labels = State().objectLabels;
+        const auto it = labels.find(MakeObjectLabelKey(identifier, name));
+        const String& text = it != labels.end() ? it->second : String{};
+        // GL 4.6 core 20.5: the returned length excludes the NUL, and an unlabelled object hands
+        // back an empty string with length 0 rather than an error.
+        SizeT copied = 0;
+        if (label != nullptr && bufSize > 0) {
+            copied = std::min(text.size(), static_cast<SizeT>(bufSize) - 1);
+            std::memcpy(label, text.data(), copied);
+            label[copied] = '\0';
+        }
+        if (length != nullptr) {
+            *length = static_cast<GLsizei>(copied);
+        }
+    }
+} // namespace MobileGL::MG_Impl::GLImpl

MobileGL/MG_Impl/GLImpl/Debug/GL_Debug.h

@@ -0,0 +1,42 @@
+// MobileGL - MobileGL/MG_Impl/GLImpl/Debug/GL_Debug.h
+// Copyright (c) 2025-2026 MobileGL-Dev
+// Licensed under the GNU Lesser General Public License v3.0:
+//   https://www.gnu.org/licenses/gpl-3.0.txt
+//   https://www.gnu.org/licenses/lgpl-3.0.txt
+// SPDX-License-Identifier: LGPL-3.0-only
+// End of Source File Header
+
+#pragma once
+#include <Includes.h>
+
+namespace MobileGL::MG_Impl::GLImpl {
+    // KHR_debug, core since GL 4.3 (GL 4.6 core 20). Applications use these to annotate a capture
+    // and to name their objects; Better Clouds calls all four for exactly that.
+    //
+    // MobileGL implements the STATE and the ERRORS, and deliberately does not forward the calls to
+    // the host driver. Two independent reasons:
+    //
+    //  * glObjectLabel names a FRONTEND object. MobileGL's texture 5 is not the ES driver's
+    //    texture 5 (and under DirectVulkan it is not a driver object at all), so forwarding the
+    //    pair verbatim would label an unrelated object or a nonexistent one - worse than not
+    //    labelling.
+    //  * A debug GROUP is only meaningful if it brackets the commands the application issued
+    //    inside it. Neither backend emits its work at the moment the GL call arrives: DirectGLES
+    //    defers and reorders state sync and uploads around draws, and DirectVulkan is usually not
+    //    even recording a command buffer here. A forwarded push/pop would therefore enclose the
+    //    wrong commands, which is a misleading capture rather than a helpful one.
+    //
+    // What the application can rely on is the observable contract: the group stack depth is real
+    // (GL_DEBUG_GROUP_STACK_DEPTH tracks it, and over/underflow raise the errors KHR_debug
+    // specifies), and a label written with glObjectLabel comes back from glGetObjectLabel.
+    void PushDebugGroup(GLenum source, GLuint id, GLsizei length, const GLchar* message);
+    void PopDebugGroup();
+    void DebugMessageInsert(GLenum source, GLenum type, GLuint id, GLenum severity, GLsizei length,
+                            const GLchar* buf);
+    void ObjectLabel(GLenum identifier, GLuint name, GLsizei length, const GLchar* label);
+    void GetObjectLabel(GLenum identifier, GLuint name, GLsizei bufSize, GLsizei* length, GLchar* label);
+
+    // Current depth of the debug group stack, for GL_DEBUG_GROUP_STACK_DEPTH. The base group the
+    // context is created with counts, so this is never below 1 (GL 4.6 core 20.6).
+    GLint GetDebugGroupStackDepth();
+} // namespace MobileGL::MG_Impl::GLImpl

MobileGL/MG_Impl/GLImpl/Drawing/GL_Drawing.cpp

@@ -34,8 +34,81 @@ namespace MobileGL::MG_Impl::GLImpl {
         return true;
     }
 
-    static Bool ValidateCurrentProgramForExecution(const char* functionName) {
-        return ValidateProgramForExecution(MG_State::pGLContext->GetProgramForDraw(), functionName);
+    // Takes the ALREADY-RESOLVED draw program rather than looking it up: GLContext::GetProgramForDraw
+    // is not a plain getter (it settles the program's link and SPIR-V jobs so every version a
+    // backend samples during this draw describes the program it is drawing), so the draw funnel
+    // below resolves it exactly once and hands it to both users.
+    static Bool ValidateResolvedProgramForDraw(const SharedPtr<MG_State::GLState::ProgramObject>& currentProgram,
+                                               const char* functionName) {
+        // "If there is no current program object or bound program pipeline object, the results of
+        // a draw are UNDEFINED" - and undefined is not an error (GL 4.6 core 7.3, ES 3.1 7.3).
+        // The draw is dropped, silently, which is one of the shapes "undefined" is allowed to
+        // take; recording INVALID_OPERATION here is not, and es31cSeparateShaderObjsTests'
+        // StateInteraction reads exactly that error back after useProgram(0) + bindProgramPipeline(0).
+        // A DISPATCH is the opposite rule ("INVALID_OPERATION if there is no active program for
+        // the compute shader stage"), which is why this lives on the draw path and not in the
+        // shared ValidateProgramForExecution below.
+        if (!currentProgram) return false;
+        if (!ValidateProgramForExecution(currentProgram, functionName)) return false;
+
+        // GL 4.6 core 7.4.1, the pipeline validation rule every vertex-transferring command
+        // inherits: it is an INVALID_OPERATION when a tessellation control, tessellation
+        // evaluation or geometry stage has an executable but no program supplies an executable
+        // VERTEX shader. A non-separable program cannot reach this - the link rule forbids the
+        // shape - so in practice it catches a program pipeline assembled out of stage programs,
+        // which today draws happily and renders nothing.
+        //
+        // Asked of the EXECUTABLE, like the compute check below: for a pipeline the resolved
+        // program is the graphics composite, whose linked-shader snapshot is built out of exactly
+        // the pipeline's own graphics stage programs (GLContext::GetProgramForDraw), and the only
+        // stage compositing ever invents is a default FRAGMENT shader. A fragment-only pipeline is
+        // deliberately NOT rejected: the rule above names the three pre-rasterization stages, and
+        // nothing else here should start refusing draws GL accepts.
+        //
+        // On the DRAW path only, never in ValidateProgramForExecution itself, so a dispatch -
+        // which shares that helper and legitimately has no vertex stage - is untouched.
+        const Bool hasPreRasterizationStage = currentProgram->HasLinkedShaderStage(ShaderStage::Geometry) ||
+                                              currentProgram->HasLinkedShaderStage(ShaderStage::TessControl) ||
+                                              currentProgram->HasLinkedShaderStage(ShaderStage::TessEval);
+        if (hasPreRasterizationStage && !currentProgram->HasLinkedShaderStage(ShaderStage::Vertex)) {
+            MG_State::pGLContext->RecordError(
+                ErrorCode::InvalidOperation,
+                MakeUnique<GenericErrorInfo>(
+                    "MG_Impl/GLImpl", functionName,
+                    "The program in use runs a geometry or tessellation stage but has no vertex shader stage."));
+            return false;
+        }
+
+        return true;
+    }
+
+    // gl_NumSamples has no SPIR-V built-in, so the source pipeline lowers it onto a reserved
+    // default-block uniform (see InjectNumSamplesBuiltinShim). This is where that uniform is paid
+    // for: the value is a property of the DRAW FRAMEBUFFER, not of the program, so one program
+    // drawn into a 4x target and then into the default framebuffer must see 4 and then 1 - which
+    // rules out baking it at link time.
+    //
+    // Per draw rather than on framebuffer changes because the pair (program, framebuffer) is what
+    // decides the value and either half can move between draws. It costs a phase-A flag read for
+    // every program that has no shim, and a 4-byte compare for the ones that do: the write only
+    // bumps the UBO content version when the number actually changes, so a run of draws into one
+    // framebuffer re-uploads nothing.
+    static void PublishDrawFramebufferSampleCount(const SharedPtr<MG_State::GLState::ProgramObject>& program) {
+        if (!program || !program->UsesReservedNumSamples()) return;
+        // GL 4.6 core 15.2.2: gl_NumSamples is the number of samples in the framebuffer, or ONE
+        // when the target is not multisampled - where glGetIntegerv(GL_SAMPLES) answers zero.
+        program->WriteReservedNumSamples(static_cast<Int>(std::max<GLint>(ResolveDrawFramebufferSampleCount(), 1)));
+    }
+
+    // The one funnel every drawing command passes through. Order is load-bearing: validate first
+    // (a rejected draw must leave state alone), then publish the sample count - which reads the
+    // DRAW FRAMEBUFFER binding, so it has to run after the caller's framebuffer state is settled
+    // and before the backend consumes the program's UBO content version.
+    static Bool PrepareCurrentProgramForDraw(const char* functionName) {
+        const auto& currentProgram = MG_State::pGLContext->GetProgramForDraw();
+        if (!ValidateResolvedProgramForDraw(currentProgram, functionName)) return false;
+        PublishDrawFramebufferSampleCount(currentProgram);
+        return true;
     }
 
     // A dispatch resolves its program through the DISPATCH accessor: with a pipeline bound
@@ -382,6 +455,31 @@ namespace MobileGL::MG_Impl::GLImpl {
         }
     }
 
+    // GL 4.6 core 10.3.9: every DrawElements-family count is a sizei and "if count is negative, an
+    // INVALID_VALUE error is generated". The same sentence covers instancecount and the
+    // MultiDraw* drawcount, so one helper serves all of them; the parameter is named for the
+    // caller so the message says which argument the application actually got wrong.
+    static Bool ValidateNonNegativeDrawArgument(const char* functionName, const char* argumentName, GLsizei value) {
+        if (value >= 0) return true;
+        MG_State::pGLContext->RecordError(
+            ErrorCode::InvalidValue,
+            MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", functionName,
+                                         String(argumentName) + " must be non-negative."));
+        return false;
+    }
+
+    // GL 4.6 core 10.3.9 for DrawRangeElements*: "if end < start, an INVALID_VALUE error is
+    // generated". Both are uints, so a caller that passes -1 for start arrives here as
+    // 0xFFFFFFFF and is caught by the same comparison - which is exactly what
+    // KHR-GL4x.draw_elements_base_vertex_tests.invalid_count_argument checks.
+    static Bool ValidateDrawElementsRange(const char* functionName, GLuint start, GLuint end) {
+        if (end >= start) return true;
+        MG_State::pGLContext->RecordError(
+            ErrorCode::InvalidValue,
+            MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", functionName, "end must not be less than start."));
+        return false;
+    }
+
     // GL 4.6 core 10.9: inside a conditional block whose predicate did not pass, the drawing
     // commands, Clear, ClearBuffer* and the compute dispatches are DISCARDED. The gate sits on the
     // wrappers that ISSUE the backend call rather than at the top of each entry point, so that
@@ -688,7 +786,65 @@ namespace MobileGL::MG_Impl::GLImpl {
         }
     }
 
+    // GL 4.6 core 11.2.2. The default tessellation levels a program with an evaluation stage and
+    // NO control stage tessellates at; both backends have to synthesize that control stage
+    // themselves (ES 3.2 and Vulkan both require one), and they compile these numbers into it, so
+    // there is no backend entry point to forward to - ES has none at all. INVALID_ENUM on a bad
+    // pname is the only error the spec lists: any float values are accepted, negatives and NaN
+    // included, and it is the tessellator that clamps them.
+    //
+    // This used to be a stub, which is why the two synthesizers hardcoded 1.0.
+    void PatchParameterfv(GLenum pname, const GLfloat* values) {
+        if (pname != GL_PATCH_DEFAULT_OUTER_LEVEL && pname != GL_PATCH_DEFAULT_INNER_LEVEL) {
+            MG_State::pGLContext->RecordError(
+                ErrorCode::InvalidEnum,
+                MakeUnique<GenericErrorInfo>(
+                    "MG_Impl/GLImpl", __func__,
+                    "pname must be GL_PATCH_DEFAULT_OUTER_LEVEL or GL_PATCH_DEFAULT_INNER_LEVEL."));
+            return;
+        }
+        if (!values) {
+            MG_State::pGLContext->RecordError(
+                ErrorCode::InvalidValue,
+                MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", __func__, "values pointer cannot be null"));
+            return;
+        }
+        if (pname == GL_PATCH_DEFAULT_OUTER_LEVEL) {
+            MG_State::pGLContext->SetPatchDefaultOuterLevel(
+                FloatVec4(values[0], values[1], values[2], values[3]));
+        } else {
+            MG_State::pGLContext->SetPatchDefaultInnerLevel(FloatVec2(values[0], values[1]));
+        }
+    }
+
+    namespace {
+        // GL 4.6 core 7.11.2 (and ARB_shader_image_load_store, which introduced the call): the
+        // barrier bitfield is INVALID_VALUE unless every bit is one of the defined ones, with
+        // GL_ALL_BARRIER_BITS - which is 0xFFFFFFFF, not the union of the list - accepted whole.
+        // Forwarding an undefined bit to the host driver let a caller that had computed its mask
+        // wrongly (or reused an ES-only bit) get silence instead of the error the spec promises.
+        constexpr GLbitfield kAllDefinedBarrierBits =
+            GL_VERTEX_ATTRIB_ARRAY_BARRIER_BIT | GL_ELEMENT_ARRAY_BARRIER_BIT | GL_UNIFORM_BARRIER_BIT |
+            GL_TEXTURE_FETCH_BARRIER_BIT | GL_SHADER_IMAGE_ACCESS_BARRIER_BIT | GL_COMMAND_BARRIER_BIT |
+            GL_PIXEL_BUFFER_BARRIER_BIT | GL_TEXTURE_UPDATE_BARRIER_BIT | GL_BUFFER_UPDATE_BARRIER_BIT |
+            GL_FRAMEBUFFER_BARRIER_BIT | GL_TRANSFORM_FEEDBACK_BARRIER_BIT | GL_ATOMIC_COUNTER_BARRIER_BIT |
+            GL_SHADER_STORAGE_BARRIER_BIT | GL_CLIENT_MAPPED_BUFFER_BARRIER_BIT | GL_QUERY_BUFFER_BARRIER_BIT;
+
+        Bool ValidateMemoryBarrierBits(const char* function, GLbitfield barriers) {
+            if (barriers == GL_ALL_BARRIER_BITS) return true;
+            if ((barriers & ~kAllDefinedBarrierBits) != 0) {
+                MG_State::pGLContext->RecordError(
+                    ErrorCode::InvalidValue,
+                    MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", function,
+                                                 "barriers contains bits that are not defined barrier bits."));
+                return false;
+            }
+            return true;
+        }
+    } // namespace
+
     void MemoryBarrier(GLbitfield barriers) {
+        if (!ValidateMemoryBarrierBits(__func__, barriers)) return;
         auto memoryBarrier = MG_Backend::gBackendFunctionsTable.GL.MemoryBarrier;
         if (!memoryBarrier) {
             MG_State::pGLContext->RecordError(
@@ -699,7 +855,29 @@ namespace MobileGL::MG_Impl::GLImpl {
         memoryBarrier(barriers);
     }
 
+    void TextureBarrier() {
+        // GL 4.5 core 8.26 / GL_ARB_texture_barrier: order every write the fixed-function
+        // framebuffer has already issued ahead of every subsequent texture fetch, so a shader may
+        // read texels of a texture that is also attached to the current framebuffer.
+        //
+        // Both backends serve this through their existing memory-barrier hook rather than a new
+        // entry point of their own: GL_FRAMEBUFFER_BARRIER_BIT is the source half (framebuffer
+        // writes) and GL_TEXTURE_FETCH_BARRIER_BIT the destination half (texture fetches), which
+        // is exactly the dependency ARB_texture_barrier defines - just expressed with the wider
+        // scope glMemoryBarrier gives it. That is a superset of the required ordering, never a
+        // subset, so it cannot under-synchronize.
+        auto memoryBarrier = MG_Backend::gBackendFunctionsTable.GL.MemoryBarrier;
+        if (!memoryBarrier) {
+            MG_State::pGLContext->RecordError(
+                ErrorCode::InvalidOperation,
+                MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", __func__, "Backend does not support memory barriers."));
+            return;
+        }
+        memoryBarrier(GL_TEXTURE_FETCH_BARRIER_BIT | GL_FRAMEBUFFER_BARRIER_BIT);
+    }
+
     void MemoryBarrierByRegion(GLbitfield barriers) {
+        if (!ValidateMemoryBarrierBits(__func__, barriers)) return;
         auto memoryBarrierByRegion = MG_Backend::gBackendFunctionsTable.GL.MemoryBarrierByRegion;
         if (!memoryBarrierByRegion) {
             MG_State::pGLContext->RecordError(
@@ -713,14 +891,14 @@ namespace MobileGL::MG_Impl::GLImpl {
 
     void MultiDrawElementsIndirect(GLenum mode, GLenum type, const void* indirect, GLsizei drawcount, GLsizei stride) {
         if (!ValidatePrimitiveModeEnum(__func__, mode)) return;
-        if (!ValidateCurrentProgramForExecution(__func__)) return;
+        if (!PrepareCurrentProgramForDraw(__func__)) return;
         if (!ValidatePrimitiveModeForBackend(__func__, mode)) return;
         MultiDrawElementsIndirect_Backend(mode, type, indirect, drawcount, stride);
     }
 
     void MultiDrawArraysIndirect(GLenum mode, const void* indirect, GLsizei drawcount, GLsizei stride) {
         if (!ValidatePrimitiveModeEnum(__func__, mode)) return;
-        if (!ValidateCurrentProgramForExecution(__func__)) return;
+        if (!PrepareCurrentProgramForDraw(__func__)) return;
         if (!ValidatePrimitiveModeForBackend(__func__, mode)) return;
         MultiDrawArraysIndirect_Backend(mode, indirect, drawcount, stride);
     }
@@ -798,7 +976,7 @@ namespace MobileGL::MG_Impl::GLImpl {
         // NegativeApiErrorsTest.IndirectParameterDrawsCheckBothBuffers pins the INVALID_VALUE
         // they produce for a call made with no program bound. Same precedence decision, and
         // the same reason, as DispatchComputeIndirect above.
-        if (!ValidateCurrentProgramForExecution(__func__)) return;
+        if (!PrepareCurrentProgramForDraw(__func__)) return;
         auto multiDrawElementsIndirectCount = MG_Backend::gBackendFunctionsTable.GL.MultiDrawElementsIndirectCount;
         if (!multiDrawElementsIndirectCount) {
             MG_State::pGLContext->RecordError(
@@ -819,7 +997,7 @@ namespace MobileGL::MG_Impl::GLImpl {
             return;
         }
         // See MultiDrawElementsIndirectCount, including why this one goes last.
-        if (!ValidateCurrentProgramForExecution(__func__)) return;
+        if (!PrepareCurrentProgramForDraw(__func__)) return;
         auto multiDrawArraysIndirectCount = MG_Backend::gBackendFunctionsTable.GL.MultiDrawArraysIndirectCount;
         if (!multiDrawArraysIndirectCount) {
             MG_State::pGLContext->RecordError(
@@ -834,14 +1012,17 @@ namespace MobileGL::MG_Impl::GLImpl {
     void DrawRangeElementsBaseVertex(GLenum mode, GLuint start, GLuint end, GLsizei count, GLenum type,
                                      const void* indices, GLint basevertex) {
         if (!ValidatePrimitiveModeEnum(__func__, mode)) return;
-        if (!ValidateCurrentProgramForExecution(__func__)) return;
+        if (!PrepareCurrentProgramForDraw(__func__)) return;
         if (!ValidatePrimitiveModeForBackend(__func__, mode)) return;
+        if (!ValidateDrawElementsIndexType(__func__, type)) return;
+        if (!ValidateNonNegativeDrawArgument(__func__, "count", count)) return;
+        if (!ValidateDrawElementsRange(__func__, start, end)) return;
         DrawRangeElementsBaseVertex_Backend(mode, start, end, count, type, indices, basevertex);
     }
 
     void DrawRangeElements(GLenum mode, GLuint start, GLuint end, GLsizei count, GLenum type, const void* indices) {
         if (!ValidatePrimitiveModeEnum(__func__, mode)) return;
-        if (!ValidateCurrentProgramForExecution(__func__)) return;
+        if (!PrepareCurrentProgramForDraw(__func__)) return;
         if (!ValidatePrimitiveModeForBackend(__func__, mode)) return;
         DrawRangeElements_Backend(mode, start, end, count, type, indices);
     }
@@ -849,7 +1030,7 @@ namespace MobileGL::MG_Impl::GLImpl {
     void DrawElementsInstancedBaseVertexBaseInstance(GLenum mode, GLsizei count, GLenum type, const void* indices,
                                                      GLsizei instancecount, GLint basevertex, GLuint baseinstance) {
         if (!ValidatePrimitiveModeEnum(__func__, mode)) return;
-        if (!ValidateCurrentProgramForExecution(__func__)) return;
+        if (!PrepareCurrentProgramForDraw(__func__)) return;
         if (!ValidatePrimitiveModeForBackend(__func__, mode)) return;
         DrawElementsInstancedBaseVertexBaseInstance_Backend(mode, count, type, indices, instancecount, basevertex,
                                                             baseinstance);
@@ -858,29 +1039,32 @@ namespace MobileGL::MG_Impl::GLImpl {
     void DrawElementsInstancedBaseVertex(GLenum mode, GLsizei count, GLenum type, const void* indices,
                                          GLsizei instancecount, GLint basevertex) {
         if (!ValidatePrimitiveModeEnum(__func__, mode)) return;
-        if (!ValidateCurrentProgramForExecution(__func__)) return;
+        if (!PrepareCurrentProgramForDraw(__func__)) return;
         if (!ValidatePrimitiveModeForBackend(__func__, mode)) return;
+        if (!ValidateDrawElementsIndexType(__func__, type)) return;
+        if (!ValidateNonNegativeDrawArgument(__func__, "count", count)) return;
+        if (!ValidateNonNegativeDrawArgument(__func__, "instancecount", instancecount)) return;
         DrawElementsInstancedBaseVertex_Backend(mode, count, type, indices, instancecount, basevertex);
     }
 
     void DrawElementsInstancedBaseInstance(GLenum mode, GLsizei count, GLenum type, const void* indices,
                                            GLsizei instancecount, GLuint baseinstance) {
         if (!ValidatePrimitiveModeEnum(__func__, mode)) return;
-        if (!ValidateCurrentProgramForExecution(__func__)) return;
+        if (!PrepareCurrentProgramForDraw(__func__)) return;
         if (!ValidatePrimitiveModeForBackend(__func__, mode)) return;
         DrawElementsInstancedBaseInstance_Backend(mode, count, type, indices, instancecount, baseinstance);
     }
 
     void DrawElementsInstanced(GLenum mode, GLsizei count, GLenum type, const void* indices, GLsizei instancecount) {
         if (!ValidatePrimitiveModeEnum(__func__, mode)) return;
-        if (!ValidateCurrentProgramForExecution(__func__)) return;
+        if (!PrepareCurrentProgramForDraw(__func__)) return;
         if (!ValidatePrimitiveModeForBackend(__func__, mode)) return;
         DrawElementsInstanced_Backend(mode, count, type, indices, instancecount);
     }
 
     void DrawElementsIndirect(GLenum mode, GLenum type, const void* indirect) {
         if (!ValidatePrimitiveModeEnum(__func__, mode)) return;
-        if (!ValidateCurrentProgramForExecution(__func__)) return;
+        if (!PrepareCurrentProgramForDraw(__func__)) return;
         if (!ValidatePrimitiveModeForBackend(__func__, mode)) return;
         if (!ValidateDrawElementsIndexType(__func__, type)) return;
         if (!ValidateIndirectDrawSource(__func__, indirect, kDrawElementsIndirectCommandBytes)) return;
@@ -890,21 +1074,21 @@ namespace MobileGL::MG_Impl::GLImpl {
     void DrawArraysInstancedBaseInstance(GLenum mode, GLint first, GLsizei count, GLsizei instancecount,
                                          GLuint baseinstance) {
         if (!ValidatePrimitiveModeEnum(__func__, mode)) return;
-        if (!ValidateCurrentProgramForExecution(__func__)) return;
+        if (!PrepareCurrentProgramForDraw(__func__)) return;
         if (!ValidatePrimitiveModeForBackend(__func__, mode)) return;
         DrawArraysInstancedBaseInstance_Backend(mode, first, count, instancecount, baseinstance);
     }
 
     void DrawArraysInstanced(GLenum mode, GLint first, GLsizei count, GLsizei instancecount) {
         if (!ValidatePrimitiveModeEnum(__func__, mode)) return;
-        if (!ValidateCurrentProgramForExecution(__func__)) return;
+        if (!PrepareCurrentProgramForDraw(__func__)) return;
         if (!ValidatePrimitiveModeForBackend(__func__, mode)) return;
         DrawArraysInstanced_Backend(mode, first, count, instancecount);
     }
 
     void DrawArraysIndirect(GLenum mode, const void* indirect) {
         if (!ValidatePrimitiveModeEnum(__func__, mode)) return;
-        if (!ValidateCurrentProgramForExecution(__func__)) return;
+        if (!PrepareCurrentProgramForDraw(__func__)) return;
         if (!ValidatePrimitiveModeForBackend(__func__, mode)) return;
         if (!ValidateIndirectDrawSource(__func__, indirect, kDrawArraysIndirectCommandBytes)) return;
         DrawArraysIndirect_Backend(mode, indirect);
@@ -912,15 +1096,17 @@ namespace MobileGL::MG_Impl::GLImpl {
 
     void DrawElementsBaseVertex(GLenum mode, GLsizei count, GLenum type, const void* indices, GLint basevertex) {
         if (!ValidatePrimitiveModeEnum(__func__, mode)) return;
-        if (!ValidateCurrentProgramForExecution(__func__)) return;
+        if (!PrepareCurrentProgramForDraw(__func__)) return;
         if (!ValidatePrimitiveModeForBackend(__func__, mode)) return;
+        if (!ValidateDrawElementsIndexType(__func__, type)) return;
+        if (!ValidateNonNegativeDrawArgument(__func__, "count", count)) return;
         AccountTransformFeedbackPrimitives(mode, count);
         DrawElementsBaseVertex_Backend(mode, count, type, indices, basevertex);
     }
 
     void DrawArrays(GLenum mode, GLint first, GLsizei count) {
         if (!ValidatePrimitiveModeEnum(__func__, mode)) return;
-        if (!ValidateCurrentProgramForExecution(__func__)) return;
+        if (!PrepareCurrentProgramForDraw(__func__)) return;
         if (!ValidatePrimitiveModeForBackend(__func__, mode)) return;
         AccountTransformFeedbackPrimitives(mode, count);
         DrawArrays_Backend(mode, first, count);
@@ -928,7 +1114,7 @@ namespace MobileGL::MG_Impl::GLImpl {
 
     void MultiDrawArrays(GLenum mode, const GLint* first, const GLsizei* count, GLsizei drawcount) {
         if (!ValidatePrimitiveModeEnum(__func__, mode)) return;
-        if (!ValidateCurrentProgramForExecution(__func__)) return;
+        if (!PrepareCurrentProgramForDraw(__func__)) return;
         if (!ValidatePrimitiveModeForBackend(__func__, mode)) return;
         if (drawcount < 0) {
             MG_State::pGLContext->RecordError(
@@ -942,7 +1128,7 @@ namespace MobileGL::MG_Impl::GLImpl {
     void MultiDrawElements(GLenum mode, const GLsizei* count, GLenum type, const void* const* indices,
                            GLsizei drawcount) {
         if (!ValidatePrimitiveModeEnum(__func__, mode)) return;
-        if (!ValidateCurrentProgramForExecution(__func__)) return;
+        if (!PrepareCurrentProgramForDraw(__func__)) return;
         if (!ValidatePrimitiveModeForBackend(__func__, mode)) return;
         MultiDrawElements_Backend(mode, count, type, indices, drawcount);
     }
@@ -950,8 +1136,21 @@ namespace MobileGL::MG_Impl::GLImpl {
     void MultiDrawElementsBaseVertex(GLenum mode, const GLsizei* count, GLenum type, const void* const* indices,
                                      GLsizei drawcount, const GLint* basevertex) {
         if (!ValidatePrimitiveModeEnum(__func__, mode)) return;
-        if (!ValidateCurrentProgramForExecution(__func__)) return;
+        if (!PrepareCurrentProgramForDraw(__func__)) return;
         if (!ValidatePrimitiveModeForBackend(__func__, mode)) return;
+        if (!ValidateDrawElementsIndexType(__func__, type)) return;
+        if (!ValidateNonNegativeDrawArgument(__func__, "drawcount", drawcount)) return;
+        // GL 4.6 core 10.5 defines MultiDrawElementsBaseVertex as drawcount separate
+        // DrawElementsBaseVertex calls, so each element of the count array carries the same
+        // non-negative requirement the single-draw entry point applies to its own count. The
+        // whole call is rejected before any sub-draw is issued, which is what makes the error
+        // observable at all - a driver that drew the valid prefix first would leave the
+        // framebuffer half-written.
+        if (count != nullptr) {
+            for (GLsizei draw = 0; draw < drawcount; ++draw) {
+                if (!ValidateNonNegativeDrawArgument(__func__, "every element of count", count[draw])) return;
+            }
+        }
         MultiDrawElementsBaseVertex_Backend(mode, count, type, indices, drawcount, basevertex);
     }
 
@@ -961,7 +1160,7 @@ namespace MobileGL::MG_Impl::GLImpl {
 
     void DrawElements(GLenum mode, GLsizei count, GLenum type, const void* indices) {
         if (!ValidatePrimitiveModeEnum(__func__, mode)) return;
-        if (!ValidateCurrentProgramForExecution(__func__)) return;
+        if (!PrepareCurrentProgramForDraw(__func__)) return;
         if (!ValidatePrimitiveModeForBackend(__func__, mode)) return;
         AccountTransformFeedbackPrimitives(mode, count);
         DrawElements_Backend(mode, count, type, indices);
@@ -1385,7 +1584,7 @@ namespace MobileGL::MG_Impl::GLImpl {
     // (GL 4.6 core 10.3.7).
     static void DrawTransformFeedbackImpl(const char* functionName, GLenum mode, GLuint id, GLuint stream,
                                           GLsizei instancecount) {
-        if (!ValidateCurrentProgramForExecution(functionName)) return;
+        if (!PrepareCurrentProgramForDraw(functionName)) return;
         if (!ValidatePrimitiveModeForBackend(functionName, mode)) return;
         if (instancecount < 0) {
             MG_State::pGLContext->RecordError(
@@ -1408,8 +1607,13 @@ namespace MobileGL::MG_Impl::GLImpl {
                                              std::to_string(id) + " is not a transform feedback object name."));
             return;
         }
-        // GL_MAX_VERTEX_STREAMS is 1, so stream 0 is the only one that exists.
-        if (stream != 0) {
+        // GL 4.6 core 10.3.7 bounds `stream` by GL_MAX_VERTEX_STREAMS, which this implementation
+        // answers as 1 - so stream 0 is the only one that exists and anything else is
+        // INVALID_VALUE. Read from the getter rather than written as `stream != 0` so the two can
+        // never drift: if vertex-stream support ever lands, this bound moves with the limit.
+        GLint maxVertexStreams = 1;
+        GetIntegerv(GL_MAX_VERTEX_STREAMS, &maxVertexStreams);
+        if (stream >= static_cast<GLuint>(std::max(maxVertexStreams, 1))) {
             MG_State::pGLContext->RecordError(
                 ErrorCode::InvalidValue,
                 MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", functionName,
@@ -1427,6 +1631,7 @@ namespace MobileGL::MG_Impl::GLImpl {
             return;
         }
 
+        // `stream` is provably 0 here (the bound above is 1), so this is stream 0's record.
         const Uint64 vertices = MG_State::pGLContext->GetTransformFeedbackRecordedVertices(id);
         if (vertices == 0) return;
         const auto count = static_cast<GLsizei>(vertices);

MobileGL/MG_Impl/GLImpl/Drawing/GL_Drawing.h

@@ -32,8 +32,10 @@ namespace MobileGL::MG_Impl::GLImpl {
     void DispatchCompute(GLuint numGroupsX, GLuint numGroupsY, GLuint numGroupsZ);
     void DispatchComputeIndirect(GLintptr indirect);
     void PatchParameteri(GLenum pname, GLint value);
+    void PatchParameterfv(GLenum pname, const GLfloat* values);
     void MemoryBarrier(GLbitfield barriers);
     void MemoryBarrierByRegion(GLbitfield barriers);
+    void TextureBarrier();
     void MultiDrawElementsIndirect(GLenum mode, GLenum type, const void* indirect, GLsizei drawcount, GLsizei stride);
     void MultiDrawArraysIndirect(GLenum mode, const void* indirect, GLsizei drawcount, GLsizei stride);
     void MultiDrawElementsIndirectCount(GLenum mode, GLenum type, const void* indirect, GLintptr drawcount,

MobileGL/MG_Impl/GLImpl/Exporting/Definitions.cpp

@@ -20,6 +20,7 @@
 #include "../Framebuffer/GL_Framebuffer.h"
 #include "../VertexArray/GL_VertexArray.h"
 #include "../Sync/GL_Sync.h"
+#include "../Debug/GL_Debug.h"
 #include <MG_State/GLState/Core.h>
 
 #define DECLARE_GL_FUNCTION_STUB_HEAD(type, name, ...) MOBILEGL_GL_API type gl##name(__VA_ARGS__) {
@@ -159,7 +160,7 @@ DECLARE_GL_FUNCTION_STUB_HEAD(void, ReleaseShaderCompiler) DECLARE_GL_FUNCTION_S
 DECLARE_GL_FUNCTION_HEAD(void, RenderbufferStorage, GLenum target, GLenum internalformat, GLsizei width, GLsizei height) DECLARE_GL_FUNCTION_END_NO_RETURN(void, RenderbufferStorage, target, internalformat, width, height)
 DECLARE_GL_FUNCTION_HEAD(void, SampleCoverage, GLfloat value, GLboolean invert) DECLARE_GL_FUNCTION_END_NO_RETURN(void, SampleCoverage, value, invert)
 DECLARE_GL_FUNCTION_HEAD(void, Scissor, GLint x, GLint y, GLsizei width, GLsizei height) DECLARE_GL_FUNCTION_END_NO_RETURN(void, Scissor, x, y, width, height)
-DECLARE_GL_FUNCTION_STUB_HEAD(void, ShaderBinary, GLsizei count, const GLuint* shaders, GLenum binaryformat, const void* binary, GLsizei length) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, ShaderBinary, count, shaders, binaryformat, binary, length)
+DECLARE_GL_FUNCTION_HEAD(void, ShaderBinary, GLsizei count, const GLuint* shaders, GLenum binaryformat, const void* binary, GLsizei length) DECLARE_GL_FUNCTION_END_NO_RETURN(void, ShaderBinary, count, shaders, binaryformat, binary, length)
 DECLARE_GL_FUNCTION_HEAD(void, ShaderSource, GLuint shader, GLsizei count, const GLchar* const* string, const GLint* length) DECLARE_GL_FUNCTION_END_NO_RETURN(void, ShaderSource, shader, count, string, length)
 DECLARE_GL_FUNCTION_HEAD(void, StencilFunc, GLenum func, GLint ref, GLuint mask) DECLARE_GL_FUNCTION_END_NO_RETURN(void, StencilFunc, func, ref, mask)
 DECLARE_GL_FUNCTION_HEAD(void, StencilFuncSeparate, GLenum face, GLenum func, GLint ref, GLuint mask) DECLARE_GL_FUNCTION_END_NO_RETURN(void, StencilFuncSeparate, face, func, ref, mask)
@@ -378,27 +379,13 @@ DECLARE_GL_FUNCTION_HEAD(void, VertexBindingDivisor, GLuint bindingindex, GLuint
 DECLARE_GL_FUNCTION_STUB_HEAD(void, BlendBarrier) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, BlendBarrier)
 DECLARE_GL_FUNCTION_HEAD(void, CopyImageSubData, GLuint srcName, GLenum srcTarget, GLint srcLevel, GLint srcX, GLint srcY, GLint srcZ, GLuint dstName, GLenum dstTarget, GLint dstLevel, GLint dstX, GLint dstY, GLint dstZ, GLsizei srcWidth, GLsizei srcHeight, GLsizei srcDepth) DECLARE_GL_FUNCTION_END_NO_RETURN(void, CopyImageSubData, srcName, srcTarget, srcLevel, srcX, srcY, srcZ, dstName, dstTarget, dstLevel, dstX, dstY, dstZ, srcWidth, srcHeight, srcDepth)
 DECLARE_GL_FUNCTION_STUB_HEAD(void, DebugMessageControl, GLenum source, GLenum type, GLenum severity, GLsizei count, const GLuint* ids, GLboolean enabled) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, DebugMessageControl, source, type, severity, count, ids, enabled)
-DECLARE_GL_FUNCTION_STUB_HEAD(void, DebugMessageInsert, GLenum source, GLenum type, GLuint id, GLenum severity, GLsizei length, const GLchar* buf) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, DebugMessageInsert, source, type, id, severity, length, buf)
+DECLARE_GL_FUNCTION_HEAD(void, DebugMessageInsert, GLenum source, GLenum type, GLuint id, GLenum severity, GLsizei length, const GLchar* buf) DECLARE_GL_FUNCTION_END_NO_RETURN(void, DebugMessageInsert, source, type, id, severity, length, buf)
 DECLARE_GL_FUNCTION_STUB_HEAD(void, DebugMessageCallback, GLDEBUGPROC callback, const void* userParam) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, DebugMessageCallback, callback, userParam)
 DECLARE_GL_FUNCTION_STUB_HEAD(GLuint, GetDebugMessageLog, GLuint count, GLsizei bufSize, GLenum* sources, GLenum* types, GLuint* ids, GLenum* severities, GLsizei* lengths, GLchar* messageLog) DECLARE_GL_FUNCTION_STUB_END(GLuint, GetDebugMessageLog, count, bufSize, sources, types, ids, severities, lengths, messageLog)
-DECLARE_GL_FUNCTION_STUB_HEAD(void, PushDebugGroup, GLenum source, GLuint id, GLsizei length, const GLchar* message) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, PushDebugGroup, source, id, length, message)
-DECLARE_GL_FUNCTION_STUB_HEAD(void, PopDebugGroup) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, PopDebugGroup)
-MOBILEGL_GL_API void glObjectLabel(GLenum identifier, GLuint name, GLsizei length, const GLchar* label) {
-    (void)identifier;
-    (void)name;
-    (void)length;
-    (void)label;
-}
-MOBILEGL_GL_API void glGetObjectLabel(GLenum identifier, GLuint name, GLsizei bufSize, GLsizei* length, GLchar* label) {
-    (void)identifier;
-    (void)name;
-    if (length) {
-        *length = 0;
-    }
-    if (label && bufSize > 0) {
-        label[0] = '\0';
-    }
-}
+DECLARE_GL_FUNCTION_HEAD(void, PushDebugGroup, GLenum source, GLuint id, GLsizei length, const GLchar* message) DECLARE_GL_FUNCTION_END_NO_RETURN(void, PushDebugGroup, source, id, length, message)
+DECLARE_GL_FUNCTION_HEAD(void, PopDebugGroup) DECLARE_GL_FUNCTION_END_NO_RETURN(void, PopDebugGroup)
+DECLARE_GL_FUNCTION_HEAD(void, ObjectLabel, GLenum identifier, GLuint name, GLsizei length, const GLchar* label) DECLARE_GL_FUNCTION_END_NO_RETURN(void, ObjectLabel, identifier, name, length, label)
+DECLARE_GL_FUNCTION_HEAD(void, GetObjectLabel, GLenum identifier, GLuint name, GLsizei bufSize, GLsizei* length, GLchar* label) DECLARE_GL_FUNCTION_END_NO_RETURN(void, GetObjectLabel, identifier, name, bufSize, length, label)
 DECLARE_GL_FUNCTION_STUB_HEAD(void, ObjectPtrLabel, const void* ptr, GLsizei length, const GLchar* label) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, ObjectPtrLabel, ptr, length, label)
 DECLARE_GL_FUNCTION_STUB_HEAD(void, GetObjectPtrLabel, const void* ptr, GLsizei bufSize, GLsizei* length, GLchar* label) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, GetObjectPtrLabel, ptr, bufSize, length, label)
 DECLARE_GL_FUNCTION_STUB_HEAD(void, GetPointerv, GLenum pname, void** params) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, GetPointerv, pname, params)
@@ -424,7 +411,7 @@ DECLARE_GL_FUNCTION_HEAD(void, ReadnPixels, GLint x, GLint y, GLsizei width, GLs
 DECLARE_GL_FUNCTION_STUB_HEAD(void, GetnUniformfv, GLuint program, GLint location, GLsizei bufSize, GLfloat* params) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, GetnUniformfv, program, location, bufSize, params)
 DECLARE_GL_FUNCTION_STUB_HEAD(void, GetnUniformiv, GLuint program, GLint location, GLsizei bufSize, GLint* params) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, GetnUniformiv, program, location, bufSize, params)
 DECLARE_GL_FUNCTION_STUB_HEAD(void, GetnUniformuiv, GLuint program, GLint location, GLsizei bufSize, GLuint* params) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, GetnUniformuiv, program, location, bufSize, params)
-DECLARE_GL_FUNCTION_STUB_HEAD(void, MinSampleShading, GLfloat value) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, MinSampleShading, value)
+DECLARE_GL_FUNCTION_HEAD(void, MinSampleShading, GLfloat value) DECLARE_GL_FUNCTION_END_NO_RETURN(void, MinSampleShading, value)
 DECLARE_GL_FUNCTION_HEAD(void, PatchParameteri, GLenum pname, GLint value) DECLARE_GL_FUNCTION_END_NO_RETURN(void, PatchParameteri, pname, value)
 DECLARE_GL_FUNCTION_HEAD(void, TexParameterIiv, GLenum target, GLenum pname, const GLint* params) DECLARE_GL_FUNCTION_END_NO_RETURN(void, TexParameterIiv, target, pname, params)
 DECLARE_GL_FUNCTION_HEAD(void, TexParameterIuiv, GLenum target, GLenum pname, const GLuint* params) DECLARE_GL_FUNCTION_END_NO_RETURN(void, TexParameterIuiv, target, pname, params)
@@ -936,7 +923,7 @@ DECLARE_GL_FUNCTION_STUB_HEAD(void, GetActiveSubroutineName, GLuint program, GLe
 DECLARE_GL_FUNCTION_STUB_HEAD(void, UniformSubroutinesuiv, GLenum shadertype, GLsizei count, const GLuint* indices) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, UniformSubroutinesuiv, shadertype, count, indices)
 DECLARE_GL_FUNCTION_STUB_HEAD(void, GetUniformSubroutineuiv, GLenum shadertype, GLint location, GLuint* params) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, GetUniformSubroutineuiv, shadertype, location, params)
 DECLARE_GL_FUNCTION_STUB_HEAD(void, GetProgramStageiv, GLuint program, GLenum shadertype, GLenum pname, GLint* values) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, GetProgramStageiv, program, shadertype, pname, values)
-DECLARE_GL_FUNCTION_STUB_HEAD(void, PatchParameterfv, GLenum pname, const GLfloat* values) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, PatchParameterfv, pname, values)
+DECLARE_GL_FUNCTION_HEAD(void, PatchParameterfv, GLenum pname, const GLfloat* values) DECLARE_GL_FUNCTION_END_NO_RETURN(void, PatchParameterfv, pname, values)
 DECLARE_GL_FUNCTION_HEAD(void, DrawTransformFeedback, GLenum mode, GLuint id) DECLARE_GL_FUNCTION_END_NO_RETURN(void, DrawTransformFeedback, mode, id)
 DECLARE_GL_FUNCTION_HEAD(void, DrawTransformFeedbackStream, GLenum mode, GLuint id, GLuint stream) DECLARE_GL_FUNCTION_END_NO_RETURN(void, DrawTransformFeedbackStream, mode, id, stream)
 DECLARE_GL_FUNCTION_HEAD(void, BeginQueryIndexed, GLenum target, GLuint index, GLuint id) DECLARE_GL_FUNCTION_END_NO_RETURN(void, BeginQueryIndexed, target, index, id)
@@ -996,7 +983,7 @@ DECLARE_GL_FUNCTION_HEAD(void, MultiDrawArraysIndirect, GLenum mode, const void*
 DECLARE_GL_FUNCTION_HEAD(void, MultiDrawElementsIndirect, GLenum mode, GLenum type, const void* indirect, GLsizei drawcount, GLsizei stride) DECLARE_GL_FUNCTION_END_NO_RETURN(void, MultiDrawElementsIndirect, mode, type, indirect, drawcount, stride)
 DECLARE_GL_FUNCTION_HEAD(GLint, GetProgramResourceLocationIndex, GLuint program, GLenum programInterface, const GLchar* name) DECLARE_GL_FUNCTION_END(GLint, GetProgramResourceLocationIndex, program, programInterface, name)
 DECLARE_GL_FUNCTION_HEAD(void, ShaderStorageBlockBinding, GLuint program, GLuint storageBlockIndex, GLuint storageBlockBinding) DECLARE_GL_FUNCTION_END_NO_RETURN(void, ShaderStorageBlockBinding, program, storageBlockIndex, storageBlockBinding)
-DECLARE_GL_FUNCTION_STUB_HEAD(void, TextureView, GLuint texture, GLenum target, GLuint origtexture, GLenum internalformat, GLuint minlevel, GLuint numlevels, GLuint minlayer, GLuint numlayers) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, TextureView, texture, target, origtexture, internalformat, minlevel, numlevels, minlayer, numlayers)
+DECLARE_GL_FUNCTION_HEAD(void, TextureView, GLuint texture, GLenum target, GLuint origtexture, GLenum internalformat, GLuint minlevel, GLuint numlevels, GLuint minlayer, GLuint numlayers) DECLARE_GL_FUNCTION_END_NO_RETURN(void, TextureView, texture, target, origtexture, internalformat, minlevel, numlevels, minlayer, numlayers)
 DECLARE_GL_FUNCTION_HEAD(void, VertexAttribLFormat, GLuint attribindex, GLint size, GLenum type, GLuint relativeoffset) DECLARE_GL_FUNCTION_END_NO_RETURN(void, VertexAttribLFormat, attribindex, size, type, relativeoffset)
 DECLARE_GL_FUNCTION_HEAD(void, BufferStorage, GLenum target, GLsizeiptr size, const void* data, GLbitfield flags) DECLARE_GL_FUNCTION_END_NO_RETURN(void, BufferStorage, target, size, data, flags)
 DECLARE_GL_FUNCTION_HEAD(void, ClearTexImage, GLuint texture, GLint level, GLenum format, GLenum type, const void* data) DECLARE_GL_FUNCTION_END_NO_RETURN(void, ClearTexImage, texture, level, format, type, data)
@@ -1007,7 +994,7 @@ DECLARE_GL_FUNCTION_HEAD(void, BindTextures, GLuint first, GLsizei count, const
 DECLARE_GL_FUNCTION_HEAD(void, BindSamplers, GLuint first, GLsizei count, const GLuint* samplers) DECLARE_GL_FUNCTION_END_NO_RETURN(void, BindSamplers, first, count, samplers)
 DECLARE_GL_FUNCTION_HEAD(void, BindImageTextures, GLuint first, GLsizei count, const GLuint* textures) DECLARE_GL_FUNCTION_END_NO_RETURN(void, BindImageTextures, first, count, textures)
 DECLARE_GL_FUNCTION_HEAD(void, BindVertexBuffers, GLuint first, GLsizei count, const GLuint* buffers, const GLintptr* offsets, const GLsizei* strides) DECLARE_GL_FUNCTION_END_NO_RETURN(void, BindVertexBuffers, first, count, buffers, offsets, strides)
-DECLARE_GL_FUNCTION_STUB_HEAD(void, ClipControl, GLenum origin, GLenum depth) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, ClipControl, origin, depth)
+DECLARE_GL_FUNCTION_HEAD(void, ClipControl, GLenum origin, GLenum depth) DECLARE_GL_FUNCTION_END_NO_RETURN(void, ClipControl, origin, depth)
 DECLARE_GL_FUNCTION_HEAD(void, CreateTransformFeedbacks, GLsizei n, GLuint* ids) DECLARE_GL_FUNCTION_END_NO_RETURN(void, CreateTransformFeedbacks, n, ids)
 DECLARE_GL_FUNCTION_HEAD(void, TransformFeedbackBufferBase, GLuint xfb, GLuint index, GLuint buffer) DECLARE_GL_FUNCTION_END_NO_RETURN(void, TransformFeedbackBufferBase, xfb, index, buffer)
 DECLARE_GL_FUNCTION_HEAD(void, TransformFeedbackBufferRange, GLuint xfb, GLuint index, GLuint buffer, GLintptr offset, GLsizeiptr size) DECLARE_GL_FUNCTION_END_NO_RETURN(void, TransformFeedbackBufferRange, xfb, index, buffer, offset, size)
@@ -1060,9 +1047,9 @@ DECLARE_GL_FUNCTION_HEAD(void, TextureStorage3DMultisample, GLuint texture, GLsi
 DECLARE_GL_FUNCTION_HEAD(void, TextureSubImage1D, GLuint texture, GLint level, GLint xoffset, GLsizei width, GLenum format, GLenum type, const void* pixels) DECLARE_GL_FUNCTION_END_NO_RETURN(void, TextureSubImage1D, texture, level, xoffset, width, format, type, pixels)
 DECLARE_GL_FUNCTION_HEAD(void, TextureSubImage2D, GLuint texture, GLint level, GLint xoffset, GLint yoffset, GLsizei width, GLsizei height, GLenum format, GLenum type, const void* pixels) DECLARE_GL_FUNCTION_END_NO_RETURN(void, TextureSubImage2D, texture, level, xoffset, yoffset, width, height, format, type, pixels)
 DECLARE_GL_FUNCTION_HEAD(void, TextureSubImage3D, GLuint texture, GLint level, GLint xoffset, GLint yoffset, GLint zoffset, GLsizei width, GLsizei height, GLsizei depth, GLenum format, GLenum type, const void* pixels) DECLARE_GL_FUNCTION_END_NO_RETURN(void, TextureSubImage3D, texture, level, xoffset, yoffset, zoffset, width, height, depth, format, type, pixels)
-DECLARE_GL_FUNCTION_STUB_HEAD(void, CompressedTextureSubImage1D, GLuint texture, GLint level, GLint xoffset, GLsizei width, GLenum format, GLsizei imageSize, const void* data) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, CompressedTextureSubImage1D, texture, level, xoffset, width, format, imageSize, data)
+DECLARE_GL_FUNCTION_HEAD(void, CompressedTextureSubImage1D, GLuint texture, GLint level, GLint xoffset, GLsizei width, GLenum format, GLsizei imageSize, const void* data) DECLARE_GL_FUNCTION_END_NO_RETURN(void, CompressedTextureSubImage1D, texture, level, xoffset, width, format, imageSize, data)
 DECLARE_GL_FUNCTION_HEAD(void, CompressedTextureSubImage2D, GLuint texture, GLint level, GLint xoffset, GLint yoffset, GLsizei width, GLsizei height, GLenum format, GLsizei imageSize, const void* data) DECLARE_GL_FUNCTION_END_NO_RETURN(void, CompressedTextureSubImage2D, texture, level, xoffset, yoffset, width, height, format, imageSize, data)
-DECLARE_GL_FUNCTION_STUB_HEAD(void, CompressedTextureSubImage3D, GLuint texture, GLint level, GLint xoffset, GLint yoffset, GLint zoffset, GLsizei width, GLsizei height, GLsizei depth, GLenum format, GLsizei imageSize, const void* data) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, CompressedTextureSubImage3D, texture, level, xoffset, yoffset, zoffset, width, height, depth, format, imageSize, data)
+DECLARE_GL_FUNCTION_HEAD(void, CompressedTextureSubImage3D, GLuint texture, GLint level, GLint xoffset, GLint yoffset, GLint zoffset, GLsizei width, GLsizei height, GLsizei depth, GLenum format, GLsizei imageSize, const void* data) DECLARE_GL_FUNCTION_END_NO_RETURN(void, CompressedTextureSubImage3D, texture, level, xoffset, yoffset, zoffset, width, height, depth, format, imageSize, data)
 DECLARE_GL_FUNCTION_HEAD(void, CopyTextureSubImage1D, GLuint texture, GLint level, GLint xoffset, GLint x, GLint y, GLsizei width) DECLARE_GL_FUNCTION_END_NO_RETURN(void, CopyTextureSubImage1D, texture, level, xoffset, x, y, width)
 DECLARE_GL_FUNCTION_HEAD(void, CopyTextureSubImage2D, GLuint texture, GLint level, GLint xoffset, GLint yoffset, GLint x, GLint y, GLsizei width, GLsizei height) DECLARE_GL_FUNCTION_END_NO_RETURN(void, CopyTextureSubImage2D, texture, level, xoffset, yoffset, x, y, width, height)
 DECLARE_GL_FUNCTION_HEAD(void, CopyTextureSubImage3D, GLuint texture, GLint level, GLint xoffset, GLint yoffset, GLint zoffset, GLint x, GLint y, GLsizei width, GLsizei height) DECLARE_GL_FUNCTION_END_NO_RETURN(void, CopyTextureSubImage3D, texture, level, xoffset, yoffset, zoffset, x, y, width, height)
@@ -1120,11 +1107,11 @@ DECLARE_GL_FUNCTION_STUB_HEAD(void, GetnConvolutionFilter, GLenum target, GLenum
 DECLARE_GL_FUNCTION_STUB_HEAD(void, GetnSeparableFilter, GLenum target, GLenum format, GLenum type, GLsizei rowBufSize, void* row, GLsizei columnBufSize, void* column, void* span) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, GetnSeparableFilter, target, format, type, rowBufSize, row, columnBufSize, column, span)
 DECLARE_GL_FUNCTION_STUB_HEAD(void, GetnHistogram, GLenum target, GLboolean reset, GLenum format, GLenum type, GLsizei bufSize, void* values) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, GetnHistogram, target, reset, format, type, bufSize, values)
 DECLARE_GL_FUNCTION_STUB_HEAD(void, GetnMinmax, GLenum target, GLboolean reset, GLenum format, GLenum type, GLsizei bufSize, void* values) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, GetnMinmax, target, reset, format, type, bufSize, values)
-DECLARE_GL_FUNCTION_STUB_HEAD(void, TextureBarrier, void) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, TextureBarrier, )
-DECLARE_GL_FUNCTION_STUB_HEAD(void, SpecializeShader, GLuint shader, const GLchar* pEntryPoint, GLuint numSpecializationConstants, const GLuint* pConstantIndex, const GLuint* pConstantValue) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, SpecializeShader, shader, pEntryPoint, numSpecializationConstants, pConstantIndex, pConstantValue)
+DECLARE_GL_FUNCTION_HEAD(void, TextureBarrier, void) DECLARE_GL_FUNCTION_END_NO_RETURN(void, TextureBarrier, )
+DECLARE_GL_FUNCTION_HEAD(void, SpecializeShader, GLuint shader, const GLchar* pEntryPoint, GLuint numSpecializationConstants, const GLuint* pConstantIndex, const GLuint* pConstantValue) DECLARE_GL_FUNCTION_END_NO_RETURN(void, SpecializeShader, shader, pEntryPoint, numSpecializationConstants, pConstantIndex, pConstantValue)
 DECLARE_GL_FUNCTION_HEAD(void, MultiDrawArraysIndirectCount, GLenum mode, const void* indirect, GLintptr drawcount, GLsizei maxdrawcount, GLsizei stride) DECLARE_GL_FUNCTION_END_NO_RETURN(void, MultiDrawArraysIndirectCount, mode, indirect, drawcount, maxdrawcount, stride)
 DECLARE_GL_FUNCTION_HEAD(void, MultiDrawElementsIndirectCount, GLenum mode, GLenum type, const void* indirect, GLintptr drawcount, GLsizei maxdrawcount, GLsizei stride) DECLARE_GL_FUNCTION_END_NO_RETURN(void, MultiDrawElementsIndirectCount, mode, type, indirect, drawcount, maxdrawcount, stride)
-DECLARE_GL_FUNCTION_STUB_HEAD(void, PolygonOffsetClamp, GLfloat factor, GLfloat units, GLfloat clamp) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, PolygonOffsetClamp, factor, units, clamp)
+DECLARE_GL_FUNCTION_HEAD(void, PolygonOffsetClamp, GLfloat factor, GLfloat units, GLfloat clamp) DECLARE_GL_FUNCTION_END_NO_RETURN(void, PolygonOffsetClamp, factor, units, clamp)
 DECLARE_GL_FUNCTION_STUB_HEAD(void, PrimitiveBoundingBoxARB, GLfloat minX, GLfloat minY, GLfloat minZ, GLfloat minW, GLfloat maxX, GLfloat maxY, GLfloat maxZ, GLfloat maxW) DECLARE_GL_FUNCTION_STUB_END(void, PrimitiveBoundingBoxARB, minX, minY, minZ, minW, maxX, maxY, maxZ, maxW)
 DECLARE_GL_FUNCTION_STUB_HEAD(GLuint64, GetTextureHandleARB, GLuint texture) DECLARE_GL_FUNCTION_STUB_END(GLuint64, GetTextureHandleARB, texture)
 DECLARE_GL_FUNCTION_STUB_HEAD(GLuint64, GetTextureSamplerHandleARB, GLuint texture, GLuint sampler) DECLARE_GL_FUNCTION_STUB_END(GLuint64, GetTextureSamplerHandleARB, texture, sampler)
@@ -1163,7 +1150,7 @@ DECLARE_GL_FUNCTION_STUB_HEAD(void, GetProgramLocalParameterdvARB, GLenum target
 DECLARE_GL_FUNCTION_STUB_HEAD(void, GetProgramLocalParameterfvARB, GLenum target, GLuint index, GLfloat* params) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, GetProgramLocalParameterfvARB, target, index, params)
 DECLARE_GL_FUNCTION_STUB_HEAD(void, GetProgramStringARB, GLenum target, GLenum pname, void* string) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, GetProgramStringARB, target, pname, string)
 DECLARE_GL_FUNCTION_STUB_HEAD(void, FramebufferTextureFaceARB, GLenum target, GLenum attachment, GLuint texture, GLint level, GLenum face) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, FramebufferTextureFaceARB, target, attachment, texture, level, face)
-DECLARE_GL_FUNCTION_STUB_HEAD(void, SpecializeShaderARB, GLuint shader, const GLchar* pEntryPoint, GLuint numSpecializationConstants, const GLuint* pConstantIndex, const GLuint* pConstantValue) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, SpecializeShaderARB, shader, pEntryPoint, numSpecializationConstants, pConstantIndex, pConstantValue)
+DECLARE_GL_FUNCTION_HEAD(void, SpecializeShaderARB, GLuint shader, const GLchar* pEntryPoint, GLuint numSpecializationConstants, const GLuint* pConstantIndex, const GLuint* pConstantValue) DECLARE_GL_FUNCTION_END_NO_RETURN(void, SpecializeShader, shader, pEntryPoint, numSpecializationConstants, pConstantIndex, pConstantValue)
 DECLARE_GL_FUNCTION_STUB_HEAD(void, Uniform1i64ARB, GLint location, GLint64 x) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, Uniform1i64ARB, location, x)
 DECLARE_GL_FUNCTION_STUB_HEAD(void, Uniform2i64ARB, GLint location, GLint64 x, GLint64 y) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, Uniform2i64ARB, location, x, y)
 DECLARE_GL_FUNCTION_STUB_HEAD(void, Uniform3i64ARB, GLint location, GLint64 x, GLint64 y, GLint64 z) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, Uniform3i64ARB, location, x, y, z)
@@ -1848,9 +1835,9 @@ DECLARE_GL_FUNCTION_STUB_HEAD(void, GetBooleanIndexedvEXT, GLenum target, GLuint
 DECLARE_GL_FUNCTION_STUB_HEAD(void, CompressedTextureImage3DEXT, GLuint texture, GLenum target, GLint level, GLenum internalformat, GLsizei width, GLsizei height, GLsizei depth, GLint border, GLsizei imageSize, const void* bits) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, CompressedTextureImage3DEXT, texture, target, level, internalformat, width, height, depth, border, imageSize, bits)
 DECLARE_GL_FUNCTION_STUB_HEAD(void, CompressedTextureImage2DEXT, GLuint texture, GLenum target, GLint level, GLenum internalformat, GLsizei width, GLsizei height, GLint border, GLsizei imageSize, const void* bits) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, CompressedTextureImage2DEXT, texture, target, level, internalformat, width, height, border, imageSize, bits)
 DECLARE_GL_FUNCTION_STUB_HEAD(void, CompressedTextureImage1DEXT, GLuint texture, GLenum target, GLint level, GLenum internalformat, GLsizei width, GLint border, GLsizei imageSize, const void* bits) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, CompressedTextureImage1DEXT, texture, target, level, internalformat, width, border, imageSize, bits)
-DECLARE_GL_FUNCTION_STUB_HEAD(void, CompressedTextureSubImage3DEXT, GLuint texture, GLenum target, GLint level, GLint xoffset, GLint yoffset, GLint zoffset, GLsizei width, GLsizei height, GLsizei depth, GLenum format, GLsizei imageSize, const void* bits) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, CompressedTextureSubImage3DEXT, texture, target, level, xoffset, yoffset, zoffset, width, height, depth, format, imageSize, bits)
+DECLARE_GL_FUNCTION_HEAD(void, CompressedTextureSubImage3DEXT, GLuint texture, GLenum target, GLint level, GLint xoffset, GLint yoffset, GLint zoffset, GLsizei width, GLsizei height, GLsizei depth, GLenum format, GLsizei imageSize, const void* bits) DECLARE_GL_FUNCTION_END_NO_RETURN(void, CompressedTextureSubImage3D, texture, level, xoffset, yoffset, zoffset, width, height, depth, format, imageSize, bits)
 DECLARE_GL_FUNCTION_HEAD(void, CompressedTextureSubImage2DEXT, GLuint texture, GLenum target, GLint level, GLint xoffset, GLint yoffset, GLsizei width, GLsizei height, GLenum format, GLsizei imageSize, const void* bits) DECLARE_GL_FUNCTION_END_NO_RETURN(void, CompressedTextureSubImage2D, texture, level, xoffset, yoffset, width, height, format, imageSize, bits)
-DECLARE_GL_FUNCTION_STUB_HEAD(void, CompressedTextureSubImage1DEXT, GLuint texture, GLenum target, GLint level, GLint xoffset, GLsizei width, GLenum format, GLsizei imageSize, const void* bits) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, CompressedTextureSubImage1DEXT, texture, target, level, xoffset, width, format, imageSize, bits)
+DECLARE_GL_FUNCTION_HEAD(void, CompressedTextureSubImage1DEXT, GLuint texture, GLenum target, GLint level, GLint xoffset, GLsizei width, GLenum format, GLsizei imageSize, const void* bits) DECLARE_GL_FUNCTION_END_NO_RETURN(void, CompressedTextureSubImage1D, texture, level, xoffset, width, format, imageSize, bits)
 DECLARE_GL_FUNCTION_STUB_HEAD(void, GetCompressedTextureImageEXT, GLuint texture, GLenum target, GLint lod, void* img) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, GetCompressedTextureImageEXT, texture, target, lod, img)
 DECLARE_GL_FUNCTION_STUB_HEAD(void, CompressedMultiTexImage3DEXT, GLenum texunit, GLenum target, GLint level, GLenum internalformat, GLsizei width, GLsizei height, GLsizei depth, GLint border, GLsizei imageSize, const void* bits) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, CompressedMultiTexImage3DEXT, texunit, target, level, internalformat, width, height, depth, border, imageSize, bits)
 DECLARE_GL_FUNCTION_STUB_HEAD(void, CompressedMultiTexImage2DEXT, GLenum texunit, GLenum target, GLint level, GLenum internalformat, GLsizei width, GLsizei height, GLint border, GLsizei imageSize, const void* bits) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, CompressedMultiTexImage2DEXT, texunit, target, level, internalformat, width, height, border, imageSize, bits)
@@ -2062,7 +2049,7 @@ DECLARE_GL_FUNCTION_STUB_HEAD(void, GetPixelTransformParameterivEXT, GLenum targ
 DECLARE_GL_FUNCTION_STUB_HEAD(void, GetPixelTransformParameterfvEXT, GLenum target, GLenum pname, GLfloat* params) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, GetPixelTransformParameterfvEXT, target, pname, params)
 DECLARE_GL_FUNCTION_STUB_HEAD(void, PointParameterfEXT, GLenum pname, GLfloat param) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, PointParameterfEXT, pname, param)
 DECLARE_GL_FUNCTION_STUB_HEAD(void, PointParameterfvEXT, GLenum pname, const GLfloat* params) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, PointParameterfvEXT, pname, params)
-DECLARE_GL_FUNCTION_STUB_HEAD(void, PolygonOffsetClampEXT, GLfloat factor, GLfloat units, GLfloat clamp) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, PolygonOffsetClampEXT, factor, units, clamp)
+DECLARE_GL_FUNCTION_HEAD(void, PolygonOffsetClampEXT, GLfloat factor, GLfloat units, GLfloat clamp) DECLARE_GL_FUNCTION_END_NO_RETURN(void, PolygonOffsetClamp, factor, units, clamp)
 DECLARE_GL_FUNCTION_HEAD(void, ProvokingVertexEXT, GLenum mode) DECLARE_GL_FUNCTION_END_NO_RETURN(void, ProvokingVertex, mode)
 DECLARE_GL_FUNCTION_STUB_HEAD(void, RasterSamplesEXT, GLuint samples, GLboolean fixedsamplelocations) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, RasterSamplesEXT, samples, fixedsamplelocations)
 DECLARE_GL_FUNCTION_STUB_HEAD(void, SecondaryColor3bEXT, GLbyte red, GLbyte green, GLbyte blue) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, SecondaryColor3bEXT, red, green, blue)
@@ -2559,7 +2546,7 @@ DECLARE_GL_FUNCTION_STUB_HEAD(void, ShadingRateImageBarrierNV, GLboolean synchro
 DECLARE_GL_FUNCTION_STUB_HEAD(void, ShadingRateImagePaletteNV, GLuint viewport, GLuint first, GLsizei count, const GLenum* rates) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, ShadingRateImagePaletteNV, viewport, first, count, rates)
 DECLARE_GL_FUNCTION_STUB_HEAD(void, ShadingRateSampleOrderNV, GLenum order) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, ShadingRateSampleOrderNV, order)
 DECLARE_GL_FUNCTION_STUB_HEAD(void, ShadingRateSampleOrderCustomNV, GLenum rate, GLuint samples, const GLint* locations) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, ShadingRateSampleOrderCustomNV, rate, samples, locations)
-DECLARE_GL_FUNCTION_STUB_HEAD(void, TextureBarrierNV, void) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, TextureBarrierNV, )
+DECLARE_GL_FUNCTION_HEAD(void, TextureBarrierNV, void) DECLARE_GL_FUNCTION_END_NO_RETURN(void, TextureBarrier, )
 DECLARE_GL_FUNCTION_STUB_HEAD(void, TexImage2DMultisampleCoverageNV, GLenum target, GLsizei coverageSamples, GLsizei colorSamples, GLint internalFormat, GLsizei width, GLsizei height, GLboolean fixedSampleLocations) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, TexImage2DMultisampleCoverageNV, target, coverageSamples, colorSamples, internalFormat, width, height, fixedSampleLocations)
 DECLARE_GL_FUNCTION_STUB_HEAD(void, TexImage3DMultisampleCoverageNV, GLenum target, GLsizei coverageSamples, GLsizei colorSamples, GLint internalFormat, GLsizei width, GLsizei height, GLsizei depth, GLboolean fixedSampleLocations) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, TexImage3DMultisampleCoverageNV, target, coverageSamples, colorSamples, internalFormat, width, height, depth, fixedSampleLocations)
 DECLARE_GL_FUNCTION_STUB_HEAD(void, TextureImage2DMultisampleNV, GLuint texture, GLenum target, GLsizei samples, GLint internalFormat, GLsizei width, GLsizei height, GLboolean fixedSampleLocations) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, TextureImage2DMultisampleNV, texture, target, samples, internalFormat, width, height, fixedSampleLocations)

MobileGL/MG_Impl/GLImpl/Framebuffer/GL_Framebuffer.cpp

@@ -474,6 +474,75 @@ namespace MobileGL::MG_Impl::GLImpl {
             }
         }
 
+        // GL 4.6 core 9.2.8 conditions that depend only on the framebuffer and the attachment
+        // point. Shared, because glFramebufferTexture / 1D / 2D / 3D / TextureLayer are aliases of
+        // one another in that section and a CTS case that walks the family must not get five
+        // different answers - which is exactly what happened when these lived in one helper that
+        // only two of the five went through.
+        Bool ValidateFramebufferTextureAttachmentPoint(const char* functionName,
+                                                       const SharedPtr<MG_State::GLState::FramebufferObject>&
+                                                           framebufferObject,
+                                                       FramebufferAttachmentType attachmentType) {
+            // "An INVALID_OPERATION error is generated if COLOR_ATTACHMENTm is used with m greater
+            // than or equal to MAX_COLOR_ATTACHMENTS."
+            if (!FramebufferImpl::ValidateColorAttachmentInRange(attachmentType, functionName)) return false;
+            // "An INVALID_OPERATION error is generated if zero is bound to target." MobileGL keeps
+            // a real FramebufferObject for framebuffer 0, so a null test can never see this - the
+            // object is always there, and framebuffer 0 has to be recognised by identity instead,
+            // the same comparison DrawBuffers_State makes. Without this an attach onto the default
+            // framebuffer silently REPLACED its colour attachment, permanently desynchronising it
+            // from what the swapchain keeps publishing.
+            const auto& defaultFramebufferInfo = FramebufferImpl::pDefaultFramebufferInfo;
+            if (!framebufferObject ||
+                (defaultFramebufferInfo && framebufferObject == defaultFramebufferInfo->defaultFBO)) {
+                MG_State::pGLContext->RecordError(
+                    ErrorCode::InvalidOperation,
+                    MakeUnique<GenericErrorInfo>(
+                        "MG_Impl/GLImpl", functionName,
+                        "No framebuffer object is bound to the target; the default framebuffer's attachments "
+                        "cannot be named."));
+                return false;
+            }
+            return true;
+        }
+
+        // The other half of 9.2.8: "level must be greater than or equal to zero", and for a
+        // texture with immutable storage it "must be smaller than the number of levels the texture
+        // has". Split from the attachment-point half because the caller only has a texture object
+        // once the detach (texture == 0) case is behind it.
+        Bool ValidateFramebufferTextureLevel(const char* functionName,
+                                             const SharedPtr<MG_State::GLState::ITextureObject>& textureObject,
+                                             GLint level) {
+            if (level < 0) {
+                MG_State::pGLContext->RecordError(
+                    ErrorCode::InvalidValue,
+                    MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", functionName,
+                                                 "Texture level must be non-negative."));
+                return false;
+            }
+            if (!textureObject || !textureObject->IsImmutable()) {
+                // A mutable texture has no level bound here: a level it has not specified yet is
+                // not an error, it just leaves the framebuffer incomplete.
+                return true;
+            }
+            // GetAddressableLevelCount(), NOT GetImmutableLevels(): for a VIEW the latter is
+            // deliberately the ORIGINAL texture's count (GL 4.6 core 8.18 defines
+            // TEXTURE_IMMUTABLE_LEVELS on a view that way), which is far too large a bound - a
+            // two-level view onto a ten-level texture would accept level 5 and attach an image
+            // nothing can draw into.
+            const Uint levelBound = textureObject->GetAddressableLevelCount();
+            if (static_cast<Uint>(level) >= levelBound) {
+                MG_State::pGLContext->RecordError(
+                    ErrorCode::InvalidValue,
+                    MakeUnique<GenericErrorInfo>(
+                        "MG_Impl/GLImpl", functionName,
+                        std::format("Texture level {} is beyond the {} level(s) this texture has.", level,
+                                    levelBound)));
+                return false;
+            }
+            return true;
+        }
+
         void AttachFramebufferTextureWithUploadTarget(const char* functionName, GLenum target, GLenum attachment,
                                                       GLuint texture, GLint level,
                                                       TextureUploadTarget textureUploadTarget, Bool layered = false) {
@@ -482,10 +551,24 @@ namespace MobileGL::MG_Impl::GLImpl {
             }
 
             if (attachment == GL_DEPTH_STENCIL_ATTACHMENT) {
+                // `layered` has to travel with the split. GL_DEPTH_STENCIL_ATTACHMENT is only a
+                // shorthand for attaching the same image to both halves (GL 4.6 core 9.2.6), so
+                // whether glFramebufferTexture made it LAYERED is a property of the call, not of
+                // which half is being recorded - and dropping it here (the parameter defaults to
+                // false) recorded a non-layered depth/stencil attachment beside a layered colour
+                // one for every layered target. That is an inconsistent framebuffer by 9.4.1's
+                // own rule, and downstream it means the depth/stencil attachment covers layer 0
+                // alone: DirectVulkan built its view with layerCount 1 under a framebuffer
+                // declaring N layers (VUID-VkFramebufferCreateInfo-flags-04535), and DirectGLES
+                // attached one layer of it beside a layered colour target, which the driver
+                // answers with GL_FRAMEBUFFER_INCOMPLETE_LAYER_TARGETS - every draw silently
+                // produced nothing. This is the shape
+                // texture_cube_map_array.stencil_attachments_*_layered and
+                // geometry_shader.layered_framebuffer.stencil_support are built on.
                 AttachFramebufferTextureWithUploadTarget(functionName, target, GL_DEPTH_ATTACHMENT, texture, level,
-                                                        textureUploadTarget);
+                                                        textureUploadTarget, layered);
                 AttachFramebufferTextureWithUploadTarget(functionName, target, GL_STENCIL_ATTACHMENT, texture, level,
-                                                        textureUploadTarget);
+                                                        textureUploadTarget, layered);
                 return;
             }
 
@@ -497,13 +580,7 @@ namespace MobileGL::MG_Impl::GLImpl {
 
             auto& bindingSlot = MG_State::pGLContext->GetFramebufferBindingSlot(framebufferTarget);
             auto& framebufferObject = bindingSlot.GetBoundObject();
-            if (!framebufferObject) {
-                MG_State::pGLContext->RecordError(
-                    ErrorCode::InvalidOperation,
-                    MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", functionName,
-                                                 "Framebuffer target is bound to no framebuffer object."));
-                return;
-            }
+            if (!ValidateFramebufferTextureAttachmentPoint(functionName, framebufferObject, attachmentType)) return;
 
             if (texture == 0) {
                 framebufferObject->Detach(attachmentType);
@@ -518,6 +595,7 @@ namespace MobileGL::MG_Impl::GLImpl {
                                                  std::format("Texture object {} is not valid.", texture)));
                 return;
             }
+            if (!ValidateFramebufferTextureLevel(functionName, textureObject, level)) return;
 
             const auto expectedTextureTarget = MG_Util::ConvertTextureUploadTargetToTextureTarget(textureUploadTarget);
             if (expectedTextureTarget == TextureTarget::Unknown ||
@@ -624,16 +702,33 @@ namespace MobileGL::MG_Impl::GLImpl {
     // GL_MAX_SAMPLES is the ceiling over all formats; an integer format has its own
     // (GL_MAX_INTEGER_SAMPLES) and GL 4.6 core 9.2.4 makes exceeding it INVALID_OPERATION.
     // The multisample TEXTURE path resolves the limit per format the same way
-    // (GL_Texture.cpp, GetMaxSupportedTextureSamples). Both are floored to the value MobileGL
-    // advertises: on a driver where the two differ - Adreno reports GL_MAX_SAMPLES 4 and
-    // GL_MAX_INTEGER_SAMPLES 1 - rejecting the advertised count here only moves the failure
-    // from the driver into MobileGL, so the frontend accepts it and the backend clamps the
-    // count it actually hands the driver.
+    // (GL_Texture.cpp, GetMaxSupportedTextureSamples), and both now enforce exactly what their
+    // pname advertises. The integer ceiling used to be floored at GL_MAX_SAMPLES so that the
+    // frontend would accept a count it had advertised globally - but on Adreno and Mali the
+    // integer path is genuinely one sample, and accepting four only moved the failure from an
+    // honest INVALID_OPERATION here to a silently under-allocated renderbuffer.
+    // The head of the per-format renderbuffer sample list the backend probed, or 0 when nothing
+    // was probed for it. Same shape as GetProbedMaxTextureSamples in GL_Texture.cpp, and reads
+    // the same cache glGetInternalformativ(GL_RENDERBUFFER, ..., GL_SAMPLES) answers from.
+    static Int GetProbedMaxRenderbufferSamples(TextureInternalFormat format) {
+        if (MG_Backend::pActiveBackendObject == nullptr) {
+            return 0;
+        }
+        const SizeT targetIndex = MG_Backend::GetRenderbufferFormatCapabilityTargetIndex();
+        const SizeT formatIndex = static_cast<SizeT>(format);
+        if (targetIndex >= MG_Backend::kFormatCapabilityTargetCount ||
+            formatIndex >= MG_Backend::kFormatCapabilityFormatCount) {
+            return 0;
+        }
+        const auto& sampleCounts =
+            MG_Backend::pActiveBackendObject->GetFormatCapabilities().SampleCounts[targetIndex][formatIndex];
+        return sampleCounts.empty() ? 0 : sampleCounts.front();
+    }
+
     Int GetMaxRenderbufferSamplesForFormat_State(TextureInternalFormat format) {
         if (MG_Backend::pActiveBackendObject == nullptr) {
             return std::numeric_limits<Int>::max();
         }
-        const auto& dynamicParameters = MG_Backend::pActiveBackendObject->GetDynamicParameters();
 
         GLenum normalizedInternalFormat = MG_Util::ConvertTextureInternalFormatToGLEnum(format);
         GLenum normalizedFormat = GL_RGBA;
@@ -644,13 +739,24 @@ namespace MobileGL::MG_Impl::GLImpl {
                                                               &normalizedType);
         const Bool isIntegerFormat = normalizedFormat == GL_RED_INTEGER || normalizedFormat == GL_RG_INTEGER ||
                                      normalizedFormat == GL_RGB_INTEGER || normalizedFormat == GL_RGBA_INTEGER;
+        // The per-format probe first, for the same reason the texture path takes it first: GL 4.6
+        // core 9.2.4 words the error as "samples is greater than the maximum number of samples
+        // supported for internalformat (see GetInternalformativ)", and
+        // glGetInternalformativ(GL_RENDERBUFFER, ..., GL_SAMPLES) is answered from exactly this
+        // list. It was never consulted here - the TODO that deferred it was written before the
+        // query was backed and had gone stale - so a format whose multisample probes fail inside
+        // a category that allows four was accepted at four, quietly allocated at one by
+        // ClampSamplesToBackendSupport, and then reported as four by
+        // glGetRenderbufferParameteriv(GL_RENDERBUFFER_SAMPLES).
+        const Int probedMaxSamples = GetProbedMaxRenderbufferSamples(format);
+        if (probedMaxSamples > 0) {
+            return probedMaxSamples;
+        }
         if (!isIntegerFormat) {
             return GetMaxRenderbufferSamples_State();
         }
-        // Per-format still, but never below the ceiling glGetIntegerv(GL_MAX_SAMPLES) promised:
-        // the driver's raw GL_MAX_INTEGER_SAMPLES stays the *backend* limit and the backend
-        // clamps to it, while the frontend honours what it advertised.
-        return std::max(dynamicParameters.MaxIntegerSamples, GetAdvertisedMaxSamples());
+        // Exactly what glGetIntegerv(GL_MAX_INTEGER_SAMPLES) reports.
+        return GetAdvertisedIntegerMaxSamples();
     }
 
     Bool ValidateRenderbufferStorageSize_State(GLsizei width, GLsizei height, const char* caller) {
@@ -682,8 +788,10 @@ namespace MobileGL::MG_Impl::GLImpl {
             return false;
         }
 
-        // TODO: Resolve the remaining per-internalformat renderbuffer sample limits once
-        // glGetInternalformativ is backed; integer formats are handled below.
+        // Per-internalformat, from the probe list glGetInternalformativ answers with, falling back
+        // to the format's category pname where nothing was probed. (This carried a TODO deferring
+        // the per-format resolution "once glGetInternalformativ is backed"; it has been backed for
+        // both renderbuffers and multisample textures since, so the deferral was collected.)
         const Int maxSamples = GetMaxRenderbufferSamplesForFormat_State(format);
         if (samples > maxSamples) {
             // GL 4.6 core 9.2.4 makes asking for more samples than the format supports
@@ -1048,13 +1156,7 @@ namespace MobileGL::MG_Impl::GLImpl {
 
         auto& bindingSlot = MG_State::pGLContext->GetFramebufferBindingSlot(framebufferTarget);
         auto& framebufferObject = bindingSlot.GetBoundObject();
-        if (!framebufferObject) {
-            MG_State::pGLContext->RecordError(
-                ErrorCode::InvalidOperation,
-                MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", functionName,
-                                             "Framebuffer target is bound to no framebuffer object."));
-            return;
-        }
+        if (!ValidateFramebufferTextureAttachmentPoint(functionName, framebufferObject, attachmentType)) return;
 
         if (texture == 0) {
             framebufferObject->Detach(attachmentType);
@@ -1069,6 +1171,7 @@ namespace MobileGL::MG_Impl::GLImpl {
                                              std::format("Texture object {} is not valid.", texture)));
             return;
         }
+        if (!ValidateFramebufferTextureLevel(functionName, textureObject, level)) return;
         if (layer < 0) {
             MG_State::pGLContext->RecordError(
                 ErrorCode::InvalidValue,
@@ -1191,6 +1294,13 @@ namespace MobileGL::MG_Impl::GLImpl {
                                              "Framebuffer target is bound to no framebuffer object."));
             return;
         }
+        // glFramebufferTexture2D is by far the most-used member of the family and the only one
+        // that inlines its own logic instead of going through the shared helper, so the 9.2.8
+        // conditions have to be asked here explicitly.
+        if (!ValidateFramebufferTextureAttachmentPoint("FramebufferTexture2D_State", framebufferObject,
+                                                       attachmentType)) {
+            return;
+        }
 
         if (texture == 0) {
             framebufferObject->Detach(attachmentType);
@@ -1205,6 +1315,7 @@ namespace MobileGL::MG_Impl::GLImpl {
                                              std::format("Texture object {} is not valid.", texture)));
             return;
         }
+        if (!ValidateFramebufferTextureLevel("FramebufferTexture2D_State", textureObject, level)) return;
 
         const auto expectedTextureTarget = MG_Util::ConvertTextureUploadTargetToTextureTarget(textureUploadTarget);
         if (expectedTextureTarget == TextureTarget::Unknown ||
@@ -1241,6 +1352,12 @@ namespace MobileGL::MG_Impl::GLImpl {
             return;
         }
 
+        // The name's validity is an INVALID_VALUE condition (GL 4.6 core 9.2.8), and it has to be
+        // asked BEFORE the object is resolved: reporting the miss as the INVALID_OPERATION below
+        // pre-empted the shared helper's ValidateTextureName and answered the wrong error code for
+        // every texture name that was never generated.
+        if (!TextureImpl::ValidateTextureName(texture, true)) return;
+
         auto& textureObject = MG_State::pGLContext->GetTextureObject(texture);
         if (!textureObject) {
             MG_State::pGLContext->RecordError(
@@ -1291,13 +1408,10 @@ namespace MobileGL::MG_Impl::GLImpl {
                                              std::format("Texture object {} is not valid.", texture)));
             return;
         }
-        if (level < 0) {
-            MG_State::pGLContext->RecordError(
-                ErrorCode::InvalidValue,
-                MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", "NamedFramebufferTexture_State",
-                                             "Texture level must be non-negative."));
-            return;
-        }
+        // The whole level condition, not just its negative half: glNamedFramebufferTexture and
+        // glFramebufferTexture are equivalent in 9.2.8, so an out-of-range immutable level has to
+        // be rejected on both or a CTS case gets two answers for one rule.
+        if (!ValidateFramebufferTextureLevel("NamedFramebufferTexture_State", textureObject, level)) return;
 
         TextureUploadTarget textureUploadTarget = TextureUploadTarget::Unknown;
         Bool layered = false;

MobileGL/MG_Impl/GLImpl/Getter/GL_Getter.cpp

@@ -7,9 +7,12 @@
 // End of Source File Header
 
 #include "GL_Getter.h"
+#include <algorithm>
 #include <cmath>
+#include <limits>
 #include <Config.h>
 #include <MGGitHash.h>
+#include <MG_Impl/GLImpl/Debug/GL_Debug.h>
 #include <MG_Impl/GLImpl/VertexArray/Validators.h>
 #include <MG_State/EGLState/Core.h>
 #include <MG_State/GLState/Core.h>
@@ -92,8 +95,15 @@ namespace MobileGL::MG_Impl::GLImpl {
         // limits they advertise still have to be legal.
         constexpr GLint kFrontendMaxDebugGroupStackDepth = 64;
         constexpr GLint kFrontendMaxDebugLoggedMessages = 1;
-        constexpr GLint kFrontendMaxVertexUniformComponents = 4096;
-        constexpr GLint kFrontendMaxVertexUniformVectors = 128;
+        // The *_VECTORS answers are the *_COMPONENTS ones divided by four, never a second
+        // literal: they used to be independent (4096 components against 128 vectors, 64 varying
+        // components against 8 varying vectors) and could not both be describing the same
+        // capacity. Both are shared with BuildTBuiltInResource through Types.h, because
+        // gl_MaxVertexUniformVectors and gl_MaxVaryingVectors expand from the same numbers.
+        constexpr GLint kFrontendMaxVertexUniformComponents =
+            static_cast<GLint>(MG_Util::ShaderTranspiler::MAX_VERTEX_UNIFORM_COMPONENTS);
+        constexpr GLint kFrontendMaxVertexUniformVectors =
+            static_cast<GLint>(MG_Util::ShaderTranspiler::MAX_VERTEX_UNIFORM_VECTORS);
         constexpr GLint kFrontendMaxVertexUniformBlocks = 14;
         constexpr GLint kFrontendMaxVertexOutputComponents = 64;
         constexpr GLint kFrontendMaxFragmentInputComponents = 128;
@@ -105,21 +115,61 @@ namespace MobileGL::MG_Impl::GLImpl {
         constexpr GLint kFrontendMaxGeometryTextureImageUnits = 16;
         constexpr GLint kFrontendMaxGeometryUniformComponents = 1024;
         constexpr GLint kFrontendMaxGeometryUniformBlocks = 14;
-        constexpr GLint kFrontendMaxCombinedUniformBlocks = kFrontendMaxVertexUniformBlocks +
-                                                            kFrontendMaxGeometryUniformBlocks +
-                                                            kFrontendMaxFragmentUniformBlocks;
-        constexpr GLint kFrontendMaxVaryingComponents = 64;
-        constexpr GLint kFrontendMaxVaryingVectors = 8;
+        // ARB_geometry_shader4's per-invocation count. No TBuiltInResource field and no
+        // gl_MaxGeometryShaderInvocations built-in exists to keep in step, so this is a getter
+        // answer only; 32 is the GL 4.6 core minimum (table 23.57).
+        constexpr GLint kFrontendMaxGeometryShaderInvocations = 32;
+        constexpr GLint kFrontendMaxTessControlUniformBlocks = 14;
+        constexpr GLint kFrontendMaxTessEvaluationUniformBlocks = 14;
+        // The compute stage's share of the combined sum below. Compute's own per-stage answer is
+        // backend-derived (GL_MAX_COMPUTE_UNIFORM_BLOCKS reads dynamicParameters), so this is not
+        // what that query returns - it is the GL 4.3 core minimum, present here only so the
+        // combined total covers all SIX stages.
+        constexpr GLint kFrontendMaxComputeUniformBlocksShare = 14;
+        // GL 4.6 table 23.64 orders MAX_UNIFORM_BUFFER_BINDINGS >= MAX_COMBINED_UNIFORM_BLOCKS >=
+        // every per-stage count, and the sum has to run over SIX stages, not three and not five.
+        // Three (42) was the original bug. Five (70) replaced it and broke the middle term the
+        // other way: compute's per-stage count is backend-derived and clamps at the binding count,
+        // so a device reporting descriptor-indexing-scale uniform buffers (Adreno reports
+        // maxPerStageDescriptorUniformBuffers = 16777216) advertised 84 compute blocks against a
+        // combined 70. Six stages x 14 = 84, which is also exactly the binding-point count and the
+        // arithmetic the GL 4.5 minimum of 84 bindings is built from, so the ordering is now tight
+        // rather than accidental.
+        constexpr GLint kFrontendMaxCombinedUniformBlocks =
+            kFrontendMaxVertexUniformBlocks + kFrontendMaxTessControlUniformBlocks +
+            kFrontendMaxTessEvaluationUniformBlocks + kFrontendMaxGeometryUniformBlocks +
+            kFrontendMaxFragmentUniformBlocks + kFrontendMaxComputeUniformBlocksShare;
+        constexpr GLint kFrontendMaxVaryingComponents =
+            static_cast<GLint>(MG_Util::ShaderTranspiler::MAX_VARYING_COMPONENTS);
+        constexpr GLint kFrontendMaxVaryingVectors =
+            static_cast<GLint>(MG_Util::ShaderTranspiler::MAX_VARYING_VECTORS);
         constexpr GLint kFrontendMaxProgramTexelOffset = 7;
         constexpr GLint kFrontendMinProgramTexelOffset = -8;
         constexpr GLint kFrontendMaxTransformFeedbackInterleavedComponents = 64;
         constexpr GLint kFrontendMaxTransformFeedbackSeparateAttribs = 4;
         constexpr GLint kFrontendMaxTransformFeedbackSeparateComponents = 4;
+        // ARB_transform_feedback3's vertex-stream count. One is what this implementation can
+        // actually emit to; see the GL_MAX_VERTEX_STREAMS case for why it is not four.
+        constexpr GLint kFrontendMaxVertexStreams = 1;
         constexpr GLint kFrontendMaxGeometryOutputVertices = 256;
         constexpr GLint kFrontendMaxGeometryTotalOutputComponents = 1024;
-        constexpr GLint kFrontendMinUniformBufferBindings = 36;
+        // GL 4.5 core table 23.64 requires 84 indexed uniform binding points, and that is exactly
+        // how wide the state layer's array is (BufferState::BufferBindingPointCount) - see the
+        // GL_MAX_UNIFORM_BUFFER_BINDINGS case for why the ES driver's own, smaller count is not
+        // the ceiling here.
+        constexpr GLint kFrontendMinUniformBufferBindings = 84;
         constexpr GLint kFrontendSubpixelBits = 4;
-        constexpr GLint kFrontendMaxSamples = 4;
+        constexpr GLint kFrontendMaxSamples =
+            static_cast<GLint>(MG_Util::ShaderTranspiler::MIN_ADVERTISED_MAX_SAMPLES);
+        // ARB_shader_subroutine's two limits. NOTHING IMPLEMENTS SUBROUTINES: there is no
+        // glGetSubroutineIndex / glUniformSubroutinesuiv, only the program-interface enum
+        // plumbing. These are answered - with the GL 4.5 core minimums - because the conformance
+        // suite queries them before it checks for the feature and an INVALID_ENUM both leaves the
+        // caller reading its own uninitialised stack slot and strands an error for the next
+        // unrelated call to trip over. The extension is deliberately NOT advertised, so the
+        // numbers are a table entry, not a capability claim.
+        constexpr GLint kFrontendMaxSubroutines = 256;
+        constexpr GLint kFrontendMaxSubroutineUniformLocations = 1024;
 
         // The floors under GL_MAX_COMPUTE_WORK_GROUP_COUNT / _SIZE. Shared with the compile
         // pipeline (CaptureCompileEnv floors the same driver answers at them, and
@@ -133,9 +183,19 @@ namespace MobileGL::MG_Impl::GLImpl {
             return index < 3 ? static_cast<GLint>(MG_Util::ShaderTranspiler::MIN_COMPUTE_WORK_GROUP_SIZE[index]) : 0;
         }
 
+        // GL 4.6 core table 23.64: components + blocks * (blockSize / 4). The product has to be
+        // formed in 64 bits and saturated on the way out - it overflowed a signed 32-bit int on
+        // every Vulkan host that reports a large maxUniformBufferRange. A Mali driver answering
+        // 0xFFFFFFFF saturates to INT32_MAX in the loader, and 14 * (2147483647 / 4) + 4096 wraps
+        // to -1073737742, which the conformance suite read back as a limit "smaller than 58368".
+        // Saturating instead of wrapping is also the only honest answer: an implementation that
+        // can serve more components than a GLint holds still has to report a GLint.
         GLint GetMaxCombinedUniformComponents(GLint maxDefaultUniformComponents, GLint maxUniformBlocks,
                                               GLint maxUniformBlockSizeBytes) {
-            return maxDefaultUniformComponents + maxUniformBlocks * (maxUniformBlockSizeBytes / 4);
+            const Int64 blocks = std::max<Int64>(static_cast<Int64>(maxUniformBlocks), 0);
+            const Int64 componentsPerBlock = std::max<Int64>(static_cast<Int64>(maxUniformBlockSizeBytes), 0) / 4;
+            const Int64 total = static_cast<Int64>(maxDefaultUniformComponents) + blocks * componentsPerBlock;
+            return static_cast<GLint>(std::min<Int64>(total, std::numeric_limits<GLint>::max()));
         }
 
         bool TryDecodeIndexedBufferQuery(GLenum pname, BufferTarget& bufferTarget, IndexedBufferQueryKind& queryKind) {
@@ -303,24 +363,6 @@ namespace MobileGL::MG_Impl::GLImpl {
             return true;
         }
 
-        GLint ResolveDrawFramebufferSampleCount() {
-            const auto& drawFbo =
-                MG_State::pGLContext->GetFramebufferBindingSlot(FramebufferTarget::Draw).GetBoundObject();
-            if (!drawFbo) return 0;
-
-            GLint maxSamples = 0;
-            for (const auto& attachment : drawFbo->GetAllAttachmentObjects()) {
-                if (attachment.IsRenderbuffer() && attachment.GetRenderbuffer()) {
-                    maxSamples = std::max(maxSamples, static_cast<GLint>(attachment.GetRenderbuffer()->GetSamples()));
-                } else if (attachment.IsTexture() && attachment.GetTexture()) {
-                    // Multisample texture attachments count too (GL_SAMPLE_BUFFERS must
-                    // report 1 for any multisampled draw framebuffer).
-                    maxSamples = std::max(maxSamples, static_cast<GLint>(attachment.GetTexture()->GetSamples()));
-                }
-            }
-            return maxSamples;
-        }
-
         void RecordIndexedOnlyGetterError(const char* functionName, GLenum pname) {
             MG_State::pGLContext->RecordError(
                 ErrorCode::InvalidEnum,
@@ -472,10 +514,18 @@ namespace MobileGL::MG_Impl::GLImpl {
     } // namespace
 
     // GL 4.6 core table 23.53 requires GL_MAX_SAMPLES >= 4, so the driver's value is floored
-    // before it is advertised. Every other multisample ceiling MobileGL advertises has to be
-    // floored the same way: promising 4 samples globally while answering GL_MAX_INTEGER_SAMPLES
-    // 1 - which is exactly what Adreno reports - makes the frontend reject the very count it
-    // just told the application to use. The backends clamp the realised count instead.
+    // before it is advertised. gl_MaxSamples expands from the same floored number
+    // (BuildTBuiltInResource), which is also what sizes gl_SampleMask[].
+    //
+    // THE FLOOR STOPS HERE, and that is the point. It used to be applied to
+    // GL_MAX_INTEGER_SAMPLES, GL_MAX_COLOR_TEXTURE_SAMPLES and GL_MAX_DEPTH_TEXTURE_SAMPLES too,
+    // on the reasoning that an application reads GL_MAX_SAMPLES once and hands that count to
+    // every glTexStorage*Multisample. Table 23.53 gives those three a minimum of ONE, and the
+    // reasoning had it backwards: Adreno and Mali back an integer multisample texture with a
+    // single sample, so flooring the query at 4 did not make four samples exist - it made the
+    // backend silently under-allocate (ClampSamplesToBackendSupport) while the application wrote
+    // per-sample data it could never read back. Reporting what was probed turns that into an
+    // honest "unsupported" the application can branch on.
     GLint GetAdvertisedMaxSamples() {
         if (MG_Backend::pActiveBackendObject == nullptr) {
             return kFrontendMaxSamples;
@@ -483,6 +533,50 @@ namespace MobileGL::MG_Impl::GLImpl {
         return std::max(MG_Backend::pActiveBackendObject->GetDynamicParameters().MaxSamples, kFrontendMaxSamples);
     }
 
+    // GL 4.6 core table 23.53 minimum for the per-category multisample ceilings. One, not four:
+    // see the note on GetAdvertisedMaxSamples. A zero would be a probe that never ran, so it is
+    // floored rather than trusted.
+    namespace {
+        GLint AdvertisedCategoryMaxSamples(Int MG_Backend::DynamicBackendParameters::*categoryLimit) {
+            if (MG_Backend::pActiveBackendObject == nullptr) {
+                return 1;
+            }
+            return std::max(MG_Backend::pActiveBackendObject->GetDynamicParameters().*categoryLimit, 1);
+        }
+    } // namespace
+
+    GLint GetAdvertisedColorTextureMaxSamples() {
+        return AdvertisedCategoryMaxSamples(&MG_Backend::DynamicBackendParameters::MaxColorTextureSamples);
+    }
+
+    GLint GetAdvertisedDepthTextureMaxSamples() {
+        return AdvertisedCategoryMaxSamples(&MG_Backend::DynamicBackendParameters::MaxDepthTextureSamples);
+    }
+
+    GLint GetAdvertisedIntegerMaxSamples() {
+        return AdvertisedCategoryMaxSamples(&MG_Backend::DynamicBackendParameters::MaxIntegerSamples);
+    }
+
+    // Declared in GL_Getter.h, so that the draw path can feed the same number to the reserved
+    // gl_NumSamples stand-in that glGetIntegerv(GL_SAMPLES) reports.
+    GLint ResolveDrawFramebufferSampleCount() {
+        const auto& drawFbo =
+            MG_State::pGLContext->GetFramebufferBindingSlot(FramebufferTarget::Draw).GetBoundObject();
+        if (!drawFbo) return 0;
+
+        GLint maxSamples = 0;
+        for (const auto& attachment : drawFbo->GetAllAttachmentObjects()) {
+            if (attachment.IsRenderbuffer() && attachment.GetRenderbuffer()) {
+                maxSamples = std::max(maxSamples, static_cast<GLint>(attachment.GetRenderbuffer()->GetSamples()));
+            } else if (attachment.IsTexture() && attachment.GetTexture()) {
+                // Multisample texture attachments count too (GL_SAMPLE_BUFFERS must
+                // report 1 for any multisampled draw framebuffer).
+                maxSamples = std::max(maxSamples, static_cast<GLint>(attachment.GetTexture()->GetSamples()));
+            }
+        }
+        return maxSamples;
+    }
+
     /* @INSERTION_POINT:FUNCTION_IMPLEMENTATION@ */
     const GLubyte* GetString(GLenum name) {
         static String vendorString;
@@ -679,12 +773,30 @@ namespace MobileGL::MG_Impl::GLImpl {
             return;
         case GL_MIN_FRAGMENT_INTERPOLATION_OFFSET:
         case GL_MAX_FRAGMENT_INTERPOLATION_OFFSET:
-        case GL_FRAGMENT_INTERPOLATION_OFFSET_BITS: {
+        case GL_FRAGMENT_INTERPOLATION_OFFSET_BITS:
+        // Same reason as the three above: the integer fallback would round the fraction to 0
+        // or 1 first, so a 0.25 sample-shading rate would answer GL_FALSE.
+        case GL_MIN_SAMPLE_SHADING_VALUE: {
             GLfloat value = 0.0f;
             GetFloatv(pname, &value);
             *params = value != 0.0f ? GL_TRUE : GL_FALSE;
             return;
         }
+        // Float-native state, so GL 4.6 core 2.2.2's "zero becomes FALSE, every other value
+        // becomes TRUE" has to be applied to the VALUE. Answering these through the integer getter
+        // below instead - which rounds - reported GL_FALSE for a perfectly non-zero level of 0.25,
+        // and every other float state in this function already reads through GetFloatv for exactly
+        // that reason.
+        case GL_PATCH_DEFAULT_OUTER_LEVEL:
+        case GL_PATCH_DEFAULT_INNER_LEVEL: {
+            const GLsizei componentCount = pname == GL_PATCH_DEFAULT_OUTER_LEVEL ? 4 : 2;
+            GLfloat levels[4] = {};
+            GetFloatv(pname, levels);
+            for (GLsizei i = 0; i < componentCount; ++i) {
+                params[i] = levels[i] != 0.0f ? GL_TRUE : GL_FALSE;
+            }
+            return;
+        }
         default:
             break;
         }
@@ -734,6 +846,22 @@ namespace MobileGL::MG_Impl::GLImpl {
             params[1] = depthRange.y();
             return;
         }
+        // glPatchParameterfv's two states. Float-native, so they are answered here rather than
+        // through the integer fallback below - which rounds, and would report 0 for a level of 0.5.
+        case GL_PATCH_DEFAULT_OUTER_LEVEL: {
+            const FloatVec4& outer = MG_State::pGLContext->GetPatchDefaultOuterLevel();
+            params[0] = outer.x();
+            params[1] = outer.y();
+            params[2] = outer.z();
+            params[3] = outer.w();
+            return;
+        }
+        case GL_PATCH_DEFAULT_INNER_LEVEL: {
+            const FloatVec2& inner = MG_State::pGLContext->GetPatchDefaultInnerLevel();
+            params[0] = inner.x();
+            params[1] = inner.y();
+            return;
+        }
         case GL_VIEWPORT_BOUNDS_RANGE: {
             const auto& dynamicParameters = MG_Backend::pActiveBackendObject->GetDynamicParameters();
             params[0] = dynamicParameters.ViewportBoundsRangeMin;
@@ -799,6 +927,11 @@ namespace MobileGL::MG_Impl::GLImpl {
         case GL_POLYGON_OFFSET_UNITS:
             params[0] = MG_State::pGLContext->GetPolygonOffsetUnits();
             return;
+        case GL_POLYGON_OFFSET_CLAMP:
+            // Float-native state, so it is answered here rather than through the integer
+            // fallback: glPolygonOffsetClamp(1, 1, 0.5) must read back as 0.5, not as 0.
+            params[0] = MG_State::pGLContext->GetPolygonOffsetClamp();
+            return;
         case GL_SMOOTH_LINE_WIDTH_RANGE: {
             const auto& dynamicParameters = MG_Backend::pActiveBackendObject->GetDynamicParameters();
             params[0] = dynamicParameters.SmoothLineWidthRangeMin;
@@ -814,6 +947,11 @@ namespace MobileGL::MG_Impl::GLImpl {
         case GL_SAMPLE_COVERAGE_VALUE:
             params[0] = MG_State::pGLContext->GetSampleCoverageValue();
             return;
+        case GL_MIN_SAMPLE_SHADING_VALUE:
+            // Float state, so it has to be answered here rather than through the integer
+            // fallback: glMinSampleShading(0.5) must read back as 0.5 and not as 0.
+            params[0] = MG_State::pGLContext->GetMinSampleShadingValue();
+            return;
         case GL_POINT_FADE_THRESHOLD_SIZE:
             // Float state: read it directly so the fractional part is not lost to the integer path.
             params[0] = MG_State::pGLContext->GetPointFadeThresholdSize();
@@ -1185,6 +1323,13 @@ namespace MobileGL::MG_Impl::GLImpl {
         }
 
         switch (pname) {
+        case GL_MAX_ELEMENT_INDEX:
+            // The largest value a GL_UNSIGNED_INT index may take. It has to be answered HERE and
+            // not left to the 32-bit fallback below: the conformance suite reads it with
+            // glGetInteger64v, and widening the saturated GLint would report INT32_MAX where the
+            // spec requires 2^32-1.
+            params[0] = 0xFFFFFFFFLL;
+            return;
         case GL_MAX_SHADER_STORAGE_BLOCK_SIZE:
             if (MG_Backend::pActiveBackendObject) {
                 params[0] = static_cast<GLint64>(
@@ -1221,12 +1366,17 @@ namespace MobileGL::MG_Impl::GLImpl {
         GLint ints[4] = {};
         GetIntegerv(pname, ints);
 
+        // GL 4.6 core 22.1 gives glGetInteger64v the same accepted-pname set as glGetIntegerv, so
+        // every pname the integer getter answers with several components owes them all here too.
+        // A pname that reaches the `default:` arm writes params[0] and leaves the caller's other
+        // components holding whatever they held, with no error to say so.
         switch (pname) {
         case GL_BLEND_COLOR:
         case GL_COLOR_CLEAR_VALUE:
         case GL_COLOR_WRITEMASK:
         case GL_SCISSOR_BOX:
         case GL_VIEWPORT:
+        case GL_PATCH_DEFAULT_OUTER_LEVEL:
             for (int i = 0; i < 4; ++i) {
                 params[i] = static_cast<GLint64>(ints[i]);
             }
@@ -1236,6 +1386,7 @@ namespace MobileGL::MG_Impl::GLImpl {
         case GL_MAX_VIEWPORT_DIMS:
         case GL_POINT_SIZE_RANGE:
         case GL_VIEWPORT_BOUNDS_RANGE:
+        case GL_PATCH_DEFAULT_INNER_LEVEL:
             params[0] = static_cast<GLint64>(ints[0]);
             params[1] = static_cast<GLint64>(ints[1]);
             return;
@@ -1267,6 +1418,7 @@ namespace MobileGL::MG_Impl::GLImpl {
         case GL_POINT_SIZE_RANGE:
         case GL_SMOOTH_LINE_WIDTH_RANGE:
         case GL_MAX_VIEWPORT_DIMS:
+        case GL_PATCH_DEFAULT_INNER_LEVEL:
             count = 2;
             break;
         case GL_BLEND_COLOR:
@@ -1274,6 +1426,7 @@ namespace MobileGL::MG_Impl::GLImpl {
         case GL_VIEWPORT:
         case GL_SCISSOR_BOX:
         case GL_COLOR_WRITEMASK:
+        case GL_PATCH_DEFAULT_OUTER_LEVEL:
             count = 4;
             break;
         default:
@@ -1313,6 +1466,15 @@ namespace MobileGL::MG_Impl::GLImpl {
                 *params = 0;
             return;
         }
+        // GL_TEXTURE_BUFFER_BINDING and GL_TEXTURE_BUFFER are the same token (0x8C2A): as a
+        // glGetIntegerv pname it asks which BUFFER object is bound to the buffer-texture target,
+        // not which texture is (that one is GL_TEXTURE_BINDING_BUFFER, handled by the texture-unit
+        // decoder above).
+        case GL_TEXTURE_BUFFER_BINDING: {
+            auto& obj = MG_State::pGLContext->GetBufferBindingSlot(BufferTarget::Texture).GetBoundObject();
+            *params = obj ? static_cast<GLint>(obj->GetExternalIndex()) : 0;
+            return;
+        }
         case GL_BLEND:
             *params = MG_State::pGLContext->IsCapabilityEnabled(CapabilityInput::Blend) ? GL_TRUE : GL_FALSE;
             return;
@@ -1368,6 +1530,16 @@ namespace MobileGL::MG_Impl::GLImpl {
             // this single case serves every getter flavor.
             *params = static_cast<GLint>(MG_State::pGLContext->GetClampReadColor());
             return;
+        // glClipControl's two state variables (GL 4.5 core table 23.7). They answer from the
+        // state the entry point records, which is what the conformance suite's initial-value and
+        // set-then-get cases read - the RASTERIZATION half of clip control is a separate,
+        // backend-side question and does not gate the query.
+        case GL_CLIP_ORIGIN:
+            *params = static_cast<GLint>(MG_State::pGLContext->GetClipOrigin());
+            return;
+        case GL_CLIP_DEPTH_MODE:
+            *params = static_cast<GLint>(MG_State::pGLContext->GetClipDepthMode());
+            return;
         case GL_COLOR_CLEAR_VALUE: {
             const FloatVec4& clearColor = MG_State::pGLContext->GetClearColor();
             params[0] = static_cast<GLint>(clearColor.x());
@@ -1427,19 +1599,21 @@ namespace MobileGL::MG_Impl::GLImpl {
                           : 0;
             return;
         case GL_MAX_DEBUG_GROUP_STACK_DEPTH:
-            // KHR_debug floors this at 64 even when the group entry points are stubs: the
-            // limit describes how deep glPushDebugGroup may nest, and 0 is not a legal answer.
+            // KHR_debug floors this at 64. It must agree with what GL_Debug.cpp actually enforces,
+            // or an application that nests to the reported limit would take a STACK_OVERFLOW.
             *params = kFrontendMaxDebugGroupStackDepth;
             return;
         case GL_MAX_DEBUG_MESSAGE_LENGTH:
-            *params = 1024; // debug-message entrypoints are stubbed, but KHR_debug requires a valid limit
+            *params = 1024; // agrees with GL_Debug.cpp's kMaxDebugMessageLength
             return;
         case GL_MAX_DEBUG_LOGGED_MESSAGES:
             // Size of the message log ring; KHR_debug requires at least 1.
             *params = kFrontendMaxDebugLoggedMessages;
             return;
         case GL_DEBUG_GROUP_STACK_DEPTH:
-            *params = 0; // debug-group entrypoints are stubbed
+            // The live depth, which is never 0: GL 4.6 core 20.6 creates the context with one
+            // group already on the stack, and that is the one glPopDebugGroup may not pop.
+            *params = GetDebugGroupStackDepth();
             return;
         case GL_CONTEXT_FLAGS: {
             *params = MG_State::pEGLContext ? MG_State::pEGLContext->GetCurrentContextFlags() : 0;
@@ -1654,6 +1828,9 @@ namespace MobileGL::MG_Impl::GLImpl {
         case GL_MAX_GEOMETRY_UNIFORM_COMPONENTS:
             *params = kFrontendMaxGeometryUniformComponents;
             return;
+        case GL_MAX_GEOMETRY_SHADER_INVOCATIONS:
+            *params = kFrontendMaxGeometryShaderInvocations;
+            return;
         case GL_MAX_IMAGE_SAMPLES:
             *params = 0; // multisampled image load/store is not exposed by the DirectGLES frontend
             return;
@@ -1707,6 +1884,59 @@ namespace MobileGL::MG_Impl::GLImpl {
             *params =
                 StageStorageBlockCount(&MG_Backend::DynamicBackendParameters::MaxTessEvaluationShaderStorageBlocks);
             return;
+        // The tessellation per-stage resource limits. Every one of these is ALSO a GLSL built-in
+        // constant that BuildTBuiltInResource expands, and the two must report the same number
+        // (KHR-GL45.limits.max_tess_* compares them directly) - which is why the values come from
+        // the shared block in MG_Util/ShaderTranspiler/Types.h rather than from literals here.
+        // They were the whole per-stage tess family: the table had been filled in only where the
+        // honest answer was zero (the atomic counters, the image uniforms) or where a driver
+        // query existed (GL_MAX_PATCH_VERTICES, GL_MAX_TESS_GEN_LEVEL), so every pname whose
+        // answer is a real resource count fell through to GL_INVALID_ENUM.
+        case GL_MAX_TESS_CONTROL_INPUT_COMPONENTS:
+            *params = static_cast<GLint>(MG_Util::ShaderTranspiler::MAX_TESS_CONTROL_INPUT_COMPONENTS);
+            return;
+        case GL_MAX_TESS_CONTROL_OUTPUT_COMPONENTS:
+            *params = static_cast<GLint>(MG_Util::ShaderTranspiler::MAX_TESS_CONTROL_OUTPUT_COMPONENTS);
+            return;
+        case GL_MAX_TESS_CONTROL_TOTAL_OUTPUT_COMPONENTS:
+            *params = static_cast<GLint>(MG_Util::ShaderTranspiler::MAX_TESS_CONTROL_TOTAL_OUTPUT_COMPONENTS);
+            return;
+        case GL_MAX_TESS_CONTROL_TEXTURE_IMAGE_UNITS:
+            *params = static_cast<GLint>(MG_Util::ShaderTranspiler::MAX_TESS_CONTROL_TEXTURE_IMAGE_UNITS);
+            return;
+        case GL_MAX_TESS_CONTROL_UNIFORM_COMPONENTS:
+            *params = static_cast<GLint>(MG_Util::ShaderTranspiler::MAX_TESS_CONTROL_UNIFORM_COMPONENTS);
+            return;
+        case GL_MAX_TESS_EVALUATION_INPUT_COMPONENTS:
+            *params = static_cast<GLint>(MG_Util::ShaderTranspiler::MAX_TESS_EVALUATION_INPUT_COMPONENTS);
+            return;
+        case GL_MAX_TESS_EVALUATION_OUTPUT_COMPONENTS:
+            *params = static_cast<GLint>(MG_Util::ShaderTranspiler::MAX_TESS_EVALUATION_OUTPUT_COMPONENTS);
+            return;
+        case GL_MAX_TESS_EVALUATION_TEXTURE_IMAGE_UNITS:
+            *params = static_cast<GLint>(MG_Util::ShaderTranspiler::MAX_TESS_EVALUATION_TEXTURE_IMAGE_UNITS);
+            return;
+        case GL_MAX_TESS_EVALUATION_UNIFORM_COMPONENTS:
+            *params = static_cast<GLint>(MG_Util::ShaderTranspiler::MAX_TESS_EVALUATION_UNIFORM_COMPONENTS);
+            return;
+        case GL_MAX_TESS_PATCH_COMPONENTS:
+            *params = static_cast<GLint>(MG_Util::ShaderTranspiler::MAX_TESS_PATCH_COMPONENTS);
+            return;
+        // Routed through the same clamp as every other per-stage block count so the
+        // MAX_UNIFORM_BUFFER_BINDINGS >= MAX_COMBINED_UNIFORM_BLOCKS >= per-stage ordering of
+        // GL 4.6 table 23.64 cannot be broken by the two families moving independently.
+        case GL_MAX_TESS_CONTROL_UNIFORM_BLOCKS:
+            *params = ClampUniformBlockCount(kFrontendMaxTessControlUniformBlocks);
+            return;
+        case GL_MAX_TESS_EVALUATION_UNIFORM_BLOCKS:
+            *params = ClampUniformBlockCount(kFrontendMaxTessEvaluationUniformBlocks);
+            return;
+        case GL_MAX_SUBROUTINES:
+            *params = kFrontendMaxSubroutines;
+            return;
+        case GL_MAX_SUBROUTINE_UNIFORM_LOCATIONS:
+            *params = kFrontendMaxSubroutineUniformLocations;
+            return;
         case GL_MAX_TEXTURE_LOD_BIAS:
             *params = 15; // TODO
             return;
@@ -1752,8 +1982,21 @@ namespace MobileGL::MG_Impl::GLImpl {
         case GL_NUM_PROGRAM_BINARY_FORMATS:
             *params = 0;
             return;
+        // GL_ARB_spirv_extensions / GL 4.6 core 22.2. An implementation that advertises no
+        // SPIR-V extension answers zero here, and glGetStringi(GL_SPIR_V_EXTENSIONS, i) is then
+        // never legally called - MobileGL runs the module through its own translation pipeline
+        // and relies on no SPIR-V extension to do it, so zero is the true answer rather than a
+        // placeholder.
+        case GL_NUM_SPIR_V_EXTENSIONS:
+            *params = 0;
+            return;
+        // GL_ARB_gl_spirv, core since 4.6: exactly one shader binary format, and the pair has to
+        // agree - an application sizes its GL_SHADER_BINARY_FORMATS array from the count.
         case GL_NUM_SHADER_BINARY_FORMATS:
-            *params = 0; // ShaderBinary entrypoints are stubbed
+            *params = 1;
+            return;
+        case GL_SHADER_BINARY_FORMATS:
+            *params = static_cast<GLint>(GL_SHADER_BINARY_FORMAT_SPIR_V);
             return;
         case GL_PACK_ALIGNMENT:
             *params = MG_State::pGLContext->GetPixelStoreParam(PixelStoreParam::PackAlignment);
@@ -1812,6 +2055,11 @@ namespace MobileGL::MG_Impl::GLImpl {
         case GL_PRIMITIVE_RESTART_INDEX:
             *params = static_cast<GLint>(MG_State::pGLContext->GetPrimitiveRestartIndex());
             return;
+        case GL_POLYGON_OFFSET_CLAMP:
+            // Float state (see GetFloatv); rounded to nearest for the integer query per GL 4.6
+            // core 22.1's float-to-integer rule.
+            *params = static_cast<GLint>(std::lround(MG_State::pGLContext->GetPolygonOffsetClamp()));
+            return;
         case GL_PROGRAM_BINARY_FORMATS:
             *params = 0; // program-binary entrypoints are stubbed
             return;
@@ -1897,6 +2145,13 @@ namespace MobileGL::MG_Impl::GLImpl {
         case GL_SAMPLE_MASK:
             *params = MG_State::pGLContext->IsCapabilityEnabled(CapabilityInput::SampleMask) ? GL_TRUE : GL_FALSE;
             return;
+        case GL_SAMPLE_SHADING:
+            *params = MG_State::pGLContext->IsCapabilityEnabled(CapabilityInput::SampleShading) ? GL_TRUE : GL_FALSE;
+            return;
+        case GL_MIN_SAMPLE_SHADING_VALUE:
+            // GL 4.6 core 22.2: a floating-point value queried as an integer rounds to nearest.
+            *params = static_cast<GLint>(std::lround(MG_State::pGLContext->GetMinSampleShadingValue()));
+            return;
         case GL_SAMPLE_MASK_VALUE:
             *params = static_cast<GLint>(MG_State::pGLContext->GetSampleMaskValue());
             return;
@@ -2115,7 +2370,12 @@ namespace MobileGL::MG_Impl::GLImpl {
             return;
         }
         case GL_MAX_ELEMENT_INDEX:
-            *params = 1024 * 1024; // TODO
+            // 64-bit state (see GetInteger64v); the 32-bit query saturates, per the GL
+            // state-query conversion rules - the same shape GL_MAX_SHADER_STORAGE_BLOCK_SIZE
+            // uses. The real answer is 2^32-1 because both backends draw with GL_UNSIGNED_INT
+            // indices and neither bounds an index value; the old `1024 * 1024` was a placeholder
+            // that no draw path ever consulted.
+            *params = INT32_MAX;
             return;
         case GL_CONTEXT_PROFILE_MASK:
             // Reports the requested context profile (EGL defaults 3.x contexts to core);
@@ -2171,8 +2431,12 @@ namespace MobileGL::MG_Impl::GLImpl {
             *params = dynamicParameters.MaxComputeTextureImageUnits;
             break;
         case GL_MAX_COMBINED_COMPUTE_UNIFORM_COMPONENTS:
+            // The CLAMPED block count, i.e. exactly what GL_MAX_COMPUTE_UNIFORM_BLOCKS answers.
+            // GL 4.6 table 23.64 defines this as the components reachable through the blocks a
+            // stage may declare, so deriving it from the raw backend number described 256 blocks
+            // an application is only ever allowed 84 of.
             *params = GetMaxCombinedUniformComponents(kFrontendMaxComputeUniformComponents,
-                                                      dynamicParameters.MaxComputeUniformBlocks,
+                                                      ClampUniformBlockCount(dynamicParameters.MaxComputeUniformBlocks),
                                                       dynamicParameters.MaxUniformBlockSize);
             break;
         case GL_MAX_COMPUTE_WORK_GROUP_INVOCATIONS:
@@ -2216,16 +2480,16 @@ namespace MobileGL::MG_Impl::GLImpl {
             *params = static_cast<GLint>(dynamicParameters.ViewportIndexProvokingVertex);
             break;
         case GL_MAX_COLOR_TEXTURE_SAMPLES:
-            *params = std::max(dynamicParameters.MaxColorTextureSamples, GetAdvertisedMaxSamples());
+            *params = GetAdvertisedColorTextureMaxSamples();
             break;
         case GL_MAX_COMBINED_FRAGMENT_UNIFORM_COMPONENTS:
             *params = GetMaxCombinedUniformComponents(kFrontendMaxFragmentUniformComponents,
-                                                      kFrontendMaxFragmentUniformBlocks,
+                                                      ClampUniformBlockCount(kFrontendMaxFragmentUniformBlocks),
                                                       dynamicParameters.MaxUniformBlockSize);
             break;
         case GL_MAX_COMBINED_GEOMETRY_UNIFORM_COMPONENTS:
             *params = GetMaxCombinedUniformComponents(kFrontendMaxGeometryUniformComponents,
-                                                      kFrontendMaxGeometryUniformBlocks,
+                                                      ClampUniformBlockCount(kFrontendMaxGeometryUniformBlocks),
                                                       dynamicParameters.MaxUniformBlockSize);
             break;
         case GL_MAX_GEOMETRY_OUTPUT_VERTICES:
@@ -2239,14 +2503,14 @@ namespace MobileGL::MG_Impl::GLImpl {
             break;
         case GL_MAX_COMBINED_VERTEX_UNIFORM_COMPONENTS:
             *params = GetMaxCombinedUniformComponents(kFrontendMaxVertexUniformComponents,
-                                                      kFrontendMaxVertexUniformBlocks,
+                                                      ClampUniformBlockCount(kFrontendMaxVertexUniformBlocks),
                                                       dynamicParameters.MaxUniformBlockSize);
             break;
         case GL_MAX_CUBE_MAP_TEXTURE_SIZE:
             *params = dynamicParameters.MaxCubeMapTextureSize;
             break;
         case GL_MAX_DEPTH_TEXTURE_SAMPLES:
-            *params = std::max(dynamicParameters.MaxDepthTextureSamples, GetAdvertisedMaxSamples());
+            *params = GetAdvertisedDepthTextureMaxSamples();
             break;
         case GL_MAX_FRAMEBUFFER_WIDTH:
             *params = dynamicParameters.MaxFramebufferWidth;
@@ -2273,7 +2537,7 @@ namespace MobileGL::MG_Impl::GLImpl {
             *params = dynamicParameters.MaxComputeImageUniforms;
             break;
         case GL_MAX_INTEGER_SAMPLES:
-            *params = std::max(dynamicParameters.MaxIntegerSamples, GetAdvertisedMaxSamples());
+            *params = GetAdvertisedIntegerMaxSamples();
             break;
         case GL_MAX_RENDERBUFFER_SIZE:
             *params = dynamicParameters.MaxRenderbufferSize;
@@ -2284,12 +2548,56 @@ namespace MobileGL::MG_Impl::GLImpl {
         case GL_PATCH_VERTICES:
             *params = static_cast<GLint>(MG_State::pGLContext->GetPatchVertices());
             break;
+        // Float state, so glGetIntegerv rounds it (GL 4.6 core 2.2.2) - the exact values come back
+        // through glGetFloatv. Answered here so glGetBooleanv, which delegates to this getter for
+        // everything its own switch does not handle, does not report INVALID_ENUM for them.
+        case GL_PATCH_DEFAULT_OUTER_LEVEL: {
+            const FloatVec4& outer = MG_State::pGLContext->GetPatchDefaultOuterLevel();
+            for (Uint i = 0; i < 4; ++i) params[i] = static_cast<GLint>(std::lround(outer[i]));
+            break;
+        }
+        case GL_PATCH_DEFAULT_INNER_LEVEL: {
+            const FloatVec2& inner = MG_State::pGLContext->GetPatchDefaultInnerLevel();
+            for (Uint i = 0; i < 2; ++i) params[i] = static_cast<GLint>(std::lround(inner[i]));
+            break;
+        }
+        // GL 4.6 core table 23.66: whether the primitive-restart index terminates a patch.
+        // GL_FALSE is a legal answer and the true one - neither backend cuts a patch short, and
+        // the DirectVulkan draw path relies on this staying false (it resolves primitive restart
+        // to "never" for a PATCH_LIST topology on the strength of it).
+        case GL_PRIMITIVE_RESTART_FOR_PATCHES_SUPPORTED:
+            *params = GL_FALSE;
+            break;
         case GL_MAX_PATCH_VERTICES:
             *params = dynamicParameters.MaxPatchVertices;
             break;
         case GL_MAX_TESS_GEN_LEVEL:
             *params = dynamicParameters.MaxTessGenLevel;
             break;
+        // Same helper, and so the same arithmetic, as every other GL_MAX_COMBINED_*_UNIFORM_
+        // COMPONENTS: default-block components + blocks * (block size / 4). It reproduces the
+        // conformance suite's own formula exactly, so the two cannot drift.
+        case GL_MAX_COMBINED_TESS_CONTROL_UNIFORM_COMPONENTS:
+            *params = GetMaxCombinedUniformComponents(
+                static_cast<GLint>(MG_Util::ShaderTranspiler::MAX_TESS_CONTROL_UNIFORM_COMPONENTS),
+                ClampUniformBlockCount(kFrontendMaxTessControlUniformBlocks), dynamicParameters.MaxUniformBlockSize);
+            break;
+        case GL_MAX_COMBINED_TESS_EVALUATION_UNIFORM_COMPONENTS:
+            *params = GetMaxCombinedUniformComponents(
+                static_cast<GLint>(MG_Util::ShaderTranspiler::MAX_TESS_EVALUATION_UNIFORM_COMPONENTS),
+                ClampUniformBlockCount(kFrontendMaxTessEvaluationUniformBlocks), dynamicParameters.MaxUniformBlockSize);
+            break;
+        // ARB_cull_distance. Backend-derived exactly like GL_MAX_CLIP_DISTANCES beside it, and
+        // for a stronger reason: a cull distance discards the whole primitive, so advertising
+        // eight the rasterizer cannot serve turns every culling draw into a silent no-op. Zero is
+        // the honest answer on a host with no cull-distance route, and the conformance suite then
+        // skips the functional cases instead of failing them deep inside a pixel comparison.
+        case GL_MAX_CULL_DISTANCES:
+            *params = dynamicParameters.MaxCullDistances;
+            break;
+        case GL_MAX_COMBINED_CLIP_AND_CULL_DISTANCES:
+            *params = dynamicParameters.MaxCombinedClipAndCullDistances;
+            break;
         case GL_MIN_PROGRAM_TEXTURE_GATHER_OFFSET:
             *params = dynamicParameters.MinProgramTextureGatherOffset;
             break;
@@ -2340,7 +2648,25 @@ namespace MobileGL::MG_Impl::GLImpl {
             *params = kFrontendMaxTransformFeedbackSeparateAttribs;
             break;
         case GL_MAX_VERTEX_STREAMS:
-            *params = 1;
+            // ONE, which is under the GL 4.5 core table 23.62 minimum of four and is a known,
+            // deliberate non-conformance. It was briefly raised to 4 on the theory that streams
+            // 1..3 could exist and be permanently empty; measuring that decision refuted it.
+            // Raising the limit un-gates two CTS cases per package across KHR-GL40..GL46 -
+            // transform_feedback.draw_xfb_stream_test (which stops being skipped) and
+            // transform_feedback3.multiple_streams (which stops reporting NotSupported) - and
+            // both then fail, because nothing in the shader pipeline supports layout(stream = N),
+            // EmitStreamVertex or EndStreamPrimitive, and because the query state machine tracks
+            // one active query per TARGET rather than per (target, stream). That is 14 new
+            // failures against 2 gained limits passes, and a 4 nothing can back is the
+            // advertised-caps lie with the sign flipped.
+            //
+            // The real fix is the feature, not the number: per-stream capture needs
+            // layout(stream = N) through the transpiler plus per-(target, stream) query slots,
+            // which DirectVulkan could back with VK_EXT_transform_feedback's geometryStreams and
+            // DirectGLES cannot back at all (ES has no vertex streams). Until that lands, one is
+            // the honest count and every stream-addressing entry point bounds itself by THIS
+            // query, so raising it later moves them all together.
+            *params = kFrontendMaxVertexStreams;
             break;
         case GL_TRANSFORM_FEEDBACK_ACTIVE:
             *params = MG_State::pGLContext->IsTransformFeedbackActive() ? 1 : 0;
@@ -2357,15 +2683,36 @@ namespace MobileGL::MG_Impl::GLImpl {
         case GL_MAX_TEXTURE_SIZE:
             *params = dynamicParameters.MaxTextureSize;
             break;
-        case GL_MAX_UNIFORM_BUFFER_BINDINGS:
+        case GL_MAX_UNIFORM_BUFFER_BINDINGS: {
             // Never advertise more bindings than the state layer's indexed-binding array can track
             // (BufferState::BufferBindingPointCount): glBindBufferBase rejects indices past that
             // capacity, and the GL CTS per-case state reset calls glBindBufferBase on every
-            // advertised index and expects no error. The floor equals the GL 3.3 core minimum
-            // (36), so the clamp never under-advertises.
+            // advertised index and expects no error. The floor is the GL 4.5 core minimum, and
+            // the array was widened to exactly it, so the two coincide by construction.
+            //
+            // WHY THE BACKEND'S OWN COUNT IS NOT THE CEILING HERE, unlike the shader-storage
+            // family. A GL uniform binding point is where an APPLICATION parks a buffer; it is
+            // not a driver binding point. Neither backend forwards it as one on the draw path:
+            // DirectGLES rebinds the blocks a program declares onto COMPACTED ES points
+            // (BindCurrentProgramWithResources maps block i to ES point i+1) and DirectVulkan
+            // resolves each block to a descriptor. So what the host driver's count bounds is how
+            // many blocks ONE PROGRAM may use, not how many points an application may bind.
+            //
+            // That per-program number is NOT GL_MAX_COMBINED_UNIFORM_BLOCKS (84, the six-stage
+            // sum): no single program can reach it. A graphics program is bounded by the five
+            // graphics stages' per-stage counts, 14 each, so 70 blocks plus the global UBO at ES
+            // point 0 = 71 - inside the ES 3.2 minimum of 72. A compute program is bounded by
+            // GL_MAX_COMPUTE_UNIFORM_BLOCKS, which on DirectGLES is the ES driver's own count
+            // (GL-scale, ~14) and on DirectVulkan is served from descriptors with no ES binding
+            // points involved. Raising any per-stage graphics count past 14 is what would break
+            // this, so that is the edit to check against the ES ceiling - not this one.
+            static_assert(static_cast<GLint>(MG_State::GLState::BufferBindingPointCount) >=
+                              kFrontendMinUniformBufferBindings,
+                          "the indexed-binding array must be able to hold every advertised uniform binding point");
             *params = std::clamp(dynamicParameters.MaxUniformBufferBindings, kFrontendMinUniformBufferBindings,
                                  static_cast<GLint>(MG_State::GLState::BufferBindingPointCount));
             break;
+        }
         case GL_MAX_UNIFORM_BLOCK_SIZE:
             *params = dynamicParameters.MaxUniformBlockSize;
             break;
@@ -2403,7 +2750,12 @@ namespace MobileGL::MG_Impl::GLImpl {
             *params = static_cast<GLint>(dynamicParameters.PointSizeGranularity);
             break;
         case GL_SHADER_STORAGE_BUFFER_OFFSET_ALIGNMENT:
-            *params = static_cast<GLint>(dynamicParameters.UniformBufferOffsetAlignment);
+            // The STORAGE alignment, which is its own limit - this used to answer with the
+            // uniform one. They differ on real hardware (Adreno 830: 32 uniform, 64 storage), and
+            // under-reporting it is silent: ValidateBindBufferRange accepts the offset, the ES
+            // driver accepts it too without raising an error, and the shader's writes then land
+            // at an address the application never bound.
+            *params = static_cast<GLint>(dynamicParameters.ShaderStorageBufferOffsetAlignment);
             break;
         case GL_SMOOTH_LINE_WIDTH_RANGE:
             params[0] = static_cast<GLint>(dynamicParameters.SmoothLineWidthRangeMin);

MobileGL/MG_Impl/GLImpl/Getter/GL_Getter.h

@@ -25,7 +25,24 @@ namespace MobileGL::MG_Impl::GLImpl {
     GLenum GetError();
     GLenum GetGraphicsResetStatus();
     // The GL_MAX_SAMPLES value MobileGL advertises, i.e. the driver's value floored to the GL
-    // core minimum. Frontend multisample validators have to honour this ceiling for every
-    // format, otherwise MobileGL rejects a sample count it advertised itself.
+    // core minimum of 4. This is the RENDERBUFFER ceiling; the three per-category texture
+    // ceilings below have a minimum of one and are reported as probed.
     GLint GetAdvertisedMaxSamples();
+    // Exactly what GL_MAX_COLOR_TEXTURE_SAMPLES / GL_MAX_DEPTH_TEXTURE_SAMPLES /
+    // GL_MAX_INTEGER_SAMPLES report: the probed backend limit floored at the GL 4.6 core minimum
+    // of ONE (table 23.53). Exported so the frontend's storage validation enforces exactly what
+    // the query promised - it used to floor both at 4 and then let the backend quietly
+    // under-allocate whatever the driver could not actually provide.
+    GLint GetAdvertisedColorTextureMaxSamples();
+    GLint GetAdvertisedDepthTextureMaxSamples();
+    GLint GetAdvertisedIntegerMaxSamples();
+    // What glGetIntegerv(GL_SAMPLES) answers for the CURRENT draw framebuffer: the largest sample
+    // count over its attachments, and 0 for a single-sample or default framebuffer (GL 4.6 core
+    // 9.2.3 / 22.2 - GL_SAMPLE_BUFFERS is 1 exactly when this is non-zero).
+    //
+    // Shared rather than duplicated because two callers need the identical number and disagreeing
+    // would be a silent bug: the query itself, and the draw path's write of the reserved
+    // gl_NumSamples stand-in - a shader comparing gl_NumSamples against glGetIntegerv(GL_SAMPLES)
+    // is exactly what the sample_variables CTS does.
+    GLint ResolveDrawFramebufferSampleCount();
 } // namespace MobileGL::MG_Impl::GLImpl

MobileGL/MG_Impl/GLImpl/Program/GL_Program.cpp

@@ -11,6 +11,8 @@
 #include "Config.h"
 #include <cmath>
 #include <limits>
+#include <set>
+#include <MG_Util/ShaderTranspiler/ShaderCompiler.h>
 #include <MG_Impl/GLImpl/VertexArray/Validators.h>
 #include <MG_State/GLState/Core.h>
 #include <MG_Util/Converters/GLToStr/GLEnumConverter.h>
@@ -30,10 +32,22 @@ namespace MobileGL::MG_Impl::GLImpl {
 
     static bool CheckShaderNameValidity(Uint shader) {
         if (shader == 0 || !MG_State::pGLContext->ValidateShaderName(shader)) {
+            // The mirror of CheckProgramNameValidity below, and for the same reason: programs and
+            // shaders are drawn from ONE name space (ProgramState hands both out of a single
+            // generator), so a name that exists but belongs to a PROGRAM is the wrong kind of
+            // object - GL 3.3 core 2.11.x makes that INVALID_OPERATION - while a name GL never
+            // handed out is INVALID_VALUE. This half of the split was missing, so every shader
+            // entry point handed a program name reported INVALID_VALUE; the conformance suite
+            // reads exactly that code back from glSpecializeShader.
+            const ErrorCode error = (shader != 0 && MG_State::pGLContext->ValidateProgramName(shader))
+                ? ErrorCode::InvalidOperation
+                : ErrorCode::InvalidValue;
             MG_State::pGLContext->RecordError(
-                ErrorCode::InvalidValue,
+                error,
                 MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", __func__,
-                                             std::to_string(shader) + " is not a valid name."));
+                                             std::to_string(shader) +
+                                                 (error == ErrorCode::InvalidOperation ? " is not a shader object."
+                                                                                       : " is not a valid name.")));
             return false;
         }
         return true;
@@ -245,6 +259,30 @@ namespace MobileGL::MG_Impl::GLImpl {
         return true;
     }
 
+    // GL 4.6 core 7.6.3: INVALID_VALUE when uniformBlockBinding >= MAX_UNIFORM_BUFFER_BINDINGS.
+    // The storage-block twin below has always had this check; the uniform one never did, and the
+    // value it stores is used as a RAW SUBSCRIPT into the state layer's fixed indexed-binding
+    // array on every draw and dispatch (DirectGLES's per-program UBO rebind, DirectVulkan's
+    // descriptor resolve, whose only guard is a MOBILEGL_ASSERT that compiles away in release).
+    // An out-of-range binding therefore did not merely go unreported - it read past the array and
+    // dereferenced whatever SharedPtr it found there.
+    bool ValidateUniformBlockBinding(GLuint binding) {
+        // Exactly what glGetIntegerv(GL_MAX_UNIFORM_BUFFER_BINDINGS) advertises: the state
+        // layer's array width, which the getter clamps to as well.
+        const SizeT maxBindingCount = MG_State::pGLContext->GetBufferBindingPointCount(BufferTarget::Uniform);
+        if (binding < maxBindingCount) {
+            return true;
+        }
+
+        MG_State::pGLContext->RecordError(
+            ErrorCode::InvalidValue,
+            MakeUnique<GenericErrorInfo>(
+                "MG_Impl/GLImpl", __func__,
+                std::format("Uniform block binding {} is not less than GL_MAX_UNIFORM_BUFFER_BINDINGS ({}).", binding,
+                            maxBindingCount)));
+        return false;
+    }
+
     bool ValidateShaderStorageBlockBinding(GLuint binding) {
         SizeT maxBindingCount = MG_State::pGLContext->GetBufferBindingPointCount(BufferTarget::ShaderStorage);
         if (MG_Backend::pActiveBackendObject) {
@@ -307,9 +345,195 @@ namespace MobileGL::MG_Impl::GLImpl {
     void CompileShader_State(GLuint shader) {
         auto& shaderObject = TryToGetShaderObject(shader);
         if (!shaderObject) return;
+        // ARB_gl_spirv: "INVALID_OPERATION is generated by CompileShader if shader has been
+        // associated with a SPIR-V binary". Such an object has no GLSL source to compile - it is
+        // waiting for glSpecializeShader, which is the operation that compiles it.
+        if (shaderObject->HasSpirvBinary()) {
+            MG_State::pGLContext->RecordError(
+                ErrorCode::InvalidOperation,
+                MakeUnique<GenericErrorInfo>(
+                    "MG_Impl/GLImpl", __func__,
+                    "shader " + std::to_string(shader) +
+                        " holds a SPIR-V binary; use glSpecializeShader instead of glCompileShader."));
+            return;
+        }
         shaderObject->Compile();
     }
 
+    // ---------------------------------------------------------------------------------------
+    // GL_ARB_gl_spirv
+    // ---------------------------------------------------------------------------------------
+
+    void ShaderBinary_State(GLsizei count, const GLuint* shaders, GLenum binaryformat, const void* binary,
+                            GLsizei length) {
+        if (count < 0 || length < 0) {
+            MG_State::pGLContext->RecordError(
+                ErrorCode::InvalidValue,
+                MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", __func__, "count and length must be non-negative."));
+            return;
+        }
+        // GL_NUM_SHADER_BINARY_FORMATS advertises exactly one format, so every other value is
+        // INVALID_ENUM (GL 4.6 core 7.2). This is the check that used to be missing entirely -
+        // the entry point was a silent stub, so an application handed a format nothing supports
+        // and was told nothing.
+        if (binaryformat != GL_SHADER_BINARY_FORMAT_SPIR_V) {
+            MG_State::pGLContext->RecordError(
+                ErrorCode::InvalidEnum,
+                MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", __func__,
+                                             "binaryformat must be GL_SHADER_BINARY_FORMAT_SPIR_V."));
+            return;
+        }
+        if (count == 0) return;
+        if (shaders == nullptr || (length > 0 && binary == nullptr)) {
+            MG_State::pGLContext->RecordError(
+                ErrorCode::InvalidValue,
+                MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", __func__, "shaders and binary must not be null."));
+            return;
+        }
+        // A SPIR-V module is a sequence of 32-bit words, so a length that is not a multiple of
+        // four cannot be one (ARB_gl_spirv makes this INVALID_VALUE).
+        if ((length % 4) != 0) {
+            MG_State::pGLContext->RecordError(
+                ErrorCode::InvalidValue,
+                MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", __func__,
+                                             "length must be a multiple of four for a SPIR-V module."));
+            return;
+        }
+
+        // EVERY name is validated before ANY of them is written: the entry point is all-or-
+        // nothing, and half-applying it would leave some objects holding a module the call was
+        // rejected for. The duplicate check is the extension's own ("INVALID_VALUE ... if the
+        // same shader object is specified more than once").
+        std::set<GLuint> seen;
+        for (GLsizei i = 0; i < count; ++i) {
+            if (!seen.insert(shaders[i]).second) {
+                MG_State::pGLContext->RecordError(
+                    ErrorCode::InvalidValue,
+                    MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", __func__,
+                                                 "shader " + std::to_string(shaders[i]) +
+                                                     " appears more than once in `shaders`."));
+                return;
+            }
+            if (!MG_State::pGLContext->ValidateShaderName(shaders[i])) {
+                MG_State::pGLContext->RecordError(
+                    ErrorCode::InvalidValue,
+                    MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", __func__,
+                                                 std::to_string(shaders[i]) + " is not the name of a shader object."));
+                return;
+            }
+        }
+
+        const SizeT wordCount = static_cast<SizeT>(length) / 4;
+        Vector<Uint32> module(wordCount);
+        if (wordCount != 0) {
+            Memcpy(module.data(), binary, static_cast<SizeT>(length));
+        }
+        // spirv-val here, not at glSpecializeShader: this is where the words arrive, and past it
+        // they reach SPIRV-Cross, which parses rather than validates. ARB_gl_spirv lets an
+        // implementation reject an invalid module at either call; rejecting at the earlier one
+        // means the application's error is reported next to the data that caused it.
+        if (const auto validated = MG_Util::ShaderTranspiler::ShaderCompiler::ValidateSpirvModule(module);
+            !validated) {
+            MGLOG_D("%s: rejected SPIR-V module: %s", __func__, validated.error().log.c_str());
+            MG_State::pGLContext->RecordError(
+                ErrorCode::InvalidValue,
+                MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", __func__, validated.error().log));
+            return;
+        }
+
+        for (GLsizei i = 0; i < count; ++i) {
+            auto& shaderObject = TryToGetShaderObject(shaders[i]);
+            if (!shaderObject) continue;
+            // A copy per object, not a shared buffer: each shader object may be specialized with
+            // different constants, and each specialization re-reads its own original words.
+            Vector<Uint32> perObject = module;
+            shaderObject->SetSpirvBinary(Move(perObject));
+        }
+    }
+
+    void SpecializeShader_State(GLuint shader, const GLchar* pEntryPoint, GLuint numSpecializationConstants,
+                                const GLuint* pConstantIndex, const GLuint* pConstantValue) {
+        auto& shaderObject = TryToGetShaderObject(shader);
+        if (!shaderObject) return;
+        if (!shaderObject->HasSpirvBinary()) {
+            MG_State::pGLContext->RecordError(
+                ErrorCode::InvalidOperation,
+                MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", __func__,
+                                             "shader " + std::to_string(shader) +
+                                                 " has no SPIR-V binary; call glShaderBinary first."));
+            return;
+        }
+        // ARB_gl_spirv: a shader that has already been specialized may not be specialized again
+        // until glShaderBinary re-associates a module with it.
+        if (shaderObject->HasBeenSpecialized()) {
+            MG_State::pGLContext->RecordError(
+                ErrorCode::InvalidOperation,
+                MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", __func__,
+                                             "shader " + std::to_string(shader) +
+                                                 " has already been specialized; re-associate its module with "
+                                                 "glShaderBinary before specializing it again."));
+            return;
+        }
+        // pEntryPoint names the entry point to specialize; there is no default. A null pointer
+        // cannot name one, and neither can the empty string.
+        if (pEntryPoint == nullptr || *pEntryPoint == '\0') {
+            MG_State::pGLContext->RecordError(
+                ErrorCode::InvalidValue,
+                MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", __func__, "pEntryPoint must name an entry point."));
+            return;
+        }
+        if (numSpecializationConstants > 0 && (pConstantIndex == nullptr || pConstantValue == nullptr)) {
+            MG_State::pGLContext->RecordError(
+                ErrorCode::InvalidValue,
+                MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", __func__,
+                                             "pConstantIndex and pConstantValue must not be null."));
+            return;
+        }
+        // "INVALID_VALUE is generated if any value in pConstantIndex is repeated" - checked before
+        // anything is applied, for the same all-or-nothing reason glShaderBinary checks its names
+        // up front.
+        Vector<Uint32> constantIds(pConstantIndex, pConstantIndex + numSpecializationConstants);
+        Vector<Uint32> constantValues(pConstantValue, pConstantValue + numSpecializationConstants);
+        {
+            std::set<Uint32> seen;
+            for (const Uint32 id : constantIds) {
+                if (seen.insert(id).second) continue;
+                MG_State::pGLContext->RecordError(
+                    ErrorCode::InvalidValue,
+                    MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", __func__,
+                                                 "constant index " + std::to_string(id) + " is repeated."));
+                return;
+            }
+        }
+
+        const String entryPoint(pEntryPoint);
+        const GLenum shaderType = MG_Util::ConvertShaderStageToGLEnum(shaderObject->GetShaderStage());
+        using SpecializationFailure = MG_Util::ShaderTranspiler::ShaderCompiler::SpecializationFailure;
+        SpecializationFailure failure = SpecializationFailure::None;
+        auto specialized = MG_Util::ShaderTranspiler::ShaderCompiler::SpecializeAndDecompileSpirvModule(
+            shaderObject->GetSpirvBinary(), shaderType, entryPoint, constantIds, constantValues, failure);
+        if (!specialized) {
+            MGLOG_D("%s: specialization failed for shader %u: %s", __func__, shader,
+                    specialized.error().log.c_str());
+            // The two conditions ARB_gl_spirv ENUMERATES are GL errors, and an erroring GL command
+            // must have no other effect - so the shader object is left exactly as it was rather
+            // than being pushed into a failed-compile state. Anything else is a genuine compile
+            // failure of a well-formed request, which the extension routes through COMPILE_STATUS
+            // and the info log exactly as glCompileShader does.
+            if (failure == SpecializationFailure::UnknownEntryPoint ||
+                failure == SpecializationFailure::UnknownConstantId) {
+                MG_State::pGLContext->RecordError(
+                    ErrorCode::InvalidValue,
+                    MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", __func__, specialized.error().log));
+                return;
+            }
+            shaderObject->RecordSpecializationFailure(String(specialized.error().log));
+            return;
+        }
+        shaderObject->SpecializeFromSpirv(Move(specialized.value().glsl), Move(specialized.value().xfbVaryings),
+                                          specialized.value().xfbBufferMode);
+    }
+
     // glMaxShaderCompilerThreadsKHR / glMaxShaderCompilerThreadsARB - one implementation,
     // because GL_KHR_parallel_shader_compile and GL_ARB_parallel_shader_compile define the
     // same entry point with the same semantics and GetProcAddress.cpp maps both spellings.
@@ -744,12 +968,77 @@ namespace MobileGL::MG_Impl::GLImpl {
             *params = programObject->GetBinaryRetrievableHint() ? GL_TRUE : GL_FALSE;
             break;
         case GL_PROGRAM_SEPARABLE:
-            *params = programObject->GetSeparable() ? GL_TRUE : GL_FALSE;
+            // The LATCHED flag, not the live one: glProgramParameteri's write takes effect at the
+            // next link (GL 4.6 core 7.3), so a program told to be separable and then never
+            // linked still reports GL_FALSE.
+            *params = programObject->GetLinkedSeparable() ? GL_TRUE : GL_FALSE;
             break;
 
+        // The geometry and tessellation link properties (GL 4.6 core table 23.35). Same shape as
+        // GL_COMPUTE_WORK_GROUP_SIZE above, and for the same reason: "a linked program object
+        // with a geometry shader" is one whose EXECUTABLE has the stage, so an
+        // attached-but-not-yet-linked shader must give INVALID_OPERATION rather than the previous
+        // link's value. The geometry three used to be listed here only to fall through into the
+        // INVALID_ENUM default, and the tessellation five were not listed at all.
         case GL_GEOMETRY_VERTICES_OUT:
         case GL_GEOMETRY_INPUT_TYPE:
         case GL_GEOMETRY_OUTPUT_TYPE:
+        case GL_GEOMETRY_SHADER_INVOCATIONS: {
+            if (!programObject->GetLinkStatus() || !programObject->HasLinkedShaderStage(ShaderStage::Geometry)) {
+                MG_State::pGLContext->RecordError(
+                    ErrorCode::InvalidOperation,
+                    MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", __func__,
+                                                 std::to_string(program) +
+                                                     " is not a linked program object with a geometry shader."));
+                return;
+            }
+            switch (pname) {
+            case GL_GEOMETRY_VERTICES_OUT: *params = programObject->GetGeometryVerticesOut(); break;
+            case GL_GEOMETRY_INPUT_TYPE: *params = static_cast<GLint>(programObject->GetGeometryInputType()); break;
+            case GL_GEOMETRY_OUTPUT_TYPE: *params = static_cast<GLint>(programObject->GetGeometryOutputType()); break;
+            default: *params = programObject->GetGeometryShaderInvocations(); break;
+            }
+            MGLOG_D("%s: %s = %d", __func__, MG_Util::ConvertGLEnumToString(pname).c_str(), *params);
+            break;
+        }
+        case GL_TESS_CONTROL_OUTPUT_VERTICES: {
+            if (!programObject->GetLinkStatus() || !programObject->HasLinkedShaderStage(ShaderStage::TessControl)) {
+                MG_State::pGLContext->RecordError(
+                    ErrorCode::InvalidOperation,
+                    MakeUnique<GenericErrorInfo>(
+                        "MG_Impl/GLImpl", __func__,
+                        std::to_string(program) +
+                            " is not a linked program object with a tessellation control shader."));
+                return;
+            }
+            *params = programObject->GetTessControlOutputVertices();
+            MGLOG_D("%s: %s = %d", __func__, MG_Util::ConvertGLEnumToString(pname).c_str(), *params);
+            break;
+        }
+        case GL_TESS_GEN_MODE:
+        case GL_TESS_GEN_SPACING:
+        case GL_TESS_GEN_VERTEX_ORDER:
+        case GL_TESS_GEN_POINT_MODE: {
+            if (!programObject->GetLinkStatus() || !programObject->HasLinkedShaderStage(ShaderStage::TessEval)) {
+                MG_State::pGLContext->RecordError(
+                    ErrorCode::InvalidOperation,
+                    MakeUnique<GenericErrorInfo>(
+                        "MG_Impl/GLImpl", __func__,
+                        std::to_string(program) +
+                            " is not a linked program object with a tessellation evaluation shader."));
+                return;
+            }
+            switch (pname) {
+            case GL_TESS_GEN_MODE: *params = static_cast<GLint>(programObject->GetTessGenMode()); break;
+            case GL_TESS_GEN_SPACING: *params = static_cast<GLint>(programObject->GetTessGenSpacing()); break;
+            case GL_TESS_GEN_VERTEX_ORDER:
+                *params = static_cast<GLint>(programObject->GetTessGenVertexOrder());
+                break;
+            default: *params = programObject->GetTessGenPointMode() ? GL_TRUE : GL_FALSE; break;
+            }
+            MGLOG_D("%s: %s = %d", __func__, MG_Util::ConvertGLEnumToString(pname).c_str(), *params);
+            break;
+        }
         default:
             MGLOG_D("%s: %s", __func__, MG_Util::ConvertGLEnumToString(pname).c_str());
             MG_State::pGLContext->RecordError(
@@ -811,8 +1100,19 @@ namespace MobileGL::MG_Impl::GLImpl {
             }
             *params = shaderObject->GetInfoLog().empty() ? 0 : (GLint)shaderObject->GetInfoLog().length() + 1;
             break;
-        case GL_SHADER_SOURCE_LENGTH:
-            *params = shaderObject->GetShaderSource().empty() ? 0 : (GLint)shaderObject->GetShaderSource().length() + 1;
+        case GL_SHADER_SOURCE_LENGTH: {
+            // The APPLICATION's source, which is empty for a shader that came from glShaderBinary -
+            // see ShaderObject::GetApplicationShaderSource.
+            const auto& source = shaderObject->GetApplicationShaderSource();
+            *params = source.empty() ? 0 : (GLint)source.length() + 1;
+            break;
+        }
+        // GL_ARB_gl_spirv. GL_SPIR_V_BINARY and GL_SPIR_V_BINARY_ARB are the same token: TRUE
+        // while the object stands for an application-supplied module. It is the FIRST thing the
+        // conformance suite asks after glShaderBinary, and it used to fall into the terminal
+        // default arm below and take the whole test with it.
+        case GL_SPIR_V_BINARY:
+            *params = shaderObject->HasSpirvBinary() ? GL_TRUE : GL_FALSE;
             break;
         // GL_KHR_parallel_shader_compile. THIS CASE MUST NOT JOIN - see the identical case in
         // GetProgramiv_State. GL_COMPILE_STATUS two cases up deliberately DOES join (it has
@@ -858,13 +1158,23 @@ namespace MobileGL::MG_Impl::GLImpl {
         auto& shaderObject = TryToGetShaderObject(shader);
         if (!shaderObject) return;
 
-        auto& src = shaderObject->GetShaderSource();
+        auto& src = shaderObject->GetApplicationShaderSource();
         CopyStr(bufSize, length, source, src.c_str(), (GLsizei)src.length());
     }
 
     GLint GetUniformLocation_State(GLuint program, const GLchar* name) {
         auto& programObject = TryToGetProgramObject(program);
         if (!programObject) return -1;
+        // GL 4.6 core 7.6: "INVALID_OPERATION is generated if program has not been successfully
+        // linked". Answering -1 silently is not the same thing - the conformance suite reads the
+        // error, not the location.
+        if (!programObject->GetLinkStatus()) {
+            MG_State::pGLContext->RecordError(
+                ErrorCode::InvalidOperation,
+                MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", __func__,
+                                             "program " + std::to_string(program) + " is not linked."));
+            return -1;
+        }
         auto loc = programObject->GetUniformLocation(name);
         MGLOG_D("%s: loc %02d = %s", __func__, loc, name);
         return loc;
@@ -1277,11 +1587,13 @@ namespace MobileGL::MG_Impl::GLImpl {
 
     template <GLsizei ItemCount, typename T>
     void ProgramUniformv_State(GLuint program, GLint location, GLsizei count, T* value) {
-        if (location == -1) return;
-
         auto& programObject = TryToGetProgramObject(program);
         if (!programObject) return;
 
+        // The link check comes BEFORE the location == -1 early-out, not after. GL 4.6 core 7.6
+        // makes an unlinked program INVALID_OPERATION regardless of the location, and -1 is
+        // exactly the location an application holds after glGetUniformLocation on such a program -
+        // so checking -1 first swallowed the very case the rule exists for.
         if (!programObject->GetLinkStatus()) {
             MG_State::pGLContext->RecordError(
                 ErrorCode::InvalidOperation,
@@ -1289,6 +1601,10 @@ namespace MobileGL::MG_Impl::GLImpl {
                                              "program " + std::to_string(program) + " is not linked."));
             return;
         }
+        // "If location is equal to -1, the data passed in will be silently ignored and the
+        // specified uniform variable will not be changed" - after the program itself has been
+        // found acceptable.
+        if (location == -1) return;
 
         for (GLint offset = 0; offset < count; offset++) {
             if (offset > 0 && !programObject->UniformLocationsAliasSameUniform(location, location + offset)) {
@@ -1699,8 +2015,6 @@ namespace MobileGL::MG_Impl::GLImpl {
 
     void ProgramUniformMatrix2fv_State(GLuint program, GLint location, GLsizei count, GLboolean transpose,
                                        const GLfloat* value) {
-        if (location == -1) return;
-
         auto& programObject = TryToGetProgramObject(program);
         if (!programObject) return;
 
@@ -1712,14 +2026,14 @@ namespace MobileGL::MG_Impl::GLImpl {
             return;
         }
 
+        if (location == -1) return;
+
         UniformMatrixfv_Object(*programObject, __func__, location, count, transpose, value, 2, 2,
                                "program " + std::to_string(program));
     }
 
     void ProgramUniformMatrix3fv_State(GLuint program, GLint location, GLsizei count, GLboolean transpose,
                                        const GLfloat* value) {
-        if (location == -1) return;
-
         auto& programObject = TryToGetProgramObject(program);
         if (!programObject) return;
 
@@ -1731,6 +2045,8 @@ namespace MobileGL::MG_Impl::GLImpl {
             return;
         }
 
+        if (location == -1) return;
+
         for (GLint i = 0; i < count; i++) {
             if (i > 0 && !programObject->UniformLocationsAliasSameUniform(location, location + i)) {
                 // Values for elements beyond the end of the uniform array are ignored.
@@ -1756,8 +2072,6 @@ namespace MobileGL::MG_Impl::GLImpl {
 
     void ProgramUniformMatrix4fv_State(GLuint program, GLint location, GLsizei count, GLboolean transpose,
                                        const GLfloat* value) {
-        if (location == -1) return;
-
         auto& programObject = TryToGetProgramObject(program);
         if (!programObject) return;
 
@@ -1769,6 +2083,8 @@ namespace MobileGL::MG_Impl::GLImpl {
             return;
         }
 
+        if (location == -1) return;
+
         for (GLint i = 0; i < count; i++) {
             if (i > 0 && !programObject->UniformLocationsAliasSameUniform(location, location + i)) {
                 // Values for elements beyond the end of the uniform array are ignored.
@@ -1790,8 +2106,6 @@ namespace MobileGL::MG_Impl::GLImpl {
 
     void ProgramUniformMatrixNonSquarefv_State(const char* caller, GLuint program, GLint location, GLsizei count,
                                                GLboolean transpose, const GLfloat* value, Int columns, Int rows) {
-        if (location == -1) return;
-
         auto& programObject = TryToGetProgramObject(program);
         if (!programObject) return;
 
@@ -1803,6 +2117,8 @@ namespace MobileGL::MG_Impl::GLImpl {
             return;
         }
 
+        if (location == -1) return;
+
         UniformMatrixfv_Object(*programObject, caller, location, count, transpose, value, columns, rows,
                                "program " + std::to_string(program));
     }
@@ -1836,6 +2152,7 @@ namespace MobileGL::MG_Impl::GLImpl {
                                              "Program object" + std::to_string(program) + " that has been linked."));
             return;
         }
+        if (!ValidateUniformBlockBinding(uniformBlockBinding)) return;
         if (!programObject->IsActiveGlUniformBlock(uniformBlockIndex)) {
             MG_State::pGLContext->RecordError(
                 ErrorCode::InvalidValue,
@@ -2083,6 +2400,15 @@ namespace MobileGL::MG_Impl::GLImpl {
         BindAttribLocation_State(program, index, name);
     }
 
+    void ShaderBinary(GLsizei count, const GLuint* shaders, GLenum binaryformat, const void* binary, GLsizei length) {
+        ShaderBinary_State(count, shaders, binaryformat, binary, length);
+    }
+
+    void SpecializeShader(GLuint shader, const GLchar* pEntryPoint, GLuint numSpecializationConstants,
+                          const GLuint* pConstantIndex, const GLuint* pConstantValue) {
+        SpecializeShader_State(shader, pEntryPoint, numSpecializationConstants, pConstantIndex, pConstantValue);
+    }
+
     void CompileShader(GLuint shader) {
         CompileShader_State(shader);
     }
@@ -2342,7 +2668,6 @@ namespace MobileGL::MG_Impl::GLImpl {
 
     void ProgramUniformMatrix2dv(GLuint program, GLint location, GLsizei count, GLboolean transpose,
                                       const GLdouble* value) {
-        if (location == -1) return;
         auto& programObject = TryToGetProgramObject(program);
         if (!programObject) return;
         if (!programObject->GetLinkStatus()) {
@@ -2352,6 +2677,7 @@ namespace MobileGL::MG_Impl::GLImpl {
                                              "program " + std::to_string(program) + " is not linked."));
             return;
         }
+        if (location == -1) return;
         UniformMatrixdv_Object(*programObject, location, count, transpose, value, 2, 2);
     }
     void UniformMatrix3dv(GLint location, GLsizei count, GLboolean transpose, const GLdouble* value) {
@@ -2368,7 +2694,6 @@ namespace MobileGL::MG_Impl::GLImpl {
 
     void ProgramUniformMatrix3dv(GLuint program, GLint location, GLsizei count, GLboolean transpose,
                                       const GLdouble* value) {
-        if (location == -1) return;
         auto& programObject = TryToGetProgramObject(program);
         if (!programObject) return;
         if (!programObject->GetLinkStatus()) {
@@ -2378,6 +2703,7 @@ namespace MobileGL::MG_Impl::GLImpl {
                                              "program " + std::to_string(program) + " is not linked."));
             return;
         }
+        if (location == -1) return;
         UniformMatrixdv_Object(*programObject, location, count, transpose, value, 3, 3);
     }
     void UniformMatrix4dv(GLint location, GLsizei count, GLboolean transpose, const GLdouble* value) {
@@ -2394,7 +2720,6 @@ namespace MobileGL::MG_Impl::GLImpl {
 
     void ProgramUniformMatrix4dv(GLuint program, GLint location, GLsizei count, GLboolean transpose,
                                       const GLdouble* value) {
-        if (location == -1) return;
         auto& programObject = TryToGetProgramObject(program);
         if (!programObject) return;
         if (!programObject->GetLinkStatus()) {
@@ -2404,6 +2729,7 @@ namespace MobileGL::MG_Impl::GLImpl {
                                              "program " + std::to_string(program) + " is not linked."));
             return;
         }
+        if (location == -1) return;
         UniformMatrixdv_Object(*programObject, location, count, transpose, value, 4, 4);
     }
     void UniformMatrix2x3dv(GLint location, GLsizei count, GLboolean transpose, const GLdouble* value) {
@@ -2420,7 +2746,6 @@ namespace MobileGL::MG_Impl::GLImpl {
 
     void ProgramUniformMatrix2x3dv(GLuint program, GLint location, GLsizei count, GLboolean transpose,
                                       const GLdouble* value) {
-        if (location == -1) return;
         auto& programObject = TryToGetProgramObject(program);
         if (!programObject) return;
         if (!programObject->GetLinkStatus()) {
@@ -2430,6 +2755,7 @@ namespace MobileGL::MG_Impl::GLImpl {
                                              "program " + std::to_string(program) + " is not linked."));
             return;
         }
+        if (location == -1) return;
         UniformMatrixdv_Object(*programObject, location, count, transpose, value, 2, 3);
     }
     void UniformMatrix2x4dv(GLint location, GLsizei count, GLboolean transpose, const GLdouble* value) {
@@ -2446,7 +2772,6 @@ namespace MobileGL::MG_Impl::GLImpl {
 
     void ProgramUniformMatrix2x4dv(GLuint program, GLint location, GLsizei count, GLboolean transpose,
                                       const GLdouble* value) {
-        if (location == -1) return;
         auto& programObject = TryToGetProgramObject(program);
         if (!programObject) return;
         if (!programObject->GetLinkStatus()) {
@@ -2456,6 +2781,7 @@ namespace MobileGL::MG_Impl::GLImpl {
                                              "program " + std::to_string(program) + " is not linked."));
             return;
         }
+        if (location == -1) return;
         UniformMatrixdv_Object(*programObject, location, count, transpose, value, 2, 4);
     }
     void UniformMatrix3x2dv(GLint location, GLsizei count, GLboolean transpose, const GLdouble* value) {
@@ -2472,7 +2798,6 @@ namespace MobileGL::MG_Impl::GLImpl {
 
     void ProgramUniformMatrix3x2dv(GLuint program, GLint location, GLsizei count, GLboolean transpose,
                                       const GLdouble* value) {
-        if (location == -1) return;
         auto& programObject = TryToGetProgramObject(program);
         if (!programObject) return;
         if (!programObject->GetLinkStatus()) {
@@ -2482,6 +2807,7 @@ namespace MobileGL::MG_Impl::GLImpl {
                                              "program " + std::to_string(program) + " is not linked."));
             return;
         }
+        if (location == -1) return;
         UniformMatrixdv_Object(*programObject, location, count, transpose, value, 3, 2);
     }
     void UniformMatrix3x4dv(GLint location, GLsizei count, GLboolean transpose, const GLdouble* value) {
@@ -2498,7 +2824,6 @@ namespace MobileGL::MG_Impl::GLImpl {
 
     void ProgramUniformMatrix3x4dv(GLuint program, GLint location, GLsizei count, GLboolean transpose,
                                       const GLdouble* value) {
-        if (location == -1) return;
         auto& programObject = TryToGetProgramObject(program);
         if (!programObject) return;
         if (!programObject->GetLinkStatus()) {
@@ -2508,6 +2833,7 @@ namespace MobileGL::MG_Impl::GLImpl {
                                              "program " + std::to_string(program) + " is not linked."));
             return;
         }
+        if (location == -1) return;
         UniformMatrixdv_Object(*programObject, location, count, transpose, value, 3, 4);
     }
     void UniformMatrix4x2dv(GLint location, GLsizei count, GLboolean transpose, const GLdouble* value) {
@@ -2524,7 +2850,6 @@ namespace MobileGL::MG_Impl::GLImpl {
 
     void ProgramUniformMatrix4x2dv(GLuint program, GLint location, GLsizei count, GLboolean transpose,
                                       const GLdouble* value) {
-        if (location == -1) return;
         auto& programObject = TryToGetProgramObject(program);
         if (!programObject) return;
         if (!programObject->GetLinkStatus()) {
@@ -2534,6 +2859,7 @@ namespace MobileGL::MG_Impl::GLImpl {
                                              "program " + std::to_string(program) + " is not linked."));
             return;
         }
+        if (location == -1) return;
         UniformMatrixdv_Object(*programObject, location, count, transpose, value, 4, 2);
     }
     void UniformMatrix4x3dv(GLint location, GLsizei count, GLboolean transpose, const GLdouble* value) {
@@ -2550,7 +2876,6 @@ namespace MobileGL::MG_Impl::GLImpl {
 
     void ProgramUniformMatrix4x3dv(GLuint program, GLint location, GLsizei count, GLboolean transpose,
                                       const GLdouble* value) {
-        if (location == -1) return;
         auto& programObject = TryToGetProgramObject(program);
         if (!programObject) return;
         if (!programObject->GetLinkStatus()) {
@@ -2560,6 +2885,7 @@ namespace MobileGL::MG_Impl::GLImpl {
                                              "program " + std::to_string(program) + " is not linked."));
             return;
         }
+        if (location == -1) return;
         UniformMatrixdv_Object(*programObject, location, count, transpose, value, 4, 3);
     }
     void GetUniformdv(GLuint program, GLint location, GLdouble* params) {

MobileGL/MG_Impl/GLImpl/Program/GL_Program.h

@@ -13,6 +13,12 @@ namespace MobileGL::MG_Impl::GLImpl {
     void AttachShader(GLuint program, GLuint shader);
     void BindAttribLocation(GLuint program, GLuint index, const GLchar* name);
     void CompileShader(GLuint shader);
+    // GL_ARB_gl_spirv, core since 4.6. The pair is a two-step operation: glShaderBinary attaches
+    // the module to one or more shader objects, glSpecializeShader names its entry point and
+    // supplies its specialization constants and is what actually compiles them.
+    void ShaderBinary(GLsizei count, const GLuint* shaders, GLenum binaryformat, const void* binary, GLsizei length);
+    void SpecializeShader(GLuint shader, const GLchar* pEntryPoint, GLuint numSpecializationConstants,
+                          const GLuint* pConstantIndex, const GLuint* pConstantValue);
     GLuint CreateProgram(void);
     GLuint CreateShader(GLenum type);
     void DeleteProgram(GLuint program);

MobileGL/MG_Impl/GLImpl/Program/GL_ProgramPipeline.cpp

@@ -192,6 +192,15 @@ namespace MobileGL::MG_Impl::GLImpl {
                                     std::format("Program {} has not been linked successfully.", program));
                 return;
             }
+            // GL 4.6 core 7.4: "INVALID_OPERATION is generated if program was not linked with its
+            // PROGRAM_SEPARABLE status set". The LATCHED flag is the one that decides - a program
+            // whose live flag was cleared after a separable link is still a legal stage, and a
+            // program whose live flag was set after a non-separable link is not.
+            if (!programObject->GetLinkedSeparable()) {
+                RecordPipelineError(ErrorCode::InvalidOperation, __func__,
+                                    std::format("Program {} was not linked as a separable program.", program));
+                return;
+            }
         }
 
         const GLbitfield selected = stages == GL_ALL_SHADER_BITS ? kAllStageBits : stages;

MobileGL/MG_Impl/GLImpl/Query/GL_Query.cpp

@@ -59,6 +59,37 @@ namespace MobileGL::MG_Impl::GLImpl {
         GLuint g_activePrimitivesGeneratedQueryId = 0;
         // Id of the query active on GL_SAMPLES_PASSED (0 = none).
         GLuint g_activeSamplesPassedQueryId = 0;
+        // Ids of the queries active on the GL_ARB_pipeline_statistics_query targets, one slot per
+        // target (0 = none). A map rather than a field per target: the eleven behave identically
+        // and none of them has any state beyond "which object is counting".
+        UnorderedMap<GLenum, GLuint> g_activePipelineStatisticsQueryIds;
+
+        // The eleven pipeline-statistics counters (GL 4.6 core table 4.3 / ARB_pipeline_statistics_query).
+        // A 4.6 core context has to ACCEPT all of them at glBeginQuery - the extension is core
+        // since 4.6 and there is no query by which an application could learn otherwise before
+        // calling. MobileGL instruments none of them, and says so the way GL 4.6 core 4.2.1
+        // provides for: GL_QUERY_COUNTER_BITS answers zero for these targets, which is the
+        // spec's own signal that the counter is unsupported and its results indeterminate. That
+        // is an honest zero, not an advertised capability - the alternative, GL_INVALID_ENUM on a
+        // core entry point, is both non-conformant AND less informative.
+        Bool IsPipelineStatisticsQueryTarget(GLenum target) {
+            switch (target) {
+            case GL_VERTICES_SUBMITTED:
+            case GL_PRIMITIVES_SUBMITTED:
+            case GL_VERTEX_SHADER_INVOCATIONS:
+            case GL_TESS_CONTROL_SHADER_PATCHES:
+            case GL_TESS_EVALUATION_SHADER_INVOCATIONS:
+            case GL_GEOMETRY_SHADER_INVOCATIONS:
+            case GL_GEOMETRY_SHADER_PRIMITIVES_EMITTED:
+            case GL_FRAGMENT_SHADER_INVOCATIONS:
+            case GL_COMPUTE_SHADER_INVOCATIONS:
+            case GL_CLIPPING_INPUT_PRIMITIVES:
+            case GL_CLIPPING_OUTPUT_PRIMITIVES:
+                return true;
+            default:
+                return false;
+            }
+        }
 
         Bool TimerQueryDisabled() {
             return MG_Config::Features.DisableTimerQuery;
@@ -370,6 +401,9 @@ namespace MobileGL::MG_Impl::GLImpl {
                     }
                     queryObject->active = false;
                     g_activeSamplesPassedQueryId = 0;
+                } else if (IsPipelineStatisticsQueryTarget(queryObject->target)) {
+                    queryObject->active = false;
+                    g_activePipelineStatisticsQueryIds[queryObject->target] = 0;
                 } else if (queryObject->target == GL_TRANSFORM_FEEDBACK_PRIMITIVES_WRITTEN ||
                            queryObject->target == GL_PRIMITIVES_GENERATED) {
                     queryObject->active = false;
@@ -410,7 +444,9 @@ namespace MobileGL::MG_Impl::GLImpl {
             (target == GL_SAMPLES_PASSED || target == GL_ANY_SAMPLES_PASSED ||
              target == GL_ANY_SAMPLES_PASSED_CONSERVATIVE) &&
             MG_Backend::gBackendFunctionsTable.GL.BeginOcclusionQuery != nullptr;
-        if (target != GL_TIME_ELAPSED && !isTransformFeedbackQuery && !isOcclusionQuery) {
+        const Bool isPipelineStatisticsQuery = IsPipelineStatisticsQueryTarget(target);
+        if (target != GL_TIME_ELAPSED && !isTransformFeedbackQuery && !isOcclusionQuery &&
+            !isPipelineStatisticsQuery) {
             // GL_TIMESTAMP is not a valid BeginQuery target; the occlusion targets
             // need backend support.
             RecordQueryError(ErrorCode::InvalidEnum, __FUNCTION__, "Query target is not supported.");
@@ -426,10 +462,12 @@ namespace MobileGL::MG_Impl::GLImpl {
             RecordQueryError(ErrorCode::InvalidOperation, __FUNCTION__, "Query object does not exist.");
             return;
         }
-        GLuint& activeQueryId = isTransformFeedbackQuery
-            ? (target == GL_TRANSFORM_FEEDBACK_PRIMITIVES_WRITTEN ? g_activePrimitivesWrittenQueryId
-                                                                  : g_activePrimitivesGeneratedQueryId)
-            : (isOcclusionQuery ? g_activeSamplesPassedQueryId : g_activeTimeElapsedQueryId);
+        GLuint& activeQueryId = isPipelineStatisticsQuery
+            ? g_activePipelineStatisticsQueryIds[target]
+            : (isTransformFeedbackQuery
+                   ? (target == GL_TRANSFORM_FEEDBACK_PRIMITIVES_WRITTEN ? g_activePrimitivesWrittenQueryId
+                                                                         : g_activePrimitivesGeneratedQueryId)
+                   : (isOcclusionQuery ? g_activeSamplesPassedQueryId : g_activeTimeElapsedQueryId));
         if (activeQueryId != 0) {
             RecordQueryError(ErrorCode::InvalidOperation, __FUNCTION__,
                              "A query is already active on this target.");
@@ -448,7 +486,11 @@ namespace MobileGL::MG_Impl::GLImpl {
         ResetQueryObjectLocked(queryObject); // discard any previous result
         queryObject->target = target;
         queryObject->active = true;
-        if (isTransformFeedbackQuery) {
+        if (isPipelineStatisticsQuery) {
+            // Nothing to start: the counter is uninstrumented and GL_QUERY_COUNTER_BITS says so.
+            // The object still becomes a real, target-latched query so every other rule about it
+            // (re-use with another target, double-begin, EndQuery pairing) keeps holding.
+        } else if (isTransformFeedbackQuery) {
             // Prefer real GPU transform-feedback queries (exact with geometry shaders);
             // the CPU accounting delta stays as the fallback when the backend lacks them.
             const auto beginXfbPrimitivesQuery = MG_Backend::gBackendFunctionsTable.GL.BeginXfbPrimitivesQuery;
@@ -476,15 +518,19 @@ namespace MobileGL::MG_Impl::GLImpl {
             (target == GL_SAMPLES_PASSED || target == GL_ANY_SAMPLES_PASSED ||
              target == GL_ANY_SAMPLES_PASSED_CONSERVATIVE) &&
             MG_Backend::gBackendFunctionsTable.GL.BeginOcclusionQuery != nullptr;
-        if (target != GL_TIME_ELAPSED && !isTransformFeedbackQuery && !isOcclusionQuery) {
+        const Bool isPipelineStatisticsQuery = IsPipelineStatisticsQueryTarget(target);
+        if (target != GL_TIME_ELAPSED && !isTransformFeedbackQuery && !isOcclusionQuery &&
+            !isPipelineStatisticsQuery) {
             RecordQueryError(ErrorCode::InvalidEnum, __FUNCTION__, "Query target is not supported.");
             return;
         }
         const std::lock_guard<std::mutex> lock(g_queryObjectsMutex);
-        GLuint& activeQueryId = isTransformFeedbackQuery
-            ? (target == GL_TRANSFORM_FEEDBACK_PRIMITIVES_WRITTEN ? g_activePrimitivesWrittenQueryId
-                                                                  : g_activePrimitivesGeneratedQueryId)
-            : (isOcclusionQuery ? g_activeSamplesPassedQueryId : g_activeTimeElapsedQueryId);
+        GLuint& activeQueryId = isPipelineStatisticsQuery
+            ? g_activePipelineStatisticsQueryIds[target]
+            : (isTransformFeedbackQuery
+                   ? (target == GL_TRANSFORM_FEEDBACK_PRIMITIVES_WRITTEN ? g_activePrimitivesWrittenQueryId
+                                                                         : g_activePrimitivesGeneratedQueryId)
+                   : (isOcclusionQuery ? g_activeSamplesPassedQueryId : g_activeTimeElapsedQueryId));
         if (activeQueryId == 0) {
             RecordQueryError(ErrorCode::InvalidOperation, __FUNCTION__, "No query is active on this target.");
             return;
@@ -494,6 +540,17 @@ namespace MobileGL::MG_Impl::GLImpl {
             activeQueryId = 0; // should not happen; keep state consistent
             return;
         }
+        if (isPipelineStatisticsQuery) {
+            // The result is a definite zero rather than an unread backend handle, so a later
+            // GetQueryObject* answers immediately and never waits on something that was never
+            // started. GL_QUERY_COUNTER_BITS = 0 is what marks that zero indeterminate.
+            queryObject->cachedResult = 0;
+            queryObject->resultCached = true;
+            queryObject->active = false;
+            queryObject->ended = true;
+            activeQueryId = 0;
+            return;
+        }
         if (isTransformFeedbackQuery) {
             if (queryObject->backendHandle) {
                 if (const auto endXfbPrimitivesQuery = MG_Backend::gBackendFunctionsTable.GL.EndXfbPrimitivesQuery) {
@@ -657,7 +714,12 @@ namespace MobileGL::MG_Impl::GLImpl {
                 *params = static_cast<GLint>(g_activePrimitivesGeneratedQueryId);
                 break;
             default:
-                *params = 0;
+                if (IsPipelineStatisticsQueryTarget(target)) {
+                    const auto it = g_activePipelineStatisticsQueryIds.find(target);
+                    *params = it != g_activePipelineStatisticsQueryIds.end() ? static_cast<GLint>(it->second) : 0;
+                } else {
+                    *params = 0;
+                }
                 break;
             }
             return;
@@ -668,6 +730,14 @@ namespace MobileGL::MG_Impl::GLImpl {
             // entry points / timestamp valid bits at call time, not at table
             // init), and the MOBILEGL_DISABLE_TIMERQUERY kill switch always
             // wins.
+            if (IsPipelineStatisticsQueryTarget(target)) {
+                // Zero: GL 4.6 core 4.2.1's way of saying the counter is not implemented and its
+                // results are indeterminate. The conformance suite reads exactly this and skips
+                // the functional half of each such target, which is the outcome an uninstrumented
+                // counter should produce.
+                *params = 0;
+                return;
+            }
             if (target == GL_SAMPLES_PASSED || target == GL_ANY_SAMPLES_PASSED ||
                 target == GL_ANY_SAMPLES_PASSED_CONSERVATIVE) {
                 const Bool occlusionSupported = MG_Backend::gBackendFunctionsTable.GL.BeginOcclusionQuery != nullptr;
@@ -741,14 +811,24 @@ namespace MobileGL::MG_Impl::GLImpl {
     }
 
     namespace {
+        Bool IsPerVertexStreamQueryTarget(GLenum target) {
+            return target == GL_PRIMITIVES_GENERATED || target == GL_TRANSFORM_FEEDBACK_PRIMITIVES_WRITTEN;
+        }
+
         // The indexed query entry points differ from the plain ones only in the vertex
         // stream they address (GL 4.6 core 4.2.1): index must be below GL_MAX_VERTEX_STREAMS
-        // for the two transform feedback targets and zero for every other target. With a
-        // single vertex stream both bounds are 1, so a valid call is always index 0 and
-        // forwards to the unindexed implementation.
+        // for the two transform feedback targets and zero for every other target. MobileGL
+        // implements ONE vertex stream, so both bounds are 1 and a valid call is always index 0 -
+        // which is what makes the three forwards below equivalent to the unindexed entry points.
+        //
+        // THAT EQUIVALENCE IS THE WHOLE JUSTIFICATION, and it is read out of the getter rather
+        // than assumed: the moment GL_MAX_VERTEX_STREAMS answers more than one, index 1..3 starts
+        // reaching EndQueryIndexed and GetQueryIndexediv, which resolve the active query from
+        // per-TARGET globals and would end - or report - a query begun on a different stream.
+        // Raising that limit therefore means giving each active query a stream index and
+        // comparing it here, not just changing the number.
         Bool ValidateQueryStreamIndex(const char* function, GLenum target, GLuint index) {
-            const Bool perStreamTarget =
-                target == GL_PRIMITIVES_GENERATED || target == GL_TRANSFORM_FEEDBACK_PRIMITIVES_WRITTEN;
+            const Bool perStreamTarget = IsPerVertexStreamQueryTarget(target);
             GLint maxVertexStreams = 1;
             if (perStreamTarget) {
                 GetIntegerv(GL_MAX_VERTEX_STREAMS, &maxVertexStreams);
@@ -761,6 +841,7 @@ namespace MobileGL::MG_Impl::GLImpl {
                                              : "index must be zero for this query target.");
             return false;
         }
+
     } // namespace
 
     void BeginQueryIndexed(GLenum target, GLuint index, GLuint id) {

MobileGL/MG_Impl/GLImpl/RenderState/GL_RenderState.cpp

@@ -328,10 +328,50 @@ namespace MobileGL::MG_Impl::GLImpl {
         MG_State::pGLContext->SetSampleCoverage(std::clamp(static_cast<Float>(value), 0.0f, 1.0f), invert == GL_TRUE);
     }
 
+    // ARB_sample_shading / GL 4.6 core 14.3.1: "value is clamped to [0, 1] when specified", so
+    // there is no error to raise - a caller that asks for 2.0 gets 1.0 and GL_MIN_SAMPLE_SHADING_-
+    // VALUE reads back 1.0. Was a logging no-op while ARB_sample_shading was advertised, which
+    // let an application enable GL_SAMPLE_SHADING and then quietly get the driver's default rate.
+    void MinSampleShading_State(GLfloat value) {
+        MG_State::pGLContext->SetMinSampleShadingValue(std::clamp(static_cast<Float>(value), 0.0f, 1.0f));
+    }
+
     void PolygonOffset_State(GLfloat factor, GLfloat units) {
         MG_State::pGLContext->SetPolygonOffset(static_cast<Float>(factor), static_cast<Float>(units));
     }
 
+    void PolygonOffsetClamp_State(GLfloat factor, GLfloat units, GLfloat clamp) {
+        // GL 4.6 core 14.6.5 / GL_EXT_polygon_offset_clamp. No error cases: any three floats are
+        // legal, and clamp = 0 is exactly glPolygonOffset. Whether the backend can APPLY the clamp
+        // is a separate question (see the DirectGLES/DirectVulkan forwarding); the state is
+        // recorded either way, because GL_POLYGON_OFFSET_CLAMP has to read back what was written.
+        MG_State::pGLContext->SetPolygonOffsetClamped(static_cast<Float>(factor), static_cast<Float>(units),
+                                                      static_cast<Float>(clamp));
+    }
+
+    void ClipControl_State(GLenum origin, GLenum depth) {
+        // GL 4.5 core 13.5: both arguments are strict enums, and either being wrong is
+        // GL_INVALID_ENUM with the state left untouched.
+        if (origin != GL_LOWER_LEFT && origin != GL_UPPER_LEFT) {
+            MG_State::pGLContext->RecordError(
+                ErrorCode::InvalidEnum,
+                MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", __func__,
+                                             "glClipControl origin must be GL_LOWER_LEFT or GL_UPPER_LEFT; got " +
+                                                 MG_Util::ConvertGLEnumToString(origin) + "."));
+            return;
+        }
+        if (depth != GL_NEGATIVE_ONE_TO_ONE && depth != GL_ZERO_TO_ONE) {
+            MG_State::pGLContext->RecordError(
+                ErrorCode::InvalidEnum,
+                MakeUnique<GenericErrorInfo>(
+                    "MG_Impl/GLImpl", __func__,
+                    "glClipControl depth must be GL_NEGATIVE_ONE_TO_ONE or GL_ZERO_TO_ONE; got " +
+                        MG_Util::ConvertGLEnumToString(depth) + "."));
+            return;
+        }
+        MG_State::pGLContext->SetClipControl(origin, depth);
+    }
+
     void PolygonMode_State(GLenum face, GLenum mode) {
         // GL 3.3 core: separate front/back polygon modes were removed in 3.1, so the only legal
         // face is GL_FRONT_AND_BACK. GL_FRONT / GL_BACK must be rejected (some desktop drivers
@@ -1013,10 +1053,22 @@ namespace MobileGL::MG_Impl::GLImpl {
         SampleCoverage_State(value, invert);
     }
 
+    void MinSampleShading(GLfloat value) {
+        MinSampleShading_State(value);
+    }
+
     void PolygonOffset(GLfloat factor, GLfloat units) {
         PolygonOffset_State(factor, units);
     }
 
+    void PolygonOffsetClamp(GLfloat factor, GLfloat units, GLfloat clamp) {
+        PolygonOffsetClamp_State(factor, units, clamp);
+    }
+
+    void ClipControl(GLenum origin, GLenum depth) {
+        ClipControl_State(origin, depth);
+    }
+
     void PolygonMode(GLenum face, GLenum mode) {
         PolygonMode_State(face, mode);
     }

MobileGL/MG_Impl/GLImpl/RenderState/GL_RenderState.h

@@ -38,7 +38,10 @@ namespace MobileGL::MG_Impl::GLImpl {
     void StencilFunc(GLenum func, GLint ref, GLuint mask);
     void Scissor(GLint x, GLint y, GLsizei width, GLsizei height);
     void SampleCoverage(GLfloat value, GLboolean invert);
+    void MinSampleShading(GLfloat value);
     void PolygonOffset(GLfloat factor, GLfloat units);
+    void PolygonOffsetClamp(GLfloat factor, GLfloat units, GLfloat clamp);
+    void ClipControl(GLenum origin, GLenum depth);
     void PolygonMode(GLenum face, GLenum mode);
     void PointSize(GLfloat size);
     void PointParameterf(GLenum pname, GLfloat param);

MobileGL/MG_Impl/GLImpl/Sampler/GL_Sampler.cpp

@@ -13,6 +13,7 @@
 #include <MG_State/GLState/Core.h>
 #include <MG_Util/Converters/GLToMG/TextureEnumConverter.h>
 #include <MG_Util/Converters/MGToGL/TextureEnumConverter.h>
+#include <MG_Util/Math/FixedPointConversion.h>
 
 namespace MobileGL::MG_Impl::GLImpl {
     namespace {
@@ -22,6 +23,50 @@ namespace MobileGL::MG_Impl::GLImpl {
             return static_cast<Float>(*(const GLint*)param);
         }
 
+        // GL_TEXTURE_BORDER_COLOR is the only sampler parameter with more than one component, and it
+        // is also the only one whose meaning depends on WHICH entry point wrote it. Everything else
+        // reads exactly one component and does not care.
+        Bool IsVectorOnlySamplerPname(GLenum pname) {
+            return pname == GL_TEXTURE_BORDER_COLOR;
+        }
+
+        // A state query returns the value CONVERTED to the type the caller asked for (GL 4.6 core
+        // 2.2.2 / 6.1), never the other type's bits. These two are the sampler side of the numeric
+        // casts GetTexParameterfv_State/GetTexParameteriv_State already do on the texture side; the
+        // sampler path funnels all three spellings through one void* function, which is precisely how
+        // it came to write a fixed type regardless of the caller.
+        //
+        // Truncation rather than rounding for the float -> integer direction, matching the texture
+        // twin (GetTexParameteriv_State's static_cast<GLint> on MIN_LOD/MAX_LOD/LOD_BIAS): the two
+        // spellings of the same state disagreeing is the bug being fixed here, and a texture and a
+        // sampler queried the same way must answer the same number.
+        void StoreSamplerScalar(void* params, Bool isFloat, Bool isUnsignedInteger, Float value) {
+            if (isFloat) {
+                *(GLfloat*)params = value;
+                return;
+            }
+            // Via GLint in both integer spellings: a direct float -> GLuint cast of a negative value
+            // (GL_TEXTURE_MIN_LOD defaults to -1000) is undefined behaviour, while the two-step
+            // conversion is the well-defined modular one, and it is what the texture-side
+            // GetTexParameterIuiv fallback does.
+            const GLint asInt = static_cast<GLint>(value);
+            if (isUnsignedInteger) {
+                *(GLuint*)params = static_cast<GLuint>(asInt);
+            } else {
+                *(GLint*)params = asInt;
+            }
+        }
+
+        void StoreSamplerEnum(void* params, Bool isFloat, Bool isUnsignedInteger, GLenum value) {
+            if (isFloat) {
+                *(GLfloat*)params = static_cast<GLfloat>(value);
+            } else if (isUnsignedInteger) {
+                *(GLuint*)params = value;
+            } else {
+                *(GLint*)params = static_cast<GLint>(value);
+            }
+        }
+
         Bool ValidateSamplerParameterValue(GLenum pname, const void* param, Bool isFloat, Bool isUnsignedInteger) {
             if (param == nullptr) return false;
 
@@ -56,8 +101,15 @@ namespace MobileGL::MG_Impl::GLImpl {
         }
     } // namespace
 
+    // `isIntegerCommand` distinguishes the "I" spellings (glSamplerParameterIiv / Iuiv) from the
+    // plain ones. It only matters for GL_TEXTURE_BORDER_COLOR, and there it decides everything:
+    // GL 4.6 core 8.10 says the I forms store the components unmodified with an integer internal
+    // type, while glSamplerParameteriv converts them to floating point with equation 2.2. Routing
+    // both to the same setter - which is what this file used to do - meant glSamplerParameteriv
+    // stored raw integers (so a border of 255 became float 255.0 instead of the spec's ~1.19e-7)
+    // and glSamplerParameterIiv lost the fact that it was ever an integer at all.
     void SetSamplerParam_State(GLuint sampler, GLenum pname, const void* param, bool isFloat,
-                               bool isUnsignedInteger) {
+                               bool isUnsignedInteger, bool isIntegerCommand) {
         if (param == nullptr) return;
         if (!SamplerImpl::ValidateSamplerName(sampler)) return;
 
@@ -112,6 +164,13 @@ namespace MobileGL::MG_Impl::GLImpl {
             if (isFloat) {
                 const auto* values = (const GLfloat*)param;
                 samplerObj->SetBorderColor(FloatVec4(values[0], values[1], values[2], values[3]));
+            } else if (!isIntegerCommand) {
+                // glSamplerParameteriv: GL 4.6 core equation 2.2 into the FLOAT border colour.
+                const auto* values = (const GLint*)param;
+                samplerObj->SetBorderColor(FloatVec4(MG_Util::SignedNormalizedInt32ToFloat(values[0]),
+                                                     MG_Util::SignedNormalizedInt32ToFloat(values[1]),
+                                                     MG_Util::SignedNormalizedInt32ToFloat(values[2]),
+                                                     MG_Util::SignedNormalizedInt32ToFloat(values[3])));
             } else if (isUnsignedInteger) {
                 const auto* values = (const GLuint*)param;
                 samplerObj->SetBorderColorUI(UintVec4(values[0], values[1], values[2], values[3]));
@@ -128,7 +187,7 @@ namespace MobileGL::MG_Impl::GLImpl {
     }
 
     void GetSamplerParam_State(GLuint sampler, GLenum pname, void* params, bool isFloat,
-                               bool isUnsignedInteger) {
+                               bool isUnsignedInteger, bool isIntegerCommand) {
         if (params == nullptr) return;
         if (!SamplerImpl::ValidateSamplerName(sampler)) return;
 
@@ -141,47 +200,56 @@ namespace MobileGL::MG_Impl::GLImpl {
         if (!SamplerImpl::ValidateSamplerObject(sampler)) return;
 
         using namespace MG_Util;
+        // Every scalar pname goes through StoreSamplerScalar/StoreSamplerEnum so the CALLER'S form
+        // decides the destination type. Writing a fixed type regardless - which is what these case
+        // labels used to do - hands back the other type's bit pattern rather than a converted value:
+        // glGetSamplerParameterfv(GL_TEXTURE_WRAP_S) deposited the integer 10497 into a GLfloat and
+        // the caller read 1.47e-41, and glGetSamplerParameteriv(GL_TEXTURE_MIN_LOD) deposited the
+        // IEEE bits of -1000.0f and the caller read -998637568. Sixteen (pname, entry-point) pairs
+        // were broken this way; only MAX_ANISOTROPY_EXT and BORDER_COLOR branched correctly, which is
+        // how the same bug class was already found and fixed once for a single pname.
         switch (pname) {
         case GL_TEXTURE_WRAP_S:
-            *(GLuint*)params = MG_Util::ConvertSamplerWrapModeToGLEnum(samplerObj->GetWrapS());
+            StoreSamplerEnum(params, isFloat, isUnsignedInteger,
+                             MG_Util::ConvertSamplerWrapModeToGLEnum(samplerObj->GetWrapS()));
             break;
         case GL_TEXTURE_WRAP_T:
-            *(GLuint*)params = MG_Util::ConvertSamplerWrapModeToGLEnum(samplerObj->GetWrapT());
+            StoreSamplerEnum(params, isFloat, isUnsignedInteger,
+                             MG_Util::ConvertSamplerWrapModeToGLEnum(samplerObj->GetWrapT()));
             break;
         case GL_TEXTURE_WRAP_R:
-            *(GLuint*)params = MG_Util::ConvertSamplerWrapModeToGLEnum(samplerObj->GetWrapR());
+            StoreSamplerEnum(params, isFloat, isUnsignedInteger,
+                             MG_Util::ConvertSamplerWrapModeToGLEnum(samplerObj->GetWrapR()));
             break;
         case GL_TEXTURE_MIN_FILTER:
-            *(GLuint*)params =
-                MG_Util::ConvertSamplerFilterModeToGLEnum(samplerObj->GetMinFilter(), samplerObj->GetMipmapMode());
+            StoreSamplerEnum(params, isFloat, isUnsignedInteger,
+                             MG_Util::ConvertSamplerFilterModeToGLEnum(samplerObj->GetMinFilter(),
+                                                                       samplerObj->GetMipmapMode()));
             break;
         case GL_TEXTURE_MAG_FILTER:
-            *(GLuint*)params =
-                MG_Util::ConvertSamplerFilterModeToGLEnum(samplerObj->GetMagFilter(), SamplerMipmapMode::None);
+            StoreSamplerEnum(params, isFloat, isUnsignedInteger,
+                             MG_Util::ConvertSamplerFilterModeToGLEnum(samplerObj->GetMagFilter(),
+                                                                       SamplerMipmapMode::None));
             break;
         case GL_TEXTURE_MIN_LOD:
-            *(GLfloat*)params = samplerObj->GetMinLod();
+            StoreSamplerScalar(params, isFloat, isUnsignedInteger, samplerObj->GetMinLod());
             break;
         case GL_TEXTURE_MAX_LOD:
-            *(GLfloat*)params = samplerObj->GetMaxLod();
+            StoreSamplerScalar(params, isFloat, isUnsignedInteger, samplerObj->GetMaxLod());
             break;
         case GL_TEXTURE_LOD_BIAS:
-            *(GLfloat*)params = samplerObj->GetLodBias();
+            StoreSamplerScalar(params, isFloat, isUnsignedInteger, samplerObj->GetLodBias());
             break;
         case GL_TEXTURE_MAX_ANISOTROPY_EXT:
-            if (isFloat) {
-                *(GLfloat*)params = samplerObj->GetMaxAnisotropy();
-            } else if (isUnsignedInteger) {
-                *(GLuint*)params = static_cast<GLuint>(samplerObj->GetMaxAnisotropy());
-            } else {
-                *(GLint*)params = static_cast<GLint>(samplerObj->GetMaxAnisotropy());
-            }
+            StoreSamplerScalar(params, isFloat, isUnsignedInteger, samplerObj->GetMaxAnisotropy());
             break;
         case GL_TEXTURE_COMPARE_MODE:
-            *(GLuint*)params = MG_Util::ConvertSamplerCompareModeToGLEnum(samplerObj->GetCompareMode());
+            StoreSamplerEnum(params, isFloat, isUnsignedInteger,
+                             MG_Util::ConvertSamplerCompareModeToGLEnum(samplerObj->GetCompareMode()));
             break;
         case GL_TEXTURE_COMPARE_FUNC:
-            *(GLuint*)params = MG_Util::ConvertSamplerCompareFuncToGLEnum(samplerObj->GetSamplerCompareFunc());
+            StoreSamplerEnum(params, isFloat, isUnsignedInteger,
+                             MG_Util::ConvertSamplerCompareFuncToGLEnum(samplerObj->GetSamplerCompareFunc()));
             break;
         case GL_TEXTURE_BORDER_COLOR: {
             if (isFloat) {
@@ -191,6 +259,16 @@ namespace MobileGL::MG_Impl::GLImpl {
                 out[1] = color.y();
                 out[2] = color.z();
                 out[3] = color.w();
+            } else if (!isIntegerCommand) {
+                // glGetSamplerParameteriv: the inverse of the write side, GL 4.6 core equation 2.3.
+                // Exactly inverse, so a {0,1,2,4} written with glSamplerParameteriv reads back as
+                // {0,1,2,4}; a bare truncating cast answered {0,0,0,0}.
+                const auto& color = samplerObj->GetBorderColor();
+                auto* out = (GLint*)params;
+                out[0] = MG_Util::FloatToSignedNormalizedInt32(color.x());
+                out[1] = MG_Util::FloatToSignedNormalizedInt32(color.y());
+                out[2] = MG_Util::FloatToSignedNormalizedInt32(color.z());
+                out[3] = MG_Util::FloatToSignedNormalizedInt32(color.w());
             } else if (isUnsignedInteger) {
                 const auto& color = samplerObj->GetBorderColorUI();
                 auto* out = (GLuint*)params;
@@ -293,16 +371,10 @@ namespace MobileGL::MG_Impl::GLImpl {
         if (sampler == 0) {
             textureUnit.SetSamplerObject(nullptr);
         } else {
-            // GL 3.3 core 3.8.2: BindSampler on a name GenSamplers never returned - or one already
-            // deleted - is INVALID_OPERATION. SamplerParameter* raises INVALID_VALUE for the same
-            // name, which is why this cannot go through the shared SamplerImpl validator.
-            if (!MG_State::pGLContext->ValidateSamplerName(sampler)) {
-                MG_State::pGLContext->RecordError(
-                    ErrorCode::InvalidOperation,
-                    MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", "BindSampler_State",
-                                                 std::format("Invalid sampler name {}", sampler)));
-                return;
-            }
+            // GL 4.6 core 8.2: BindSampler on a name GenSamplers never returned - or one already
+            // deleted - is INVALID_OPERATION, and so is every other sampler entry point on such a
+            // name, so the shared validator answers for all of them.
+            if (!SamplerImpl::ValidateSamplerName(sampler)) return;
             Bool doesSamplerObjectCreated = MG_State::pGLContext->ValidateSamplerObject(sampler);
             if (!doesSamplerObjectCreated) {
                 MG_State::pGLContext->CreateSamplerObject(sampler);
@@ -356,30 +428,50 @@ namespace MobileGL::MG_Impl::GLImpl {
 
     /* @INSERTION_POINT:FUNCTION_IMPLEMENTATION@ */
     void GetSamplerParameteriv(GLuint sampler, GLenum pname, GLint* params) {
-        GetSamplerParam_State(sampler, pname, params, false, false);
+        GetSamplerParam_State(sampler, pname, params, false, false, false);
     }
 
     void SamplerParameterIuiv(GLuint sampler, GLenum pname, const GLuint* param) {
-        SetSamplerParam_State(sampler, pname, param, false, true);
+        SetSamplerParam_State(sampler, pname, param, false, true, true);
     }
 
     void SamplerParameterIiv(GLuint sampler, GLenum pname, const GLint* param) {
-        SetSamplerParam_State(sampler, pname, param, false, false);
+        SetSamplerParam_State(sampler, pname, param, false, false, true);
     }
 
     void SamplerParameteriv(GLuint sampler, GLenum pname, const GLint* param) {
-        SetSamplerParam_State(sampler, pname, param, false, false);
+        SetSamplerParam_State(sampler, pname, param, false, false, false);
     }
 
     void SamplerParameterfv(GLuint sampler, GLenum pname, const GLfloat* param) {
-        SetSamplerParam_State(sampler, pname, param, true, false);
+        SetSamplerParam_State(sampler, pname, param, true, false, false);
     }
 
+    // GL 4.6 core 8.10: the scalar spellings take "the value of pname", so a pname with more than one
+    // component is INVALID_ENUM here rather than something to read four components of. Guarding at
+    // the entry point rather than downstream is also what stops the vector path reading twelve bytes
+    // past the caller's single stack scalar - taking the address of a by-value argument and handing
+    // it to a four-component reader is what these used to do. The texture-side twins already answer
+    // INVALID_ENUM for GL_TEXTURE_BORDER_COLOR (TexParameteri/f name it as unsupported outright).
     void SamplerParameteri(GLuint sampler, GLenum pname, GLint param) {
+        if (IsVectorOnlySamplerPname(pname)) {
+            MG_State::pGLContext->RecordError(
+                ErrorCode::InvalidEnum,
+                MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", "SamplerParameteri",
+                                             "pname has more than one component and needs a vector form."));
+            return;
+        }
         SamplerParameteriv(sampler, pname, &param);
     }
 
     void SamplerParameterf(GLuint sampler, GLenum pname, GLfloat param) {
+        if (IsVectorOnlySamplerPname(pname)) {
+            MG_State::pGLContext->RecordError(
+                ErrorCode::InvalidEnum,
+                MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", "SamplerParameterf",
+                                             "pname has more than one component and needs a vector form."));
+            return;
+        }
         SamplerParameterfv(sampler, pname, &param);
     }
 
@@ -388,15 +480,15 @@ namespace MobileGL::MG_Impl::GLImpl {
     }
 
     void GetSamplerParameterIuiv(GLuint sampler, GLenum pname, GLuint* params) {
-        GetSamplerParam_State(sampler, pname, params, false, true);
+        GetSamplerParam_State(sampler, pname, params, false, true, true);
     }
 
     void GetSamplerParameterIiv(GLuint sampler, GLenum pname, GLint* params) {
-        GetSamplerParam_State(sampler, pname, params, false, false);
+        GetSamplerParam_State(sampler, pname, params, false, false, true);
     }
 
     void GetSamplerParameterfv(GLuint sampler, GLenum pname, GLfloat* params) {
-        GetSamplerParam_State(sampler, pname, params, true, false);
+        GetSamplerParam_State(sampler, pname, params, true, false, false);
     }
 
     void GenSamplers(GLsizei count, GLuint* samplers) {

MobileGL/MG_Impl/GLImpl/Sampler/Validators.cpp

@@ -12,11 +12,17 @@
 #include <MG_Util/Converters/GLToMG/TextureEnumConverter.h>
 
 namespace MobileGL::MG_Impl::GLImpl::SamplerImpl {
+    // GL 4.6 core 8.2: "An INVALID_OPERATION error is generated if sampler is not the name of a
+    // sampler object previously returned from a call to GenSamplers." That class is shared by every
+    // sampler entry point - BindSampler, SamplerParameter*, GetSamplerParameter* - so this one gate
+    // answers for all of them. It used to report INVALID_VALUE (the GL 3.3 wording), which forced
+    // BindSampler to carry a bespoke duplicate of the same check just to get the class right.
     Bool ValidateSamplerName(GLuint sampler) {
         if (!MG_State::pGLContext->ValidateSamplerName(sampler)) {
             MG_State::pGLContext->RecordError(
-                ErrorCode::InvalidValue, MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", "ValidateSamplerName",
-                                                                      std::format("Invalid sampler name {}", sampler)));
+                ErrorCode::InvalidOperation,
+                MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", "ValidateSamplerName",
+                                             std::format("Invalid sampler name {}", sampler)));
             return false;
         }
         return true;

MobileGL/MG_Impl/GLImpl/Sync/GL_Sync.cpp

@@ -69,8 +69,25 @@ namespace MobileGL::MG_Impl::GLImpl {
     }
 
     GLenum ClientWaitSync(GLsync sync, GLbitfield flags, GLuint64 timeout) {
+        // GL 4.6 core 4.1.1: GL_SYNC_FLUSH_COMMANDS_BIT is the only bit this call accepts, and
+        // any other bit is INVALID_VALUE. Silently ignoring the stray bits used to make a caller
+        // that passed, say, GL_SYNC_GPU_COMMANDS_COMPLETE by mistake think it had asked for a
+        // flush it never got.
+        if ((flags & ~static_cast<GLbitfield>(GL_SYNC_FLUSH_COMMANDS_BIT)) != 0) {
+            MG_State::pGLContext->RecordError(
+                ErrorCode::InvalidValue,
+                MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", __func__,
+                                             "flags must be zero or GL_SYNC_FLUSH_COMMANDS_BIT."));
+            return GL_WAIT_FAILED;
+        }
         const auto* syncObject = FindSyncObject(sync);
         if (!syncObject) {
+            // The spec pairs the GL_WAIT_FAILED return with a recorded INVALID_VALUE; returning
+            // the enum alone left glGetError() clean and the failure indistinguishable from a
+            // genuine wait failure on a live sync.
+            MG_State::pGLContext->RecordError(
+                ErrorCode::InvalidValue,
+                MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", __func__, "sync is not the name of a sync object."));
             return GL_WAIT_FAILED;
         }
         const auto backendClientWaitSync = MG_Backend::gBackendFunctionsTable.GL.ClientWaitSync;
@@ -95,6 +112,9 @@ namespace MobileGL::MG_Impl::GLImpl {
         }
         const auto* syncObject = FindSyncObject(sync);
         if (!syncObject) {
+            MG_State::pGLContext->RecordError(
+                ErrorCode::InvalidValue,
+                MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", __func__, "sync is not the name of a sync object."));
             return;
         }
         const auto backendWaitSync = MG_Backend::gBackendFunctionsTable.GL.WaitSync;
@@ -125,8 +145,22 @@ namespace MobileGL::MG_Impl::GLImpl {
     }
 
     void GetSynciv(GLsync sync, GLenum pname, GLsizei bufSize, GLsizei* length, GLint* values) {
+        // GL 4.6 core 4.1: a negative bufSize is INVALID_VALUE, an unnamed sync is INVALID_VALUE
+        // and an unrecognised pname is INVALID_ENUM. All three used to leave glGetError() clean
+        // and write a plausible-looking zero, which is the one failure mode a caller cannot tell
+        // apart from a real answer - GL_SYNC_STATUS legitimately answers GL_UNSIGNALED (0x9118),
+        // but a mistyped pname answered a bare 0 that no query ever returns.
+        if (bufSize < 0) {
+            MG_State::pGLContext->RecordError(
+                ErrorCode::InvalidValue,
+                MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", __func__, "bufSize must not be negative."));
+            return;
+        }
         const auto* syncObject = FindSyncObject(sync);
         if (!syncObject) {
+            MG_State::pGLContext->RecordError(
+                ErrorCode::InvalidValue,
+                MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", __func__, "sync is not the name of a sync object."));
             if (length) {
                 *length = 0;
             }
@@ -152,7 +186,15 @@ namespace MobileGL::MG_Impl::GLImpl {
             value = static_cast<GLint>(syncObject->flags);
             break;
         default:
-            break;
+            MG_State::pGLContext->RecordError(
+                ErrorCode::InvalidEnum,
+                MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", __func__,
+                                             "pname must be GL_OBJECT_TYPE, GL_SYNC_STATUS, GL_SYNC_CONDITION or "
+                                             "GL_SYNC_FLAGS."));
+            if (length) {
+                *length = 0;
+            }
+            return;
         }
 
         if (length) {

MobileGL/MG_Impl/GLImpl/Texture/GL_Texture.h

@@ -8,9 +8,24 @@
 
 #pragma once
 #include <Includes.h>
+#include <MG_State/GLState/TextureState/TextureObject.h>
 
 namespace MobileGL::MG_Impl::GLImpl {
     /* @INSERTION_POINT:FUNCTION_DECLARATION@ */
+    // Answers a texture-image query straight out of the CPU shadow, into client memory or a bound
+    // PIXEL_PACK_BUFFER. This is the whole of glGetTexImage on a build with no backend readback, and
+    // it is also the sound fallback for a backend that has no GPU image to read: with no image,
+    // nothing GPU-side can ever have written the texture, so the shadow IS its content.
+    //
+    // It answers a NARROWER contract than glGetTexImage's, and refuses what it cannot do rather than
+    // answering wrongly. The copy is verbatim: it performs no format or type conversion, and it packs
+    // rows tightly, honouring only GL_PACK_SWAP_BYTES and the bitmap GL_PACK_LSB_FIRST path. A
+    // request whose (format, type) texel size differs from the texture's own, or a pixel-store state
+    // that adds row padding / a row-length override / a skip offset, is rejected with
+    // GL_INVALID_OPERATION (see ValidateShadowReadbackLayout, which spells out why each is unsafe).
+    void CopyTextureImageToClientOrPBO_State(const SharedPtr<MG_State::GLState::ITextureObject>& textureObject,
+                                             TextureUploadTarget textureUploadTarget, GLint level, GLenum format,
+                                             GLenum type, GLsizei bufSize, void* pixels, const char* caller);
     // The sized internal formats a buffer texture accepts (GL 4.6 core table 8.16). The buffer
     // clears take the same list, so it is shared rather than written out twice.
     Bool IsBufferTextureInternalFormat(GLenum internalformat);
@@ -37,8 +52,13 @@ namespace MobileGL::MG_Impl::GLImpl {
                            GLenum format, GLenum type, const void* pixels);
     void TextureSubImage3D(GLuint texture, GLint level, GLint xoffset, GLint yoffset, GLint zoffset, GLsizei width,
                            GLsizei height, GLsizei depth, GLenum format, GLenum type, const void* pixels);
+    void CompressedTextureSubImage1D(GLuint texture, GLint level, GLint xoffset, GLsizei width, GLenum format,
+                                     GLsizei imageSize, const void* data);
     void CompressedTextureSubImage2D(GLuint texture, GLint level, GLint xoffset, GLint yoffset, GLsizei width,
                                      GLsizei height, GLenum format, GLsizei imageSize, const void* data);
+    void CompressedTextureSubImage3D(GLuint texture, GLint level, GLint xoffset, GLint yoffset, GLint zoffset,
+                                     GLsizei width, GLsizei height, GLsizei depth, GLenum format, GLsizei imageSize,
+                                     const void* data);
     void TextureParameterf(GLuint texture, GLenum pname, GLfloat param);
     void TextureParameterfv(GLuint texture, GLenum pname, const GLfloat* params);
     void TextureParameteri(GLuint texture, GLenum pname, GLint param);
@@ -60,6 +80,8 @@ namespace MobileGL::MG_Impl::GLImpl {
     void GetTextureParameteriv(GLuint texture, GLenum pname, GLint* params);
     void GetTextureLevelParameterfv(GLuint texture, GLint level, GLenum pname, GLfloat* params);
     void GetTextureLevelParameteriv(GLuint texture, GLint level, GLenum pname, GLint* params);
+    void TextureView(GLuint texture, GLenum target, GLuint origtexture, GLenum internalformat, GLuint minlevel,
+                     GLuint numlevels, GLuint minlayer, GLuint numlayers);
     void TexStorage1D(GLenum target, GLsizei levels, GLenum internalformat, GLsizei width);
     void TexStorage2D(GLenum target, GLsizei levels, GLenum internalformat, GLsizei width, GLsizei height);
     void TexStorage3D(GLenum target, GLsizei levels, GLenum internalformat, GLsizei width, GLsizei height,

MobileGL/MG_Impl/GLImpl/Texture/Validators.cpp

@@ -103,6 +103,28 @@ namespace MobileGL::MG_Impl::GLImpl::TextureImpl {
         return true;
     }
 
+    Bool ValidateCubeMapArrayShape(TextureUploadTarget target, GLsizei width, GLsizei height, GLsizei depth,
+                                   const char* caller) {
+        if (target != TextureUploadTarget::CubeMapArray && target != TextureUploadTarget::ProxyCubeMapArray) {
+            return true;
+        }
+        if (width != height) {
+            MG_State::pGLContext->RecordError(
+                ErrorCode::InvalidValue,
+                MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", caller,
+                                             "Cube map array levels must be square (width == height)"));
+            return false;
+        }
+        if (depth % 6 != 0) {
+            MG_State::pGLContext->RecordError(
+                ErrorCode::InvalidValue,
+                MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", caller,
+                                             "Cube map array depth must be a multiple of six"));
+            return false;
+        }
+        return true;
+    }
+
     Bool ValidateTextureSizeWithTextureUploadTarget(TextureUploadTarget target, GLsizei width, GLsizei height) {
         if (target == TextureUploadTarget::CubeMapPositiveX || target == TextureUploadTarget::CubeMapNegativeX ||
             target == TextureUploadTarget::CubeMapPositiveY || target == TextureUploadTarget::CubeMapNegativeY ||
@@ -623,4 +645,144 @@ namespace MobileGL::MG_Impl::GLImpl::TextureImpl {
         }
         return true;
     }
+
+    // GL 4.6 core table 8.21 ("Compatible internal formats for TextureView"), transcribed whole.
+    // Written against the raw GLenum rather than TextureInternalFormat on purpose: MobileGL's own
+    // enum collapses every compressed format onto uncompressed storage and drops formats it
+    // cannot carry, so classifying the converted value would silently widen the compatibility
+    // rule - GL_COMPRESSED_RG_RGTC2 and GL_RGBA8 would end up in the same class.
+    TextureViewClass GetTextureViewClass(GLenum internalformat) {
+        switch (internalformat) {
+        case GL_RGBA32F:
+        case GL_RGBA32UI:
+        case GL_RGBA32I:
+            return TextureViewClass::Bits128;
+        case GL_RGB32F:
+        case GL_RGB32UI:
+        case GL_RGB32I:
+            return TextureViewClass::Bits96;
+        case GL_RGBA16F:
+        case GL_RG32F:
+        case GL_RGBA16UI:
+        case GL_RG32UI:
+        case GL_RGBA16I:
+        case GL_RG32I:
+        case GL_RGBA16:
+        case GL_RGBA16_SNORM:
+            return TextureViewClass::Bits64;
+        case GL_RGB16:
+        case GL_RGB16_SNORM:
+        case GL_RGB16F:
+        case GL_RGB16UI:
+        case GL_RGB16I:
+            return TextureViewClass::Bits48;
+        case GL_RG16F:
+        case GL_R11F_G11F_B10F:
+        case GL_R32F:
+        case GL_RGB10_A2UI:
+        case GL_RGBA8UI:
+        case GL_RG16UI:
+        case GL_R32UI:
+        case GL_RGBA8I:
+        case GL_RG16I:
+        case GL_R32I:
+        case GL_RGB10_A2:
+        case GL_RGBA8:
+        case GL_RG16:
+        case GL_RGBA8_SNORM:
+        case GL_RG16_SNORM:
+        case GL_SRGB8_ALPHA8:
+        case GL_RGB9_E5:
+            return TextureViewClass::Bits32;
+        case GL_RGB8:
+        case GL_RGB8_SNORM:
+        case GL_SRGB8:
+        case GL_RGB8UI:
+        case GL_RGB8I:
+            return TextureViewClass::Bits24;
+        case GL_R16F:
+        case GL_RG8UI:
+        case GL_R16UI:
+        case GL_RG8I:
+        case GL_R16I:
+        case GL_RG8:
+        case GL_R16:
+        case GL_RG8_SNORM:
+        case GL_R16_SNORM:
+            return TextureViewClass::Bits16;
+        case GL_R8UI:
+        case GL_R8I:
+        case GL_R8:
+        case GL_R8_SNORM:
+            return TextureViewClass::Bits8;
+        case GL_COMPRESSED_RED_RGTC1:
+        case GL_COMPRESSED_SIGNED_RED_RGTC1:
+            return TextureViewClass::Rgtc1Red;
+        case GL_COMPRESSED_RG_RGTC2:
+        case GL_COMPRESSED_SIGNED_RG_RGTC2:
+            return TextureViewClass::Rgtc2Rg;
+        case GL_COMPRESSED_RGBA_BPTC_UNORM:
+        case GL_COMPRESSED_SRGB_ALPHA_BPTC_UNORM:
+            return TextureViewClass::BptcUnorm;
+        case GL_COMPRESSED_RGB_BPTC_SIGNED_FLOAT:
+        case GL_COMPRESSED_RGB_BPTC_UNSIGNED_FLOAT:
+            return TextureViewClass::BptcFloat;
+        default:
+            // Every depth/stencil format, every S3TC/ETC/ASTC format and every unsized format
+            // reaches here. The caller must then demand an EXACT format match.
+            return TextureViewClass::None;
+        }
+    }
+
+    // GL 4.6 core table 8.20 ("Legal texture targets for TextureView").
+    Bool IsLegalTextureViewTargetPair(TextureTarget origTarget, TextureTarget viewTarget) {
+        switch (origTarget) {
+        case TextureTarget::Texture1D:
+            return viewTarget == TextureTarget::Texture1D || viewTarget == TextureTarget::Texture1DArray;
+        case TextureTarget::Texture2D:
+            return viewTarget == TextureTarget::Texture2D || viewTarget == TextureTarget::Texture2DArray;
+        case TextureTarget::Texture3D:
+            return viewTarget == TextureTarget::Texture3D;
+        case TextureTarget::TextureCubeMap:
+            return viewTarget == TextureTarget::TextureCubeMap || viewTarget == TextureTarget::Texture2D ||
+                   viewTarget == TextureTarget::Texture2DArray || viewTarget == TextureTarget::TextureCubeMapArray;
+        case TextureTarget::TextureRectangle:
+            return viewTarget == TextureTarget::TextureRectangle;
+        case TextureTarget::Texture1DArray:
+            return viewTarget == TextureTarget::Texture1DArray || viewTarget == TextureTarget::Texture1D;
+        case TextureTarget::Texture2DArray:
+            return viewTarget == TextureTarget::Texture2DArray || viewTarget == TextureTarget::Texture2D ||
+                   viewTarget == TextureTarget::TextureCubeMap || viewTarget == TextureTarget::TextureCubeMapArray;
+        case TextureTarget::TextureCubeMapArray:
+            return viewTarget == TextureTarget::TextureCubeMapArray || viewTarget == TextureTarget::Texture2DArray ||
+                   viewTarget == TextureTarget::Texture2D || viewTarget == TextureTarget::TextureCubeMap;
+        case TextureTarget::Texture2DMultisample:
+        case TextureTarget::Texture2DMultisampleArray:
+            return viewTarget == TextureTarget::Texture2DMultisample ||
+                   viewTarget == TextureTarget::Texture2DMultisampleArray;
+        case TextureTarget::TextureBuffer:
+            // The table lists no legal target for a buffer texture: its storage is a buffer
+            // object, and there is nothing to make a view of.
+            return false;
+        default:
+            return false;
+        }
+    }
+
+    Uint RequiredTextureViewLayerCount(TextureTarget viewTarget) {
+        switch (viewTarget) {
+        case TextureTarget::TextureCubeMap:
+            return 6;
+        case TextureTarget::Texture1D:
+        case TextureTarget::Texture2D:
+        case TextureTarget::Texture3D:
+        case TextureTarget::TextureRectangle:
+        case TextureTarget::Texture2DMultisample:
+            return 1;
+        default:
+            // 1D/2D array, cube-map array, 2D multisample array: any count (the cube-map array's
+            // "multiple of 6" is checked by the caller).
+            return 0;
+        }
+    }
 } // namespace MobileGL::MG_Impl::GLImpl::TextureImpl

MobileGL/MG_Impl/GLImpl/Texture/Validators.h

@@ -20,6 +20,13 @@ namespace MobileGL::MG_Impl::GLImpl::TextureImpl {
     Bool ValidateTexturePixelDataType(TexturePixelDataType texturePixelDataType);
     Bool ValidateTextureLevelNumber(Int level);
     Bool ValidateTextureSizeWithTextureUploadTarget(TextureUploadTarget target, GLsizei width, GLsizei height);
+    // The two shape rules a cube-map-array level owes (GL 4.6 core 8.5): its faces are square, and
+    // its depth counts whole cubes. Both are GL_INVALID_VALUE. This used to be spelled inline in
+    // glTexStorage3D only, which is why glTexImage3D let both violations through - every entry
+    // point that DEFINES a cube-array level calls this now, so the two cannot drift again. A
+    // non-cube-array upload target answers true untouched.
+    Bool ValidateCubeMapArrayShape(TextureUploadTarget target, GLsizei width, GLsizei height, GLsizei depth,
+                                   const char* caller);
     Bool ValidateTextureSizeRange(Int width, Int height, Int depth);
     Bool ValidateTextureInternalFormat(TextureInternalFormat format);
     Bool ValidateTextureBorderNumber(Int border);
@@ -79,4 +86,33 @@ namespace MobileGL::MG_Impl::GLImpl::TextureImpl {
     // GL 4.6 SS 8.6 subset rule for glCopyTexImage*: the read buffer must supply every component
     // the requested internalformat asks for, but may supply more.
     Bool ValidateCopyTexImageBaseFormatSubset(TextureInternalFormat destFormat, TextureInternalFormat srcFormat);
+
+    // ---- glTextureView (ARB_texture_view / GL 4.6 core 8.18) ----
+    // Table 8.21's view classes. `None` is not a class - it means the format has NO entry in the
+    // table, which the spec turns into a much stricter rule than "same class": such a format can
+    // only ever be viewed as ITSELF. Every depth, stencil and depth/stencil format lands here,
+    // which is why the Better Clouds D24S8 view must name GL_DEPTH24_STENCIL8 exactly.
+    enum class TextureViewClass {
+        None = 0,
+        Bits128,
+        Bits96,
+        Bits64,
+        Bits48,
+        Bits32,
+        Bits24,
+        Bits16,
+        Bits8,
+        Rgtc1Red,
+        Rgtc2Rg,
+        BptcUnorm,
+        BptcFloat,
+    };
+    TextureViewClass GetTextureViewClass(GLenum internalformat);
+    // Table 8.20: which <target> values glTextureView accepts for a given origtexture target.
+    Bool IsLegalTextureViewTargetPair(TextureTarget origTarget, TextureTarget viewTarget);
+    // Table 8.20 again, read the other way: how many layers <target> requires. Returns 0 for the
+    // targets whose layer count is unconstrained (the array targets), 6 for GL_TEXTURE_CUBE_MAP,
+    // and 1 for every single-layer target. GL_TEXTURE_CUBE_MAP_ARRAY is special-cased by the
+    // caller because its constraint is "a multiple of 6", not an exact count.
+    Uint RequiredTextureViewLayerCount(TextureTarget viewTarget);
 } // namespace MobileGL::MG_Impl::GLImpl::TextureImpl

MobileGL/MG_Impl/GLImpl/VertexArray/Validators.cpp

@@ -12,15 +12,15 @@
 #include <MG_State/GLState/ErrorState/Error.h>
 #include <MG_Util/Converters/MGToGL/DataTypeConverter.h>
 #include <MG_Util/Converters/MGToStr/DataTypeConverter.h>
+#include <MG_Util/ShaderTranspiler/CompileEnv.h>
 
 namespace MobileGL::MG_Impl::GLImpl::VertexArrayImpl {
     Uint GetMaxVertexAttribs() {
-        constexpr Uint capacity = static_cast<Uint>(MG_State::GLState::VertexArrayObject::MAX_VERTEX_ATTRIBS);
-        if (!MG_Backend::pActiveBackendObject) return capacity;
-
-        const Int backendLimit = MG_Backend::pActiveBackendObject->GetDynamicParameters().MaxVertexAttribs;
-        if (backendLimit <= 0) return capacity;
-        return std::min(static_cast<Uint>(backendLimit), capacity);
+        // Shared with reflection's limit and with gl_MaxVertexAttribs; see ResolveMaxVertexAttribs.
+        const Bool hasBackend = MG_Backend::pActiveBackendObject != nullptr;
+        const Int backendLimit =
+            hasBackend ? MG_Backend::pActiveBackendObject->GetDynamicParameters().MaxVertexAttribs : 0;
+        return static_cast<Uint>(MG_Util::ShaderTranspiler::ResolveMaxVertexAttribs(hasBackend, backendLimit));
     }
 
     Uint GetMaxVertexAttribBindings() {

MobileGL/MG_IntegrationTest/CMakeLists.txt

@@ -63,8 +63,10 @@ add_executable(MobileGLIntegrationTest
     Scenarios/PipelineFailureScenario.cpp
     Scenarios/AdvertisedLimitsScenario.cpp
     Scenarios/PixelStoreSweepScenario.cpp
+    Scenarios/PrimitiveRestartScenario.cpp
     Scenarios/FragCoordOriginScenario.cpp
     Scenarios/ClearThenReadPixelsScenario.cpp
+    Scenarios/SampleVariablesScenario.cpp
     Scenarios/DepthStencilReadbackScenario.cpp
     Scenarios/DepthStencilReadbackMatrixScenario.cpp
     Scenarios/DepthStencilReadbackAttachmentShapeScenario.cpp
@@ -98,13 +100,21 @@ add_executable(MobileGLIntegrationTest
     Scenarios/VertexArrayEnableDisableScenario.cpp
     Scenarios/CopyImageLevelRangeScenario.cpp
     Scenarios/CopyImageLayeredScenario.cpp
+    Scenarios/TextureViewScenario.cpp
     Scenarios/PackedWordReadbackScenario.cpp
     Scenarios/LayeredAttachmentBarrierScenario.cpp
+    Scenarios/LayeredAttachmentShapeScenario.cpp
     Scenarios/LayeredTextureReadbackScenario.cpp
     Scenarios/AtomicCounterScenario.cpp
     Scenarios/SsboArrayDynamicIndexScenario.cpp
     Scenarios/StorageBufferRegrowScenario.cpp
+    Scenarios/SpirvShaderBinaryScenario.cpp
     Scenarios/RelinkStageSetScenario.cpp
+    Scenarios/GuiBatchScenario.cpp
+    Scenarios/UnboundImageDescriptorScenario.cpp
+    Scenarios/IntegerBorderColorScenario.cpp
+    Scenarios/ClearTexImageUndefinedLevelZeroScenario.cpp
+    Scenarios/RenderbufferBlendFormatScenario.cpp
 )
 
 target_include_directories(MobileGLIntegrationTest PRIVATE
@@ -303,6 +313,40 @@ mgl_itest_join_environment(MGL_ITEST_VULKAN_ASYNC_ENVIRONMENT
 mgl_itest_join_environment(MGL_ITEST_GLES_FORCED_DS_ENVIRONMENT
     "MOBILEGL_BACKEND_TYPE=DirectGLES" "MOBILEGL_ESPRYT_FORCE_DS_READBACK_EMULATION=1" ${MGL_ITEST_COMMON_ENV})
 
+# The shader-compiler configurations AsyncCompileScenario needs, and the one
+# ViewportArrayScenario's negative control needs.
+#
+# These used to be poked into MG_Config::Features from inside the test bodies. They
+# cannot be any more - on Android this module links the SHIPPING libMobileGL.so, which
+# exports nothing internal - and they should not have been anyway: half of what each of
+# them decides is latched before the first GL call (the compile pool and its threads;
+# the advertised extension list, which a backend builds once from the configuration in
+# force at its first use), so an in-process write could only ever have moved the other
+# half. Every one of them is a whole-process property, and a whole-process property is
+# spelled with an environment variable and a ctest entry of its own.
+#
+# Note the shape of every list here: it APPENDS to MGL_ITEST_COMMON_ENV /
+# MGL_ITEST_VULKAN_ENV rather than standing alone. A ctest ENVIRONMENT property REPLACES
+# the job environment rather than adding to it, so an entry that lists only its mode
+# variable would silently lose the EGL vendor and Vulkan ICD pinning and run against
+# whatever the loader found first.
+mgl_itest_join_environment(MGL_ITEST_GLES_ASYNC_ON_ENVIRONMENT
+    "MOBILEGL_BACKEND_TYPE=DirectGLES" "MOBILEGL_ASYNC_SHADER_COMPILE=1" ${MGL_ITEST_COMMON_ENV})
+mgl_itest_join_environment(MGL_ITEST_GLES_ASYNC_OFF_ENVIRONMENT
+    "MOBILEGL_BACKEND_TYPE=DirectGLES" "MOBILEGL_ASYNC_SHADER_COMPILE=0" ${MGL_ITEST_COMMON_ENV})
+mgl_itest_join_environment(MGL_ITEST_VULKAN_ASYNC_ON_ENVIRONMENT
+    "MOBILEGL_BACKEND_TYPE=DirectVulkan" "MOBILEGL_ASYNC_SHADER_COMPILE=1" ${MGL_ITEST_VULKAN_ENV})
+mgl_itest_join_environment(MGL_ITEST_VULKAN_ASYNC_OFF_ENVIRONMENT
+    "MOBILEGL_BACKEND_TYPE=DirectVulkan" "MOBILEGL_ASYNC_SHADER_COMPILE=0" ${MGL_ITEST_VULKAN_ENV})
+mgl_itest_join_environment(MGL_ITEST_GLES_OPTIMISTIC_ENVIRONMENT
+    "MOBILEGL_BACKEND_TYPE=DirectGLES" "MOBILEGL_ASYNC_SHADER_COMPILE=1"
+    "MOBILEGL_ASYNC_OPTIMISTIC_SHADER_STATUS=1" ${MGL_ITEST_COMMON_ENV})
+mgl_itest_join_environment(MGL_ITEST_VULKAN_OPTIMISTIC_ENVIRONMENT
+    "MOBILEGL_BACKEND_TYPE=DirectVulkan" "MOBILEGL_ASYNC_SHADER_COMPILE=1"
+    "MOBILEGL_ASYNC_OPTIMISTIC_SHADER_STATUS=1" ${MGL_ITEST_VULKAN_ENV})
+mgl_itest_join_environment(MGL_ITEST_GLES_NO_VIEWPORT_EMULATION_ENVIRONMENT
+    "MOBILEGL_BACKEND_TYPE=DirectGLES" "MOBILEGL_FORCE_VIEWPORT_ARRAY_EMULATION=0" ${MGL_ITEST_COMMON_ENV})
+
 # TIMEOUT on every entry: a GPU test that wedges must fail the run, not hang it.
 set(MGL_ITEST_TIMEOUT 120)
 
@@ -367,3 +411,100 @@ gtest_discover_tests(MobileGLIntegrationTest
         TIMEOUT ${MGL_ITEST_TIMEOUT}
         ENVIRONMENT "${MGL_ITEST_GLES_FORCED_DS_ENVIRONMENT}"
 )
+
+# AsyncCompileScenario, with asynchronous compilation PINNED ON per backend.
+#
+# Not a duplicate of what the two ambient registrations already run: they run whatever
+# MobileGL's built-in default happens to be, and the day that default flips they would
+# stop covering the asynchronous path without anything going red. These entries are the
+# ones that keep the asynchronous half tested no matter what ships. They are also the
+# only place ExtensionStringMatchesTheConfiguration can assert that the extension IS
+# advertised - the case derives its expectation from this variable and nothing else, and
+# skips where it is unset, precisely so that it is not asserting the implementation
+# against itself.
+gtest_discover_tests(MobileGLIntegrationTest
+    TEST_PREFIX "DirectGLES.AsyncOn."
+    TEST_FILTER "AsyncCompileScenario.*"
+    DISCOVERY_TIMEOUT 30
+    PROPERTIES
+        LABELS integration-gpu
+        TIMEOUT ${MGL_ITEST_TIMEOUT}
+        ENVIRONMENT "${MGL_ITEST_GLES_ASYNC_ON_ENVIRONMENT}"
+)
+gtest_discover_tests(MobileGLIntegrationTest
+    TEST_PREFIX "DirectVulkan.AsyncOn."
+    TEST_FILTER "AsyncCompileScenario.*"
+    DISCOVERY_TIMEOUT 30
+    PROPERTIES
+        LABELS integration-gpu
+        TIMEOUT ${MGL_ITEST_TIMEOUT}
+        ENVIRONMENT "${MGL_ITEST_VULKAN_ASYNC_ON_ENVIRONMENT}"
+)
+
+# The other side of the same switch: asynchronous compilation OFF, so
+# GL_KHR_parallel_shader_compile must be WITHDRAWN from both spellings of the extension
+# list and GL_MAX_SHADER_COMPILER_THREADS_KHR must read 0. Only that one case is
+# registered here because it is the only one that has anything to say in this
+# configuration - the other four exist to observe worker-built artifacts, and there are
+# none - so registering the whole scenario would buy four guaranteed skips per backend.
+# Together with the AsyncOn. entries above, one ctest run still covers both flag states,
+# which is what the in-process forcing used to be for.
+gtest_discover_tests(MobileGLIntegrationTest
+    TEST_PREFIX "DirectGLES.AsyncOff."
+    TEST_FILTER "AsyncCompileScenario.ExtensionStringMatchesTheConfiguration"
+    DISCOVERY_TIMEOUT 30
+    PROPERTIES
+        LABELS integration-gpu
+        TIMEOUT ${MGL_ITEST_TIMEOUT}
+        ENVIRONMENT "${MGL_ITEST_GLES_ASYNC_OFF_ENVIRONMENT}"
+)
+gtest_discover_tests(MobileGLIntegrationTest
+    TEST_PREFIX "DirectVulkan.AsyncOff."
+    TEST_FILTER "AsyncCompileScenario.ExtensionStringMatchesTheConfiguration"
+    DISCOVERY_TIMEOUT 30
+    PROPERTIES
+        LABELS integration-gpu
+        TIMEOUT ${MGL_ITEST_TIMEOUT}
+        ENVIRONMENT "${MGL_ITEST_VULKAN_ASYNC_OFF_ENVIRONMENT}"
+)
+
+# The optimistic-status quirk's end-to-end shape. Its own entries and not part of the
+# AsyncOn. ones because the quirk is not neutral for the rest of the scenario: with it in
+# force glGetShaderiv(GL_COMPILE_STATUS) deliberately answers without joining, which is
+# exactly what CompletionStatusPollingThenForcedJoin asserts must NOT happen. Off by
+# default and never advertised, so - unlike asynchronous compilation, which announces
+# itself through the extension string - the variable is the only thing that can tell the
+# case it is in force.
+gtest_discover_tests(MobileGLIntegrationTest
+    TEST_PREFIX "DirectGLES.OptimisticShaderStatus."
+    TEST_FILTER "AsyncCompileScenario.IrisShapedTwoPhaseBatchRendersCorrectly"
+    DISCOVERY_TIMEOUT 30
+    PROPERTIES
+        LABELS integration-gpu
+        TIMEOUT ${MGL_ITEST_TIMEOUT}
+        ENVIRONMENT "${MGL_ITEST_GLES_OPTIMISTIC_ENVIRONMENT}"
+)
+gtest_discover_tests(MobileGLIntegrationTest
+    TEST_PREFIX "DirectVulkan.OptimisticShaderStatus."
+    TEST_FILTER "AsyncCompileScenario.IrisShapedTwoPhaseBatchRendersCorrectly"
+    DISCOVERY_TIMEOUT 30
+    PROPERTIES
+        LABELS integration-gpu
+        TIMEOUT ${MGL_ITEST_TIMEOUT}
+        ENVIRONMENT "${MGL_ITEST_VULKAN_OPTIMISTIC_ENVIRONMENT}"
+)
+
+# The negative control for the DirectGLES gl_ViewportIndex emulation, in a process that
+# has it switched off. One case, because it is the only one the switch may touch: with
+# the emulation off the three positive cases in the same fixture describe behaviour the
+# backend does not have, so a whole-scenario registration would be three guaranteed reds.
+# DirectGLES only - the flag steers nothing on DirectVulkan, which routes natively.
+gtest_discover_tests(MobileGLIntegrationTest
+    TEST_PREFIX "DirectGLES.NoViewportArrayEmulation."
+    TEST_FILTER "ViewportArrayScenario.WithoutTheEmulationEveryIndexCollapsesOntoViewportZero"
+    DISCOVERY_TIMEOUT 30
+    PROPERTIES
+        LABELS integration-gpu
+        TIMEOUT ${MGL_ITEST_TIMEOUT}
+        ENVIRONMENT "${MGL_ITEST_GLES_NO_VIEWPORT_EMULATION_ENVIRONMENT}"
+)

MobileGL/MG_IntegrationTest/Harness/HeadlessGL.cpp

@@ -552,7 +552,15 @@ namespace MGITest {
         // before the pre-flight forks - the child must measure the same platform
         // the parent will use.
         EnsureHeadlessPlatform();
-        m_backendName = EnvOr("MOBILEGL_BACKEND_TYPE", "<unset>");
+        // The backend that is actually about to come up, which is what every
+        // `BackendName() == "DirectGLES"` gate in the scenarios means by the question.
+        // MG_ConfigLoader::InitBackendType defaults an unset MOBILEGL_BACKEND_TYPE to
+        // DirectGLES, so the same default belongs here; this used to report the literal
+        // "<unset>" instead. Under ctest the variable is always set by the ENVIRONMENT
+        // property, which is why that never showed - but run straight from a device
+        // shell, where nothing sets it, DirectGLES came up and every case gated on the
+        // NAME DirectGLES skipped as though it had not.
+        m_backendName = EnvOr("MOBILEGL_BACKEND_TYPE", "DirectGLES");
         m_usable = BringUp();
     }
 

MobileGL/MG_IntegrationTest/Harness/ScenarioFixture.h

@@ -21,12 +21,49 @@
 
 #pragma once
 
+#include <cctype>
+#include <cstdlib>
+#include <string>
+
 #include <gtest/gtest.h>
 
 #include "HeadlessGL.h"
 
 namespace MGITest {
 
+    // How a MOBILEGL_* quirk variable reads in THIS process's environment.
+    //
+    // A scenario that needs a non-default configuration takes it from here and skips
+    // when the process it was launched into is not in that configuration, rather than
+    // writing MG_Config::Features itself. Two reasons, and the second one decides it:
+    //
+    //   - the feature table is an internal symbol. On Android this module links against
+    //     the SHIPPING libMobileGL.so - deliberately, so the on-device run validates the
+    //     real artifact - and that library is built -fvisibility=hidden, so nothing
+    //     internal is reachable from here at all.
+    //   - a quirk poked in-process is already too late for everything latched at
+    //     initialization: the compile pool and its threads, and the backend's advertised
+    //     extension list, which is built once from the configuration in force at first
+    //     use. The process-wide variable is the only spelling that covers the whole
+    //     configuration instead of the half of it that is still mutable afterwards.
+    //
+    // The reading rule is MG_ConfigLoader's, character for character (ConfigLoader.cpp,
+    // QueryEnvQuirkOverride / IsTruthyValue): unset is Auto - device auto-detection or a
+    // built-in default, i.e. a value only the implementation knows - a truthy value is
+    // On, and anything else that IS set ("0", "false", "") is Off.
+    enum class AmbientQuirk { Auto, On, Off };
+
+    inline AmbientQuirk AmbientQuirkFromEnvironment(const char* name) {
+        const char* value = std::getenv(name);
+        if (value == nullptr) return AmbientQuirk::Auto;
+        std::string lowered(value);
+        for (char& c : lowered) {
+            c = static_cast<char>(std::tolower(static_cast<unsigned char>(c)));
+        }
+        if (lowered.empty() || lowered == "0" || lowered == "false") return AmbientQuirk::Off;
+        return AmbientQuirk::On;
+    }
+
     class ScenarioTest : public ::testing::Test {
     protected:
         void SetUp() override {

MobileGL/MG_IntegrationTest/Scenarios/AdvertisedLimitsScenario.cpp

@@ -56,7 +56,13 @@ namespace MGITest {
 
         const std::vector<LimitBound>& BufferLimitTable() {
             static const std::vector<LimitBound> table = {
-                {GL_MAX_UNIFORM_BUFFER_BINDINGS, "GL_MAX_UNIFORM_BUFFER_BINDINGS", 36, 256},
+                // 84 is the GL 4.5 core table 23.64 minimum, and also the width of the state
+                // layer's indexed-binding array - the two were made to coincide when the array
+                // was widened from 36, which had made the clamp in GL_Getter degenerate.
+                {GL_MAX_UNIFORM_BUFFER_BINDINGS, "GL_MAX_UNIFORM_BUFFER_BINDINGS", 84, 256},
+                // 14 uniform blocks on each of the FIVE graphics stages. The sum used to count
+                // three, and the two tessellation stages were simply missing from it.
+                {GL_MAX_COMBINED_UNIFORM_BLOCKS, "GL_MAX_COMBINED_UNIFORM_BLOCKS", 70, 256},
                 {GL_MAX_COMPUTE_UNIFORM_BLOCKS, "GL_MAX_COMPUTE_UNIFORM_BLOCKS", 12, 256},
                 {GL_MAX_COMPUTE_SHADER_STORAGE_BLOCKS, "GL_MAX_COMPUTE_SHADER_STORAGE_BLOCKS", 8, 256},
                 {GL_MAX_COMBINED_SHADER_STORAGE_BLOCKS, "GL_MAX_COMBINED_SHADER_STORAGE_BLOCKS", 8, 256},
@@ -133,6 +139,49 @@ namespace MGITest {
                     << relation.blocksName << " = " << blocks << " exceeds " << relation.bindingsName << " = "
                     << bindings << "; a shader may declare more blocks than there are binding points to bind them to";
             }
+
+            // THE MIDDLE TERM, which the relation quoted above always had and this case never
+            // checked. It is the one that actually broke: widening the binding-point array to 84
+            // raised what every PER-STAGE count clamps to, while the combined value was a
+            // five-stage sum of 70 - so a device reporting descriptor-indexing-scale uniform
+            // buffers (Adreno: maxPerStageDescriptorUniformBuffers = 16777216) advertised 84
+            // compute uniform blocks inside a combined limit of 70. Per-stage <= combined is
+            // exactly the assertion that says so, and it costs one glGetIntegerv per row.
+            struct StageAgainstCombined {
+                GLenum stage;
+                const char* stageName;
+                GLenum combined;
+                const char* combinedName;
+            };
+            const StageAgainstCombined stageRelations[] = {
+                {GL_MAX_COMPUTE_UNIFORM_BLOCKS, "GL_MAX_COMPUTE_UNIFORM_BLOCKS", GL_MAX_COMBINED_UNIFORM_BLOCKS,
+                 "GL_MAX_COMBINED_UNIFORM_BLOCKS"},
+                {GL_MAX_VERTEX_UNIFORM_BLOCKS, "GL_MAX_VERTEX_UNIFORM_BLOCKS", GL_MAX_COMBINED_UNIFORM_BLOCKS,
+                 "GL_MAX_COMBINED_UNIFORM_BLOCKS"},
+                {GL_MAX_TESS_CONTROL_UNIFORM_BLOCKS, "GL_MAX_TESS_CONTROL_UNIFORM_BLOCKS",
+                 GL_MAX_COMBINED_UNIFORM_BLOCKS, "GL_MAX_COMBINED_UNIFORM_BLOCKS"},
+                {GL_MAX_TESS_EVALUATION_UNIFORM_BLOCKS, "GL_MAX_TESS_EVALUATION_UNIFORM_BLOCKS",
+                 GL_MAX_COMBINED_UNIFORM_BLOCKS, "GL_MAX_COMBINED_UNIFORM_BLOCKS"},
+                {GL_MAX_GEOMETRY_UNIFORM_BLOCKS, "GL_MAX_GEOMETRY_UNIFORM_BLOCKS", GL_MAX_COMBINED_UNIFORM_BLOCKS,
+                 "GL_MAX_COMBINED_UNIFORM_BLOCKS"},
+                {GL_MAX_FRAGMENT_UNIFORM_BLOCKS, "GL_MAX_FRAGMENT_UNIFORM_BLOCKS", GL_MAX_COMBINED_UNIFORM_BLOCKS,
+                 "GL_MAX_COMBINED_UNIFORM_BLOCKS"},
+                {GL_MAX_COMPUTE_SHADER_STORAGE_BLOCKS, "GL_MAX_COMPUTE_SHADER_STORAGE_BLOCKS",
+                 GL_MAX_COMBINED_SHADER_STORAGE_BLOCKS, "GL_MAX_COMBINED_SHADER_STORAGE_BLOCKS"},
+                {GL_MAX_FRAGMENT_SHADER_STORAGE_BLOCKS, "GL_MAX_FRAGMENT_SHADER_STORAGE_BLOCKS",
+                 GL_MAX_COMBINED_SHADER_STORAGE_BLOCKS, "GL_MAX_COMBINED_SHADER_STORAGE_BLOCKS"},
+            };
+            for (const StageAgainstCombined& relation : stageRelations) {
+                GLint stage = -1;
+                GLint combined = -1;
+                glGetIntegerv(relation.stage, &stage);
+                glGetIntegerv(relation.combined, &combined);
+                ASSERT_EQ(FirstGLError(), GLenum(GL_NO_ERROR)) << relation.stageName;
+                EXPECT_LE(stage, combined)
+                    << relation.stageName << " = " << stage << " exceeds " << relation.combinedName << " = "
+                    << combined << "; GL 4.6 table 23.64 orders MAX_*_BUFFER_BINDINGS >= MAX_COMBINED_*_BLOCKS >= "
+                       "every per-stage count, and a single-stage program may use its whole per-stage allowance";
+            }
         }
 
         // KHR-GL44.multi_bind.functional_bind_buffers_range sizes each of an indexed target's
@@ -199,6 +248,80 @@ namespace MGITest {
                    "derived component limits are computed in";
         }
 
+        // The GL 4.5 core minimums that had no case in the getter at all, or that were still
+        // carrying an ES/GL3.3-tier number. Every one of these answered GL_INVALID_ENUM or a
+        // too-small value against a context advertising 4.6, and each is the FIRST call its
+        // conformance case makes - so the case died before it could measure anything.
+        //
+        // The cull pair is deliberately absent: zero is a legal answer there (a backend with no
+        // cull-distance route MUST report it), so it is checked for answerability only, below.
+        TEST_F(AdvertisedLimitsScenario, EveryGL45CoreMinimumIsMet) {
+            const std::vector<LimitBound> table = {
+                {GL_MAX_VARYING_VECTORS, "GL_MAX_VARYING_VECTORS", 15, 256},
+                {GL_MAX_VERTEX_UNIFORM_VECTORS, "GL_MAX_VERTEX_UNIFORM_VECTORS", 256, 1 << 20},
+                {GL_MAX_VARYING_COMPONENTS, "GL_MAX_VARYING_COMPONENTS", 60, 1 << 20},
+                // GL_MAX_VERTEX_STREAMS is deliberately absent. GL 4.5 requires 4 and MobileGL
+                // answers 1, which is a KNOWN non-conformance rather than an oversight: raising
+                // the number un-gates two transform-feedback CTS cases per package across
+                // KHR-GL40..GL46 that then fail, because no part of the shader pipeline supports
+                // layout(stream = N). See the GL_MAX_VERTEX_STREAMS case in GL_Getter.cpp. Adding
+                // a row here would pin a number the implementation cannot back.
+                {GL_MAX_GEOMETRY_SHADER_INVOCATIONS, "GL_MAX_GEOMETRY_SHADER_INVOCATIONS", 32, 256},
+                {GL_MAX_SUBROUTINES, "GL_MAX_SUBROUTINES", 256, 1 << 20},
+                {GL_MAX_SUBROUTINE_UNIFORM_LOCATIONS, "GL_MAX_SUBROUTINE_UNIFORM_LOCATIONS", 1024, 1 << 20},
+                {GL_MAX_TESS_CONTROL_INPUT_COMPONENTS, "GL_MAX_TESS_CONTROL_INPUT_COMPONENTS", 128, 1 << 16},
+                {GL_MAX_TESS_CONTROL_OUTPUT_COMPONENTS, "GL_MAX_TESS_CONTROL_OUTPUT_COMPONENTS", 128, 1 << 16},
+                {GL_MAX_TESS_CONTROL_TOTAL_OUTPUT_COMPONENTS, "GL_MAX_TESS_CONTROL_TOTAL_OUTPUT_COMPONENTS", 4096,
+                 1 << 20},
+                {GL_MAX_TESS_CONTROL_TEXTURE_IMAGE_UNITS, "GL_MAX_TESS_CONTROL_TEXTURE_IMAGE_UNITS", 16, 256},
+                {GL_MAX_TESS_CONTROL_UNIFORM_COMPONENTS, "GL_MAX_TESS_CONTROL_UNIFORM_COMPONENTS", 1024, 1 << 20},
+                {GL_MAX_TESS_CONTROL_UNIFORM_BLOCKS, "GL_MAX_TESS_CONTROL_UNIFORM_BLOCKS", 14, 256},
+                {GL_MAX_TESS_EVALUATION_INPUT_COMPONENTS, "GL_MAX_TESS_EVALUATION_INPUT_COMPONENTS", 128, 1 << 16},
+                {GL_MAX_TESS_EVALUATION_OUTPUT_COMPONENTS, "GL_MAX_TESS_EVALUATION_OUTPUT_COMPONENTS", 128, 1 << 16},
+                {GL_MAX_TESS_EVALUATION_TEXTURE_IMAGE_UNITS, "GL_MAX_TESS_EVALUATION_TEXTURE_IMAGE_UNITS", 16, 256},
+                {GL_MAX_TESS_EVALUATION_UNIFORM_COMPONENTS, "GL_MAX_TESS_EVALUATION_UNIFORM_COMPONENTS", 1024,
+                 1 << 20},
+                {GL_MAX_TESS_EVALUATION_UNIFORM_BLOCKS, "GL_MAX_TESS_EVALUATION_UNIFORM_BLOCKS", 14, 256},
+                {GL_MAX_TESS_PATCH_COMPONENTS, "GL_MAX_TESS_PATCH_COMPONENTS", 120, 1 << 16},
+                {GL_MAX_COMBINED_TESS_CONTROL_UNIFORM_COMPONENTS, "GL_MAX_COMBINED_TESS_CONTROL_UNIFORM_COMPONENTS",
+                 58368, 1 << 30},
+                {GL_MAX_COMBINED_TESS_EVALUATION_UNIFORM_COMPONENTS,
+                 "GL_MAX_COMBINED_TESS_EVALUATION_UNIFORM_COMPONENTS", 58368, 1 << 30},
+            };
+            for (const LimitBound& bound : table) {
+                GLint value = -424242;
+                glGetIntegerv(bound.pname, &value);
+                const unsigned int error = FirstGLError();
+                EXPECT_EQ(error, GLenum(GL_NO_ERROR)) << bound.name << " is not answerable: " << GLErrorName(error);
+                if (error != GL_NO_ERROR) continue;
+                EXPECT_GE(value, bound.minimum) << bound.name << " = " << value << " is below the GL 4.5 minimum "
+                                                << bound.minimum;
+                EXPECT_LE(value, bound.ceiling) << bound.name << " = " << value << " exceeds the ceiling "
+                                                << bound.ceiling;
+            }
+
+            // ARB_cull_distance's pair. Zero is honest on a backend with no cull-distance route,
+            // so only answerability and the combined-limit ordering are checked here.
+            GLint cull = -1;
+            GLint clip = -1;
+            GLint combined = -1;
+            glGetIntegerv(GL_MAX_CULL_DISTANCES, &cull);
+            glGetIntegerv(GL_MAX_CLIP_DISTANCES, &clip);
+            glGetIntegerv(GL_MAX_COMBINED_CLIP_AND_CULL_DISTANCES, &combined);
+            ASSERT_EQ(FirstGLError(), GLenum(GL_NO_ERROR)) << "the ARB_cull_distance queries must not error";
+            EXPECT_GE(cull, 0);
+            EXPECT_GE(combined, cull) << "GL 4.6 core 11.1.3.10: the combined limit is at least the cull one";
+            EXPECT_GE(combined, clip) << "GL 4.6 core 11.1.3.10: the combined limit is at least the clip one";
+
+            // GL_MAX_ELEMENT_INDEX is 64-bit state: the required 2^32-1 does not fit a GLint, so
+            // the wide query must answer it and the narrow one must saturate rather than wrap.
+            GLint64 elementIndex = -1;
+            glGetInteger64v(GL_MAX_ELEMENT_INDEX, &elementIndex);
+            ASSERT_EQ(FirstGLError(), GLenum(GL_NO_ERROR));
+            EXPECT_GE(elementIndex, static_cast<GLint64>(4294967295LL))
+                << "GL 4.5 core table 23.55 sets the GL_MAX_ELEMENT_INDEX minimum at 2^32-1";
+        }
+
         // ARB_viewport_array's own limits. They are advertised from three different places -
         // GL_MAX_VIEWPORTS from the frontend's indexed state width, the bounds range and the
         // subpixel bits from the backend caps table - and each backend fills that table from a

MobileGL/MG_IntegrationTest/Scenarios/AsyncCompileScenario.cpp

@@ -25,10 +25,12 @@
 //       be able to turn this into a red.
 //   (b) Forcing the join afterwards produces the right answer for every one of them:
 //       GL_COMPILE_STATUS true, an empty info log, and a program that links.
-//   (c) The extension string matches the configuration. This is the half a recorded
-//       trace can never cover - Iris and Sodium change their submission schedule the
-//       moment they see the string - so it is asserted against a real backend's real
-//       GL_EXTENSIONS, through both glGetString and glGetStringi.
+//   (c) The extension string matches the configuration - where "the configuration" is
+//       MOBILEGL_ASYNC_SHADER_COMPILE as this process inherited it, and NOT anything the
+//       implementation says about itself. This is the half a recorded trace can never
+//       cover - Iris and Sodium change their submission schedule the moment they see the
+//       string - so it is asserted against a real backend's real GL_EXTENSIONS, through
+//       both glGetString and glGetStringi.
 //   (d) glMaxShaderCompilerThreadsKHR(0) leaves nothing in flight: every subsequent
 //       GL_COMPLETION_STATUS_KHR reads GL_TRUE immediately, and compilation after it
 //       is synchronous. That is what the extension requires of a zero count.
@@ -40,6 +42,27 @@
 //
 // Backend selection is the module's usual one process, one backend (MOBILEGL_BACKEND_TYPE),
 // so this file runs twice per ctest invocation.
+//
+// COMPILATION MODE IS PER PROCESS TOO. Every case here needs a particular configuration of
+// MobileGL's shader compiler, and takes it from the ENVIRONMENT
+// (MOBILEGL_ASYNC_SHADER_COMPILE, MOBILEGL_ASYNC_OPTIMISTIC_SHADER_STATUS) rather than by
+// writing MG_Config::Features on the way past. Half of what those variables decide is
+// latched before the first GL call - the compile pool and its threads, and the advertised
+// extension list a backend builds once from the configuration in force at its first use -
+// so an in-process poke could only ever have moved the other half; and on Android it could
+// move nothing at all, because this module links against the shipping libMobileGL.so, which
+// exports no such symbol. A case whose process is not in the configuration it needs SKIPS
+// with that as its reason. CMakeLists.txt registers the extra ctest entries that put a
+// process into each configuration (AsyncOn., AsyncOff., OptimisticShaderStatus.), so one
+// ctest run still covers both sides of every switch. Run straight from a shell with nothing
+// set - the on-device shape - the ambient configuration runs and the rest skip cleanly.
+//
+// WITHIN one process, "compiled on a worker" versus "compiled on this thread" is switched
+// through glMaxShaderCompilerThreadsKHR, the extension's own entry point: a zero count joins
+// everything outstanding and compiles inline from then on, any nonzero count lifts that
+// again, and 0xFFFFFFFF asks for the implementation maximum (GL_Program.cpp,
+// MaxShaderCompilerThreadsKHR_State). Doing it through the public call rather than the
+// feature table means the switching is itself part of what these cases exercise.
 
 #include <string>
 #include <vector>
@@ -47,9 +70,6 @@
 #include "../Harness/HeadlessGL.h"
 #include "../Harness/ScenarioFixture.h"
 
-#include "Config.h"
-#include "MG_Util/Async/ShaderCompilePool.h"
-
 #ifdef GLAPI
 #undef GLAPI
 #endif
@@ -76,8 +96,6 @@ extern "C" void glMaxShaderCompilerThreadsKHR(GLuint count);
 namespace MGITest {
     namespace {
 
-        using MobileGL::MG_Config::QuirkOverride;
-
         // Same shape as the other scenarios: a two-attribute pass-through, so the only
         // thing that can differ between the two compilation modes is the compilation.
         constexpr const char* kVertexSource = R"(#version 330 core
@@ -139,50 +157,40 @@ void main() {
             return source;
         }
 
-        // MOBILEGL_ASYNC_SHADER_COMPILE decides the ambient mode; a scenario that wants
-        // the other one says so here and gets the ambient one back on scope exit. Forcing
-        // it in-process is what lets ONE ctest run compare the two modes against each
-        // other - the whole point of (e).
-        class AsyncModeScope {
-        public:
-            explicit AsyncModeScope(bool async) : m_saved(MobileGL::MG_Config::Features.AsyncShaderCompile) {
-                MobileGL::MG_Config::Features.AsyncShaderCompile =
-                    async ? QuirkOverride::ForceOn : QuirkOverride::ForceOff;
+        // Whether this context advertises GL_KHR_parallel_shader_compile, which is exactly
+        // "MobileGL is configured to compile asynchronously" as an application can see it:
+        // the backends gate the string on AsyncShaderCompileEnabled() and on nothing else
+        // (BackendObject_DirectGLES.cpp / BackendObject_DirectVulkan.cpp), and the string
+        // is the only way MobileGL ever tells anyone. A case that needs asynchronous
+        // compilation checks for it the way an application would, and skips without it.
+        //
+        // The INDEXED form, because that is the one a core-profile application reads.
+        bool HasParallelShaderCompile() {
+            GLint count = 0;
+            glGetIntegerv(GL_NUM_EXTENSIONS, &count);
+            for (GLint i = 0; i < count; ++i) {
+                const char* name = reinterpret_cast<const char*>(glGetStringi(GL_EXTENSIONS, GLuint(i)));
+                if (name != nullptr && std::string(name) == "GL_KHR_parallel_shader_compile") return true;
             }
-            ~AsyncModeScope() { MobileGL::MG_Config::Features.AsyncShaderCompile = m_saved; }
-            AsyncModeScope(const AsyncModeScope&) = delete;
-            AsyncModeScope& operator=(const AsyncModeScope&) = delete;
+            return false;
+        }
 
-        private:
-            const QuirkOverride m_saved;
-        };
-
-        // MOBILEGL_ASYNC_OPTIMISTIC_SHADER_STATUS, forced in-process for the same reason
-        // as AsyncModeScope: one ctest run asserts the quirk against the ambient default.
-        class OptimisticStatusScope {
-        public:
-            explicit OptimisticStatusScope(const QuirkOverride mode)
-                : m_saved(MobileGL::MG_Config::Features.AsyncOptimisticShaderStatus) {
-                MobileGL::MG_Config::Features.AsyncOptimisticShaderStatus = mode;
-            }
-            ~OptimisticStatusScope() { MobileGL::MG_Config::Features.AsyncOptimisticShaderStatus = m_saved; }
-            OptimisticStatusScope(const OptimisticStatusScope&) = delete;
-            OptimisticStatusScope& operator=(const OptimisticStatusScope&) = delete;
-
-        private:
-            const QuirkOverride m_saved;
-        };
-
-        // glMaxShaderCompilerThreadsKHR writes process-wide state; a scenario that calls
-        // it has to put the pool back or it changes how every scenario after it compiles.
+        // glMaxShaderCompilerThreadsKHR writes process-wide state; a scenario that calls it
+        // has to put the pool back or it changes how every scenario after it compiles.
+        //
+        // The restore is the extension's own "implementation maximum" spelling rather than a
+        // hand-rolled poke at the pool. glMaxShaderCompilerThreadsKHR(0xFFFFFFFF) is defined
+        // (GL_Program.cpp, MaxShaderCompilerThreadsKHR_State) as precisely the two steps this
+        // used to perform through internal entry points - concurrency := the pool's full
+        // thread count, then lift any suspension a zero count had armed - in the safer order,
+        // since it raises the budget before re-admitting work rather than after. Going through
+        // the public call also puts the restore path itself under test, and it is the only
+        // spelling available on Android, where this module links the shipping shared library
+        // and can reach nothing but the GL entry points.
         class CompilerThreadScope {
         public:
             CompilerThreadScope() = default;
-            ~CompilerThreadScope() {
-                MobileGL::MG_Util::Async::SetAsyncShaderCompileSuspended(false);
-                auto& pool = MobileGL::MG_Util::Async::ShaderCompilePool::Get();
-                pool.SetMaxConcurrency(pool.GetThreadCount());
-            }
+            ~CompilerThreadScope() { glMaxShaderCompilerThreadsKHR(0xFFFFFFFFu); }
             CompilerThreadScope(const CompilerThreadScope&) = delete;
             CompilerThreadScope& operator=(const CompilerThreadScope&) = delete;
         };
@@ -293,7 +301,12 @@ void main() {
         // interesting for shaders that (a) proved were genuinely still outstanding.
         TEST_F(AsyncCompileScenario, CompletionStatusPollingThenForcedJoin) {
             if (!Ready()) return;
-            const AsyncModeScope async(true);
+            if (!HasParallelShaderCompile()) {
+                GTEST_SKIP() << "this process is configured to compile inline "
+                                "(GL_KHR_parallel_shader_compile is not advertised), so no compile can be "
+                                "outstanding; the AsyncOn. ctest entries run this case with "
+                                "MOBILEGL_ASYNC_SHADER_COMPILE=1";
+            }
             const CompilerThreadScope threads;
             // One worker, so the queue behind it is what the poll observes.
             glMaxShaderCompilerThreadsKHR(1);
@@ -341,14 +354,33 @@ void main() {
         }
 
         // ---- (c) ------------------------------------------------------------------
-        // The extension string, read from a real backend that really brought a driver
-        // up. No mode forcing here: a backend builds its advertised list once, from the
-        // configuration in force at its first use, so the meaningful assertion is
-        // against the AMBIENT configuration - which is exactly what makes this case
-        // worth running in both of the suite's flag states.
+        // The extension string, read from a real backend that really brought a driver up.
+        //
+        // The expectation comes from the ENVIRONMENT, never from the implementation. This
+        // case used to derive it by calling AsyncShaderCompileEnabled() - which is the same
+        // function the backends gate the string on, so the two halves could only ever agree
+        // and the case would have passed however wrong both of them were. Asserting an
+        // implementation against itself pins nothing.
+        //
+        // MOBILEGL_ASYNC_SHADER_COMPILE is the whole input: the process inherited it before
+        // any GL call, a backend builds its advertised list once from the configuration in
+        // force at first use, and nothing in this process can move it afterwards. So reading
+        // the variable IS reading the configuration, independently. With the variable unset
+        // the configuration in force is MobileGL's built-in default, which only the
+        // implementation knows - there is nothing independent left to compare against, and
+        // this case says so rather than inventing an expectation. The AsyncOn. and AsyncOff.
+        // ctest entries pin the variable to each of its two values, so one ctest run still
+        // asserts both the advertised and the withdrawn side.
         TEST_F(AsyncCompileScenario, ExtensionStringMatchesTheConfiguration) {
             if (!Ready()) return;
-            const bool expected = MobileGL::MG_Util::Async::AsyncShaderCompileEnabled();
+            const AmbientQuirk configured = AmbientQuirkFromEnvironment("MOBILEGL_ASYNC_SHADER_COMPILE");
+            if (configured == AmbientQuirk::Auto) {
+                GTEST_SKIP() << "MOBILEGL_ASYNC_SHADER_COMPILE is unset, so the configuration in force is "
+                                "MobileGL's built-in default and the only way to learn it would be to ask "
+                                "the implementation this case exists to check; the AsyncOn. and AsyncOff. "
+                                "ctest entries run it with the variable pinned to each of its two values";
+            }
+            const bool expected = configured == AmbientQuirk::On;
 
             const char* extensions = reinterpret_cast<const char*>(glGetString(GL_EXTENSIONS));
             ASSERT_NE(extensions, nullptr);
@@ -385,7 +417,12 @@ void main() {
         // A zero count must leave nothing in flight and keep it that way.
         TEST_F(AsyncCompileScenario, ZeroCompilerThreadsSettlesEverythingImmediately) {
             if (!Ready()) return;
-            const AsyncModeScope async(true);
+            if (!HasParallelShaderCompile()) {
+                GTEST_SKIP() << "this process is configured to compile inline "
+                                "(GL_KHR_parallel_shader_compile is not advertised), so a zero count has "
+                                "nothing to settle; the AsyncOn. ctest entries run this case with "
+                                "MOBILEGL_ASYNC_SHADER_COMPILE=1";
+            }
             const CompilerThreadScope threads;
             glMaxShaderCompilerThreadsKHR(1);
 
@@ -417,12 +454,28 @@ void main() {
         // Compared through the DEFAULT framebuffer deliberately: that is where the
         // backend's orientation and present path live, so the comparison covers the
         // whole pipeline rather than the reflection tables alone.
+        //
+        // The two modes are selected through glMaxShaderCompilerThreadsKHR, the extension's
+        // own entry point, rather than through the feature table: a zero count joins
+        // everything outstanding and makes every later glCompileShader/glLinkProgram run its
+        // body on the calling thread, and 0xFFFFFFFF lifts that again with the pool at its
+        // full thread count (GL_Program.cpp, MaxShaderCompilerThreadsKHR_State; the compile
+        // and link paths both gate on AsyncShaderCompileActive(), which is what the zero
+        // count switches). So this is still one process comparing worker-built artifacts
+        // against inline-built ones - just asked for the way an application asks.
         TEST_F(AsyncCompileScenario, AsyncAndSyncProgramsRenderIdenticalFrames) {
             if (!Ready()) return;
+            if (!HasParallelShaderCompile()) {
+                GTEST_SKIP() << "this process is configured to compile inline "
+                                "(GL_KHR_parallel_shader_compile is not advertised), so both halves would "
+                                "be the same inline build and the comparison would be vacuous; the "
+                                "AsyncOn. ctest entries run this case with MOBILEGL_ASYNC_SHADER_COMPILE=1";
+            }
+            const CompilerThreadScope threads;
 
             Image asyncImage;
             {
-                const AsyncModeScope async(true);
+                glMaxShaderCompilerThreadsKHR(0xFFFFFFFFu);
                 const GLuint program = BuildProgram();
                 ASSERT_NE(program, 0u);
                 asyncImage = DrawFrameWith(program);
@@ -431,7 +484,7 @@ void main() {
 
             Image syncImage;
             {
-                const AsyncModeScope async(false);
+                glMaxShaderCompilerThreadsKHR(0);
                 const GLuint program = BuildProgram();
                 ASSERT_NE(program, 0u);
                 syncImage = DrawFrameWith(program);
@@ -456,11 +509,16 @@ void main() {
         // candidate) shows up here and not in the single-program case above.
         TEST_F(AsyncCompileScenario, ABatchOfAsyncProgramsAllRenderCorrectly) {
             if (!Ready()) return;
+            if (!HasParallelShaderCompile()) {
+                GTEST_SKIP() << "this process is configured to compile inline "
+                                "(GL_KHR_parallel_shader_compile is not advertised), so nothing would be "
+                                "built on a worker and there is no per-worker state to leak; the AsyncOn. "
+                                "ctest entries run this case with MOBILEGL_ASYNC_SHADER_COMPILE=1";
+            }
             constexpr int kPrograms = 12;
 
             std::vector<GLuint> programs;
             {
-                const AsyncModeScope async(true);
                 const CompilerThreadScope threads;
                 glMaxShaderCompilerThreadsKHR(1);
                 // Everything enqueued before anything is read: the only shape in which
@@ -489,6 +547,21 @@ void main() {
         // then mis-renders - shows up here as a wrong quadrant signature.
         TEST_F(AsyncCompileScenario, IrisShapedTwoPhaseBatchRendersCorrectly) {
             if (!Ready()) return;
+            // The quirk is off by default and never advertised, so unlike the cases above
+            // there is no GL observable that says whether it is in force - only the variable
+            // that put it there. It also has to be set BEFORE this process started for the
+            // shape to be the real one: the optimistic answer is latched per compile, and a
+            // quirk switched on mid-process would only cover the compiles after it.
+            if (AmbientQuirkFromEnvironment("MOBILEGL_ASYNC_OPTIMISTIC_SHADER_STATUS") != AmbientQuirk::On) {
+                GTEST_SKIP() << "this case is the optimistic-status quirk's end-to-end shape and needs it on "
+                                "for the whole process; the OptimisticShaderStatus. ctest entries run it with "
+                                "MOBILEGL_ASYNC_OPTIMISTIC_SHADER_STATUS=1";
+            }
+            if (!HasParallelShaderCompile()) {
+                GTEST_SKIP() << "the optimistic status only ever applies to a compile that is still in flight "
+                                "(OptimisticShaderStatusActive() requires AsyncShaderCompileActive()), and "
+                                "this process is configured to compile inline";
+            }
             constexpr int kPrograms = 12;
 
             // Distinct per program (so neither the source memo nor the adoption map turns
@@ -508,8 +581,6 @@ void main() {
 
             std::vector<GLuint> programs;
             {
-                const AsyncModeScope async(true);
-                const OptimisticStatusScope quirk(QuirkOverride::ForceOn);
                 const CompilerThreadScope threads;
                 glMaxShaderCompilerThreadsKHR(1);
 

MobileGL/MG_IntegrationTest/Scenarios/AtomicCounterScenario.cpp

@@ -87,11 +87,6 @@ void main() {
                                  << " and GL_MAX_COMPUTE_ATOMIC_COUNTER_BUFFERS is " << buffers
                                  << "; this needs 3 and 2";
                 }
-                if (!AtomicCountersAreWired()) {
-                    GTEST_SKIP() << "atomic counter buffers are not wired up on " << Gl().BackendName()
-                                 << " yet: glslang lowers them onto a storage block and that block's descriptor "
-                                 << "is still resolved from the shader-storage binding points";
-                }
                 m_program = CompileComputeProgram(kCounterComputeSource);
                 ASSERT_NE(m_program, 0u) << m_buildLog;
             }
@@ -105,12 +100,6 @@ void main() {
                 m_program = 0;
             }
 
-            // Magma binds the lowered block as an ordinary storage-buffer descriptor resolved
-            // from GL_SHADER_STORAGE_BUFFER point N, so the counter buffer never reaches it. The
-            // frontend half (limits, reflection queries, the link-time offset rules) is
-            // backend-agnostic and is covered by the unit suites; only the VALUE is scoped here.
-            bool AtomicCountersAreWired() const { return Gl().BackendName() != "DirectVulkan"; }
-
             unsigned int CompileComputeProgram(const char* source) {
                 const GLuint shader = glCreateShader(GL_COMPUTE_SHADER);
                 glShaderSource(shader, 1, &source, nullptr);

MobileGL/MG_IntegrationTest/Scenarios/ClearTexImageUndefinedLevelZeroScenario.cpp

@@ -0,0 +1,250 @@
+// MobileGL - MobileGL/MG_IntegrationTest/Scenarios/ClearTexImageUndefinedLevelZeroScenario.cpp
+// Copyright (c) 2025-2026 MobileGL-Dev
+// Licensed under the GNU Lesser General Public License v3.0:
+//   https://www.gnu.org/licenses/gpl-3.0.txt
+//   https://www.gnu.org/licenses/lgpl-3.0.txt
+// SPDX-License-Identifier: LGPL-3.0-only
+// End of Source File Header
+//
+// Scenario - glClearTexImage ON A TEXTURE WHOSE GL LEVEL 0 WAS NEVER DEFINED.
+//
+// KHR-GL4[456].clear_tex_image.* builds exactly one shape: fillTexture() issues ONE
+// glTexImage2D(GL_TEXTURE_2D, m_texLevel, ...) - the only texImage2D in the whole format/level
+// family - sets GL_TEXTURE_MAX_LEVEL to that level, clears it and reads it back with
+// glGetTexImage(..., m_texLevel, ...). For m_texLevel > 0 the levels BELOW the defined one have no
+// storage at all, and the split in the conformance results was on that alone: every texLevel_0 body
+// passed on DirectVulkan and every texLevel != 0 body failed, across all four internal formats and
+// all three entry points.
+//
+// The frontend understands this shape - the clear is a pure CPU-shadow write, and
+// ValidateTextureImageQuery deliberately does not demand mip completeness for a readback. The
+// Vulkan backend did not: VkTextureManager takes storage mip 0 as the physical image extent, so a
+// texture with no level 0 got no VkImage, SyncTextureAndGetDescriptor answered nullptr, and
+// VulkanRenderer::GetTextureImage took a silent early return - leaving the caller's buffer exactly
+// as it found it. The conformance failures carried no <Text> at all, because nothing raised a GL
+// error: the destination was simply never written, so the test compared its own zero-initialized
+// buffer against the clear value.
+//
+// The fix this pins is the readback fallback: with NO VkImage, nothing GPU-side can ever have
+// written the texture, so the CPU shadow IS its content and is the correct answer. It is gated on
+// "no image exists at all" and not on "syncing was inconvenient - a blanket shadow answer would
+// return stale bytes for every render-to-texture result instead.
+//
+// NOT covered here, and deliberately: such a texture still has no VkImage, so it remains invisible
+// to SAMPLING and rendering on DirectVulkan. Backing the image from the lowest defined level is a
+// separate change (it moves every GL-level-to-subresource translation in the backend); this
+// scenario asserts the readback contract only, and the DirectGLES leg - which has always been able
+// to define a lone level N - is the built-in control for what the answer should be.
+
+#include <array>
+#include <cstdint>
+#include <vector>
+
+#include "../Harness/HeadlessGL.h"
+#include "../Harness/ScenarioFixture.h"
+
+#ifdef GLAPI
+#undef GLAPI
+#endif
+#define GL_GLEXT_PROTOTYPES
+#include <GL/gl.h>
+#include <GL/glcorearb.h>
+#undef GL_GLEXT_PROTOTYPES
+
+namespace MGITest {
+    namespace {
+
+        // The conformance family's own shape: a mid-chain level of a texture that has nothing else.
+        constexpr GLint kDefinedLevel = 3;
+        constexpr GLsizei kLevelExtent = 8;
+
+        struct Texel8 {
+            GLubyte r = 0, g = 0, b = 0, a = 0;
+            bool operator==(const Texel8& other) const {
+                return r == other.r && g == other.g && b == other.b && a == other.a;
+            }
+        };
+
+        std::ostream& operator<<(std::ostream& os, const Texel8& c) {
+            return os << "rgba(" << int(c.r) << "," << int(c.g) << "," << int(c.b) << "," << int(c.a) << ")";
+        }
+
+        // The conformance test's clear value is a single repeated component; 5 is what it uses, and
+        // it is deliberately neither 0 (an unwritten destination) nor 255 (a saturated one).
+        constexpr Texel8 kClearValue{5, 5, 5, 5};
+        constexpr Texel8 kInitialValue{200, 100, 50, 255};
+
+        class ClearTexImageUndefinedLevelZeroScenario : public ScenarioTest {
+        protected:
+            void SetUp() override {
+                ScenarioTest::SetUp();
+                if (!Ready()) return;
+                DrainErrors();
+            }
+
+            void TearDown() override {
+                if (!Ready()) return;
+                if (m_texture != 0) {
+                    glBindTexture(GL_TEXTURE_2D, 0);
+                    glDeleteTextures(1, &m_texture);
+                    m_texture = 0;
+                }
+                DrainErrors();
+            }
+
+            static void DrainErrors() {
+                for (int i = 0; i < 16 && glGetError() != GL_NO_ERROR; ++i) {
+                }
+            }
+
+            // One level and nothing else, through glTexImage2D - deliberately NOT glTexStorage2D,
+            // which would define the whole chain and could not express "level 0 does not exist".
+            void MakeTextureWithOnlyLevel(GLint level) {
+                if (m_texture != 0) glDeleteTextures(1, &m_texture);
+                glGenTextures(1, &m_texture);
+                glBindTexture(GL_TEXTURE_2D, m_texture);
+                const std::vector<Texel8> initial(static_cast<std::size_t>(kLevelExtent) * kLevelExtent, kInitialValue);
+                glTexImage2D(GL_TEXTURE_2D, level, GL_RGBA8, kLevelExtent, kLevelExtent, 0, GL_RGBA, GL_UNSIGNED_BYTE,
+                             initial.data());
+                // What the conformance case does: MAX_LEVEL names the one level that exists, and
+                // BASE_LEVEL is left at its default 0 - which is what makes level 0 undefined AND
+                // nominally the base level, the shape the backend could not express.
+                glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAX_LEVEL, level);
+                glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
+                glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
+                ASSERT_EQ(FirstGLError(), 0u) << "texture setup with only level " << level;
+            }
+
+            std::vector<Texel8> ReadLevel(GLint level) {
+                std::vector<Texel8> pixels(static_cast<std::size_t>(kLevelExtent) * kLevelExtent, Texel8{0, 0, 0, 0});
+                glBindTexture(GL_TEXTURE_2D, m_texture);
+                glGetTexImage(GL_TEXTURE_2D, level, GL_RGBA, GL_UNSIGNED_BYTE, pixels.data());
+                EXPECT_EQ(FirstGLError(), 0u) << "glGetTexImage(level " << level << ") left a GL error behind";
+                return pixels;
+            }
+
+            void ExpectAllTexels(const char* what, const std::vector<Texel8>& pixels, Texel8 expected) {
+                std::size_t offenders = 0;
+                Texel8 firstBad{};
+                for (const Texel8& pixel : pixels) {
+                    if (pixel == expected) continue;
+                    if (offenders == 0) firstBad = pixel;
+                    ++offenders;
+                }
+                EXPECT_EQ(offenders, 0u) << what << ": got " << firstBad << " instead of " << expected << " ("
+                                         << offenders << " of " << pixels.size() << " texels wrong)";
+            }
+
+            // Level 0 defined, a GAP, then `level` defined. GL keeps the intervening levels at a zero
+            // extent, so the backend's mip walk stops at the gap and the VkImage ends up with FEWER
+            // mip levels than the GL level count - which is a different shape from "no image at all"
+            // and is why the readback has to bound the level against the IMAGE.
+            void MakeTextureWithAGapBefore(GLint level) {
+                if (m_texture != 0) glDeleteTextures(1, &m_texture);
+                glGenTextures(1, &m_texture);
+                glBindTexture(GL_TEXTURE_2D, m_texture);
+                const std::vector<Texel8> base(static_cast<std::size_t>(kLevelExtent) * kLevelExtent, kInitialValue);
+                glTexImage2D(GL_TEXTURE_2D, 0, GL_RGBA8, kLevelExtent, kLevelExtent, 0, GL_RGBA, GL_UNSIGNED_BYTE,
+                             base.data());
+                const std::vector<Texel8> gapped(static_cast<std::size_t>(kLevelExtent) * kLevelExtent, kInitialValue);
+                glTexImage2D(GL_TEXTURE_2D, level, GL_RGBA8, kLevelExtent, kLevelExtent, 0, GL_RGBA,
+                             GL_UNSIGNED_BYTE, gapped.data());
+                glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAX_LEVEL, level);
+                glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
+                glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
+                ASSERT_EQ(FirstGLError(), 0u) << "texture setup with a gap before level " << level;
+            }
+
+            GLuint m_texture = 0;
+        };
+
+    } // namespace
+
+    // The regression. Before the fix glGetTexImage wrote nothing at all on DirectVulkan, so the
+    // caller's buffer kept whatever it already held - which is why the conformance failures showed
+    // the test's own zero-initialized memory and carried no GL error.
+    TEST_F(ClearTexImageUndefinedLevelZeroScenario, ClearAndReadBackALevelWhoseLowerLevelsDoNotExist) {
+        if (!Ready()) GTEST_SKIP();
+        MakeTextureWithOnlyLevel(kDefinedLevel);
+
+        // Pre-flight: the level reads back as what was uploaded. This is what makes the assertion
+        // after the clear falsifiable - without it, a readback that silently wrote nothing could not
+        // be told from one that wrote the right answer.
+        ExpectAllTexels("before the clear", ReadLevel(kDefinedLevel), kInitialValue);
+
+        glClearTexImage(m_texture, kDefinedLevel, GL_RGBA, GL_UNSIGNED_BYTE, &kClearValue);
+        EXPECT_EQ(FirstGLError(), 0u) << "glClearTexImage was rejected";
+
+        ExpectAllTexels("after the clear", ReadLevel(kDefinedLevel), kClearValue);
+        Gl().EndFrame();
+    }
+
+    // The same shape through glClearTexSubImage, which is a separate entry point in the conformance
+    // family and failed on exactly the same bodies.
+    TEST_F(ClearTexImageUndefinedLevelZeroScenario, ClearSubImageOfALevelWhoseLowerLevelsDoNotExist) {
+        if (!Ready()) GTEST_SKIP();
+        MakeTextureWithOnlyLevel(kDefinedLevel);
+
+        glClearTexSubImage(m_texture, kDefinedLevel, 0, 0, 0, kLevelExtent, kLevelExtent, 1, GL_RGBA,
+                           GL_UNSIGNED_BYTE, &kClearValue);
+        EXPECT_EQ(FirstGLError(), 0u) << "glClearTexSubImage was rejected";
+
+        ExpectAllTexels("after the sub-image clear", ReadLevel(kDefinedLevel), kClearValue);
+        Gl().EndFrame();
+    }
+
+    // The negative control: an ORDINARY texture, whose level 0 does exist, must keep answering from
+    // the GPU image rather than being diverted onto the shadow. A fallback that fired unconditionally
+    // would pass the two tests above and this one too - but it would also hand back stale bytes for
+    // anything the GPU had written, which is why the partial-clear check below matters: the readback
+    // has to see a region the backend cleared and a region it did not, in one image.
+    TEST_F(ClearTexImageUndefinedLevelZeroScenario, AnOrdinaryLevelZeroTextureStillReadsBackCorrectly) {
+        if (!Ready()) GTEST_SKIP();
+        MakeTextureWithOnlyLevel(0);
+
+        ExpectAllTexels("before the clear", ReadLevel(0), kInitialValue);
+
+        // Clear only the left half, so the answer is neither "all initial" nor "all cleared".
+        glClearTexSubImage(m_texture, 0, 0, 0, 0, kLevelExtent / 2, kLevelExtent, 1, GL_RGBA, GL_UNSIGNED_BYTE,
+                           &kClearValue);
+        EXPECT_EQ(FirstGLError(), 0u) << "glClearTexSubImage was rejected";
+
+        const std::vector<Texel8> pixels = ReadLevel(0);
+        ASSERT_EQ(pixels.size(), static_cast<std::size_t>(kLevelExtent) * kLevelExtent);
+        for (int y = 0; y < kLevelExtent; ++y) {
+            for (int x = 0; x < kLevelExtent; ++x) {
+                const Texel8 expected = x < kLevelExtent / 2 ? kClearValue : kInitialValue;
+                const Texel8 actual = pixels[static_cast<std::size_t>(y) * kLevelExtent + x];
+                ASSERT_EQ(actual, expected) << "at (" << x << "," << y << ")";
+            }
+        }
+        Gl().EndFrame();
+    }
+
+    // The adjacent shape the first fix did NOT cover: level 0 defined, a gap, then the level being
+    // read. This one DOES get a VkImage - just one with fewer mip levels than GL thinks the texture
+    // has - so the "no VkImage" test passes and the GL level was written straight into
+    // imageSubresource.mipLevel and into a VkImageMemoryBarrier's baseMipLevel. An out-of-range
+    // subresource is a promise the driver takes at face value; the glCopyImageSubData path two
+    // functions away grew the same guard after it SIGSEGV'd inside the Adreno driver.
+    //
+    // The level being read really does hold its own data (the shadow is its only copy, since nothing
+    // ever uploaded it), so the correct answer is the uploaded bytes - not a decline.
+    TEST_F(ClearTexImageUndefinedLevelZeroScenario, ReadBackALevelSeparatedFromLevelZeroByAGap) {
+        if (!Ready()) GTEST_SKIP();
+        MakeTextureWithAGapBefore(kDefinedLevel);
+
+        ExpectAllTexels("before the clear", ReadLevel(kDefinedLevel), kInitialValue);
+
+        glClearTexImage(m_texture, kDefinedLevel, GL_RGBA, GL_UNSIGNED_BYTE, &kClearValue);
+        EXPECT_EQ(FirstGLError(), 0u) << "glClearTexImage was rejected";
+
+        ExpectAllTexels("after the clear", ReadLevel(kDefinedLevel), kClearValue);
+
+        // Level 0 is backed by the real image and must still read back from it, so the level bound is
+        // about the level and not about the texture.
+        ExpectAllTexels("level 0 after clearing level 3", ReadLevel(0), kInitialValue);
+        Gl().EndFrame();
+    }
+
+} // namespace MGITest

MobileGL/MG_IntegrationTest/Scenarios/GuiBatchScenario.cpp

@@ -0,0 +1,794 @@
+// MobileGL - MobileGL/MG_IntegrationTest/Scenarios/GuiBatchScenario.cpp
+// Copyright (c) 2026 MobileGL-Dev
+// Licensed under the GNU Lesser General Public License v3.0:
+//   https://www.gnu.org/licenses/gpl-3.0.txt
+//   https://www.gnu.org/licenses/lgpl-3.0.txt
+// SPDX-License-Identifier: LGPL-3.0-only
+// End of Source File Header
+//
+// Scenario - A GUI QUAD DRAWN THE WAY AcceleratedRendering DRAWS ONE.
+//
+// The mod replaces every GUI blit with a compute pass: the application writes unit-space
+// vertices into a persistently mapped SSBO, a compute shader multiplies them by a shared
+// transform into a second buffer, and that second buffer is then bound as GL_ARRAY_BUFFER of a
+// DSA vertex array and drawn with glDrawElementsBaseVertex through vanilla's position_tex_color
+// program. Every element of that is replayed here, one axis per test, so a failure names the
+// element that killed the quad rather than "the GUI is broken".
+//
+// The axis that mattered is MeshesBlockLeftUnbound. The mod's vertex-transform compute shader
+// declares SIX storage blocks, and the last of them - `Meshes`, the cache of pre-uploaded model
+// geometry - is read only when a vertex says it comes from a cached mesh. A batch of plain GUI
+// blits has no cached meshes, so the mod binds nothing at that point and the shader never reads
+// it. GL 4.6 core 7.8 is explicit that this is legal: a storage block with no buffer at its
+// binding point simply has no store.
+//
+// DirectVulkan used to refuse the whole descriptor set over it, and both SetupDraw and
+// DispatchCompute skip their work on that refusal - so the transform dispatch never ran, the
+// output vertex buffer kept whatever was in it, and every hotbar and container-screen background
+// quad came out degenerate. Items were unaffected because item geometry DOES come from cached
+// meshes, which is what made the bug look like "only the backgrounds disappear".
+//
+// The assertions are whole-region, not centre-pixel: a quad that survives with three stale
+// vertices still paints its centre.
+//
+// Reproduces on DirectVulkan only. DirectGLES forwards the unbound binding to the GLES driver,
+// which does what GL says, so it is the control - every test here must stay green on both.
+
+#include <cstring>
+#include <string>
+#include <vector>
+
+#include "../Harness/HeadlessGL.h"
+#include "../Harness/ScenarioFixture.h"
+
+#ifdef GLAPI
+#undef GLAPI
+#endif
+#define GL_GLEXT_PROTOTYPES
+#include <GL/gl.h>
+#include <GL/glcorearb.h>
+#undef GL_GLEXT_PROTOTYPES
+
+namespace MGITest {
+    namespace {
+
+        constexpr const char* kTransformComputeSource = R"(#version 460 core
+
+struct Vertex {
+    float   x;
+    float   y;
+    float   z;
+    float   u0;
+    float   v0;
+    uint    color;
+};
+
+struct VaryingData {
+    int     offset;
+    int     sharing;
+    int     mesh;
+    int     shouldCull;
+};
+
+struct SharingData {
+    mat4    transform;
+    mat3    normal;
+};
+
+layout(local_size_x = 128) in;
+
+layout(binding=0, std430) restrict readonly buffer VerticesIn {
+    Vertex verticesIn[];
+};
+
+layout(binding=1, std430) restrict writeonly buffer VerticesOut {
+    Vertex verticesOut[];
+};
+
+layout(binding=2, std430) restrict readonly buffer Sharings {
+    SharingData sharings[];
+};
+
+layout(binding=3, std430) restrict readonly buffer VaryingsIn {
+    VaryingData varyingsIn[];
+};
+
+layout(binding=4, std430) restrict writeonly buffer VaryingsOut {
+    VaryingData varyingsOut[];
+};
+
+layout(binding=5, std430) restrict readonly buffer Meshes {
+    Vertex meshVertices[];
+};
+
+layout(location=0) uniform uint vertexCount;
+layout(location=1) uniform uint vertexOffset;
+layout(location=2) uniform uint varyingOffset;
+
+void main() {
+    uint    indexIn     = gl_GlobalInvocationID.x;
+    uint    vertexOut   = indexIn + vertexOffset;
+    uint    varyingOut  = indexIn + varyingOffset;
+
+    if (indexIn >= vertexCount) {
+        return;
+    }
+
+    int     offset      = varyingsIn[indexIn]   .offset;
+    uint    reference   = indexIn - offset;
+    int     sharing     = varyingsIn[reference] .sharing;
+    int     mesh        = varyingsIn[reference] .mesh;
+
+    mat4    transformMatrix;
+
+    if (sharing != -1) {
+        transformMatrix = sharings[sharing].transform;
+    } else {
+        transformMatrix = mat4(1.0);
+    }
+
+    Vertex  vertexIn;
+    vec4    colorMesh;
+
+    if (mesh != -1) {
+        vertexIn    = meshVertices[mesh + offset];
+        colorMesh   = unpackUnorm4x8    (vertexIn.color);
+    } else {
+        vertexIn    = verticesIn[indexIn];
+        colorMesh   = vec4  (1.0);
+    }
+
+    vec4    colorIn = unpackUnorm4x8    (verticesIn[reference].color);
+
+    vec4    posOut      = transformMatrix   * vec4          (vertexIn.x, vertexIn.y, vertexIn.z, 1.0);
+    vec4    colorOut    = colorMesh         * colorIn;
+
+    verticesOut[vertexOut].x        = posOut.x;
+    verticesOut[vertexOut].y        = posOut.y;
+    verticesOut[vertexOut].z        = posOut.z;
+
+    verticesOut[vertexOut].u0       = vertexIn.u0;
+    verticesOut[vertexOut].v0       = vertexIn.v0;
+
+    verticesOut[vertexOut].color    = packUnorm4x8  (colorOut);
+
+    varyingsOut[varyingOut].offset      = offset;
+    varyingsOut[varyingOut].shouldCull  = varyingsIn[reference].shouldCull;
+}
+)";
+
+        // Vanilla position_tex_color, spelled the way MC ships it: #version 150, no explicit
+        // attribute locations (they come from glBindAttribLocation in format order) and the
+        // ProjMat/ModelViewMat pair the mod re-uploads through setDefaultUniforms.
+        constexpr const char* kBlitVertexSource = R"(#version 150
+in vec3 Position;
+in vec2 UV0;
+in vec4 Color;
+uniform mat4 ModelViewMat;
+uniform mat4 ProjMat;
+out vec2 texCoord0;
+out vec4 vertexColor;
+void main() {
+    gl_Position = ProjMat * ModelViewMat * vec4(Position, 1.0);
+    texCoord0 = UV0;
+    vertexColor = Color;
+}
+)";
+
+        constexpr const char* kBlitFragmentSource = R"(#version 150
+uniform sampler2D Sampler0;
+in vec2 texCoord0;
+in vec4 vertexColor;
+out vec4 fragColor;
+void main() {
+    fragColor = texture(Sampler0, texCoord0) * vertexColor;
+}
+)";
+
+        constexpr int kFboSize = 64;
+        constexpr int kShaderStorageRestoreRange = 9;
+        constexpr int kAtomicCounterRestoreRange = 1;
+        constexpr int kBuilders = 2;
+
+        struct GuiVertex {
+            float x, y, z;
+            float u, v;
+            std::uint32_t color;
+        };
+        static_assert(sizeof(GuiVertex) == 24, "POSITION_TEX_COLOR is 24 bytes");
+
+        struct VaryingData {
+            std::int32_t offset;
+            std::int32_t sharing;
+            std::int32_t mesh;
+            std::int32_t shouldCull;
+        };
+
+        struct IndexedBinding {
+            GLint buffer = 0;
+            GLint start = 0;
+            GLint size = 0;
+        };
+
+        // Which parts of the mod's real frame this replay reproduces. Each test flips exactly
+        // one on top of the baseline so a failure names the element that killed the quad.
+        struct Fidelity {
+            bool shortIndices = false;   // MC's AutoStorageIndexBuffer is USHORT at small counts
+            bool baseVertex = false;     // ...and the second builder draws at a base vertex
+            bool twoBuilders = false;    // two render types share one output buffer
+            bool blendAndDepth = false;  // TRANSLUCENT_TRANSPARENCY + LEQUAL_DEPTH_TEST
+            bool regrow = false;         // MutableBuffer.doExpand replaces the GL name
+            bool rewriteMapEachFrame = false;
+            bool skipRelayout = false;   // bindDrawBuffers() only re-lays-out when resized
+            bool leaveMeshesUnbound = false; // a batch with no server meshes never binds binding 5
+        };
+
+        class GuiBatchScenario : public ScenarioTest {
+        protected:
+            void SetUp() override {
+                ScenarioTest::SetUp();
+                if (!Ready()) return;
+
+                m_transform = CompileComputeProgram(kTransformComputeSource);
+                ASSERT_NE(m_transform, 0u) << m_buildLog;
+                m_blit = CompileBlitProgram();
+                ASSERT_NE(m_blit, 0u) << m_buildLog;
+
+                m_target = MakeColorFbo(kFboSize, kFboSize);
+                ASSERT_NE(m_target.fbo, 0u);
+
+                MakeTexture();
+                MakeIndexBuffers();
+                MakeAcceleratedBuffers();
+                m_laidOut = false;
+                ASSERT_EQ(FirstGLError(), 0u) << "setup raised a GL error";
+            }
+
+            void TearDown() override {
+                if (!Ready()) return;
+                glUseProgram(0);
+                glBindVertexArray(0);
+                glDisable(GL_BLEND);
+                glDisable(GL_DEPTH_TEST);
+                DestroyColorFbo(m_target);
+                if (m_transform) glDeleteProgram(m_transform);
+                if (m_blit) glDeleteProgram(m_blit);
+                if (m_texture) glDeleteTextures(1, &m_texture);
+                if (!m_buffers.empty()) glDeleteBuffers((GLsizei)m_buffers.size(), m_buffers.data());
+                if (!m_vaos.empty()) glDeleteVertexArrays((GLsizei)m_vaos.size(), m_vaos.data());
+                m_buffers.clear();
+                m_vaos.clear();
+            }
+
+            unsigned int CompileComputeProgram(const char* source) {
+                const GLuint shader = glCreateShader(GL_COMPUTE_SHADER);
+                glShaderSource(shader, 1, &source, nullptr);
+                glCompileShader(shader);
+                GLint compiled = 0;
+                glGetShaderiv(shader, GL_COMPILE_STATUS, &compiled);
+                if (compiled == GL_FALSE) {
+                    char log[4096] = {};
+                    glGetShaderInfoLog(shader, sizeof(log) - 1, nullptr, log);
+                    m_buildLog = std::string("compute shader did not compile: ") + log;
+                    glDeleteShader(shader);
+                    return 0;
+                }
+                const GLuint program = glCreateProgram();
+                glAttachShader(program, shader);
+                glLinkProgram(program);
+                glDeleteShader(shader);
+                GLint linked = 0;
+                glGetProgramiv(program, GL_LINK_STATUS, &linked);
+                if (linked == GL_FALSE) {
+                    char log[4096] = {};
+                    glGetProgramInfoLog(program, sizeof(log) - 1, nullptr, log);
+                    m_buildLog = std::string("compute program did not link: ") + log;
+                    glDeleteProgram(program);
+                    return 0;
+                }
+                return program;
+            }
+
+            GLuint CompileOne(GLenum stage, const char* source) {
+                const GLuint shader = glCreateShader(stage);
+                glShaderSource(shader, 1, &source, nullptr);
+                glCompileShader(shader);
+                GLint compiled = 0;
+                glGetShaderiv(shader, GL_COMPILE_STATUS, &compiled);
+                if (compiled == GL_FALSE) {
+                    char log[4096] = {};
+                    glGetShaderInfoLog(shader, sizeof(log) - 1, nullptr, log);
+                    m_buildLog = std::string("shader did not compile: ") + log;
+                    glDeleteShader(shader);
+                    return 0;
+                }
+                return shader;
+            }
+
+            // glBindAttribLocation in format order, exactly as ShaderInstance does it.
+            unsigned int CompileBlitProgram() {
+                const GLuint vs = CompileOne(GL_VERTEX_SHADER, kBlitVertexSource);
+                if (!vs) return 0;
+                const GLuint fs = CompileOne(GL_FRAGMENT_SHADER, kBlitFragmentSource);
+                if (!fs) return 0;
+                const GLuint program = glCreateProgram();
+                glAttachShader(program, vs);
+                glAttachShader(program, fs);
+                glBindAttribLocation(program, 0, "Position");
+                glBindAttribLocation(program, 1, "UV0");
+                glBindAttribLocation(program, 2, "Color");
+                glLinkProgram(program);
+                glDeleteShader(vs);
+                glDeleteShader(fs);
+                GLint linked = 0;
+                glGetProgramiv(program, GL_LINK_STATUS, &linked);
+                if (linked == GL_FALSE) {
+                    char log[4096] = {};
+                    glGetProgramInfoLog(program, sizeof(log) - 1, nullptr, log);
+                    m_buildLog = std::string("blit program did not link: ") + log;
+                    glDeleteProgram(program);
+                    return 0;
+                }
+                return program;
+            }
+
+            void MakeTexture() {
+                std::vector<std::uint8_t> pixels(4 * 4 * 4, 0);
+                for (int i = 0; i < 16; ++i) {
+                    pixels[i * 4 + 2] = 255;
+                    pixels[i * 4 + 3] = 255;
+                }
+                glGenTextures(1, &m_texture);
+                glActiveTexture(GL_TEXTURE0);
+                glBindTexture(GL_TEXTURE_2D, m_texture);
+                glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
+                glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
+                glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
+                glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
+                glTexImage2D(GL_TEXTURE_2D, 0, GL_RGBA8, 4, 4, 0, GL_RGBA, GL_UNSIGNED_BYTE, pixels.data());
+            }
+
+            GLuint NewBuffer() {
+                GLuint b = 0;
+                glCreateBuffers(1, &b);
+                m_buffers.push_back(b);
+                return b;
+            }
+
+            GLuint NewVao() {
+                GLuint v = 0;
+                glCreateVertexArrays(1, &v);
+                m_vaos.push_back(v);
+                return v;
+            }
+
+            void MakeIndexBuffers() {
+                std::uint32_t wide[12];
+                std::uint16_t narrow[12];
+                for (int quad = 0; quad < 2; ++quad) {
+                    const std::uint32_t base = (std::uint32_t)(quad * 4);
+                    const std::uint32_t pattern[6] = {base, base + 1, base + 2, base + 2, base + 3, base};
+                    for (int i = 0; i < 6; ++i) {
+                        wide[quad * 6 + i] = pattern[i];
+                        narrow[quad * 6 + i] = (std::uint16_t)pattern[i];
+                    }
+                }
+                m_wideIndices = NewBuffer();
+                glNamedBufferStorage(m_wideIndices, sizeof(wide), wide, 0);
+                m_narrowIndices = NewBuffer();
+                glNamedBufferStorage(m_narrowIndices, sizeof(narrow), narrow, 0);
+            }
+
+            static void SetupAttributes() {
+                glVertexAttribPointer(0, 3, GL_FLOAT, GL_FALSE, 24, (const void*)0);
+                glEnableVertexAttribArray(0);
+                glVertexAttribPointer(1, 2, GL_FLOAT, GL_FALSE, 24, (const void*)12);
+                glEnableVertexAttribArray(1);
+                glVertexAttribPointer(2, 4, GL_UNSIGNED_BYTE, GL_TRUE, 24, (const void*)20);
+                glEnableVertexAttribArray(2);
+            }
+
+            void WriteInputs() {
+                const GuiVertex unit[4] = {
+                    {0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0xFFFFFFFFu},
+                    {0.0f, 1.0f, 0.0f, 0.0f, 1.0f, 0xFFFFFFFFu},
+                    {1.0f, 1.0f, 0.0f, 1.0f, 1.0f, 0xFFFFFFFFu},
+                    {1.0f, 0.0f, 0.0f, 1.0f, 0.0f, 0xFFFFFFFFu},
+                };
+                for (int b = 0; b < kBuilders; ++b) {
+                    std::memcpy(m_inVertexMap[b], unit, sizeof(unit));
+                    VaryingData varyings[4];
+                    for (int i = 0; i < 4; ++i) {
+                        varyings[i].offset = i;
+                        varyings[i].sharing = b;
+                        varyings[i].mesh = -1;
+                        varyings[i].shouldCull = 0;
+                    }
+                    std::memcpy(m_inVaryingMap[b], varyings, sizeof(varyings));
+                }
+            }
+
+            void MakeAcceleratedBuffers() {
+                const GLbitfield persistent = GL_MAP_WRITE_BIT | GL_MAP_PERSISTENT_BIT | GL_MAP_COHERENT_BIT;
+                for (int b = 0; b < kBuilders; ++b) {
+                    m_inVertices[b] = NewBuffer();
+                    glNamedBufferStorage(m_inVertices[b], 4 * (GLsizeiptr)sizeof(GuiVertex), nullptr, persistent);
+                    m_inVertexMap[b] =
+                        glMapNamedBufferRange(m_inVertices[b], 0, 4 * (GLsizeiptr)sizeof(GuiVertex), persistent);
+                    m_inVaryings[b] = NewBuffer();
+                    glNamedBufferStorage(m_inVaryings[b], 4 * (GLsizeiptr)sizeof(VaryingData), nullptr, persistent);
+                    m_inVaryingMap[b] =
+                        glMapNamedBufferRange(m_inVaryings[b], 0, 4 * (GLsizeiptr)sizeof(VaryingData), persistent);
+                }
+
+                // Two SharingData entries: builder 0 lands left of centre, builder 1 right.
+                float sharing[56] = {};
+                const float tx[2] = {-0.9f, 0.1f};
+                for (int b = 0; b < 2; ++b) {
+                    float* m = sharing + b * 28;
+                    m[0] = 0.8f;
+                    m[5] = 1.0f;
+                    m[10] = 1.0f;
+                    m[15] = 1.0f;
+                    m[12] = tx[b];
+                    m[13] = -0.5f;
+                    m[16] = 1.0f;
+                    m[20] = 1.0f;
+                    m[24] = 1.0f;
+                }
+                m_sharings = NewBuffer();
+                glNamedBufferStorage(m_sharings, sizeof(sharing), nullptr, persistent);
+                void* r = glMapNamedBufferRange(m_sharings, 0, sizeof(sharing), persistent);
+                std::memcpy(r, sharing, sizeof(sharing));
+
+                m_outSize = 8 * (GLsizeiptr)sizeof(GuiVertex);
+                m_outVertices = NewBuffer();
+                glNamedBufferStorage(m_outVertices, m_outSize, nullptr, GL_DYNAMIC_STORAGE_BIT);
+                m_outVaryings = NewBuffer();
+                glNamedBufferStorage(m_outVaryings, 8 * (GLsizeiptr)sizeof(VaryingData), nullptr,
+                                     GL_DYNAMIC_STORAGE_BIT);
+                m_meshes = NewBuffer();
+                glNamedBufferStorage(m_meshes, 4 * (GLsizeiptr)sizeof(GuiVertex), nullptr, GL_DYNAMIC_STORAGE_BIT);
+
+                m_vao = NewVao();
+                WriteInputs();
+            }
+
+            std::vector<IndexedBinding> Record(GLenum b, GLenum s, GLenum z, int range) {
+                std::vector<IndexedBinding> saved((std::size_t)range);
+                for (int i = 0; i < range; ++i) {
+                    glGetIntegeri_v(b, (GLuint)i, &saved[(std::size_t)i].buffer);
+                    glGetIntegeri_v(s, (GLuint)i, &saved[(std::size_t)i].start);
+                    glGetIntegeri_v(z, (GLuint)i, &saved[(std::size_t)i].size);
+                }
+                return saved;
+            }
+
+            void Restore(GLenum target, const std::vector<IndexedBinding>& saved) {
+                for (std::size_t i = 0; i < saved.size(); ++i) {
+                    if (saved[i].start == 0 && saved[i].size == 0) {
+                        glBindBufferBase(target, (GLuint)i, (GLuint)saved[i].buffer);
+                    } else {
+                        glBindBufferRange(target, (GLuint)i, (GLuint)saved[i].buffer, saved[i].start,
+                                          saved[i].size);
+                    }
+                }
+            }
+
+            // MutableBuffer.doExpand: a NEW immutable store, copied from the old one, old one gone.
+            void RegrowOutputBuffer() {
+                const GLsizeiptr newSize = m_outSize * 2;
+                GLuint grown = 0;
+                glCreateBuffers(1, &grown);
+                glNamedBufferStorage(grown, newSize, nullptr, GL_DYNAMIC_STORAGE_BIT);
+                glCopyNamedBufferSubData(m_outVertices, grown, 0, 0, m_outSize);
+                glDeleteBuffers(1, &m_outVertices);
+                for (auto& b : m_buffers) {
+                    if (b == m_outVertices) b = grown;
+                }
+                m_outVertices = grown;
+                m_outSize = newSize;
+            }
+
+            void Frame(const Fidelity& f, int builders) {
+                if (f.rewriteMapEachFrame) WriteInputs();
+
+                // --- prepareBuffers() -------------------------------------------------
+                const std::vector<IndexedBinding> ssbo =
+                    Record(GL_SHADER_STORAGE_BUFFER_BINDING, GL_SHADER_STORAGE_BUFFER_START,
+                           GL_SHADER_STORAGE_BUFFER_SIZE, kShaderStorageRestoreRange);
+                const std::vector<IndexedBinding> counters =
+                    Record(GL_ATOMIC_COUNTER_BUFFER_BINDING, GL_ATOMIC_COUNTER_BUFFER_START,
+                           GL_ATOMIC_COUNTER_BUFFER_SIZE, kAtomicCounterRestoreRange);
+                GLint currentProgram = 0;
+                glGetIntegerv(GL_CURRENT_PROGRAM, &currentProgram);
+
+                glMemoryBarrier(GL_SHADER_STORAGE_BARRIER_BIT);
+                glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 1, m_outVertices);
+                glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 2, m_sharings);
+                glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 4, m_outVaryings);
+                glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 5, f.leaveMeshesUnbound ? 0 : m_meshes);
+                glUseProgram(m_transform);
+                for (int b = 0; b < builders; ++b) {
+                    glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 0, m_inVertices[b]);
+                    glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 3, m_inVaryings[b]);
+                    glProgramUniform1ui(m_transform, glGetUniformLocation(m_transform, "vertexCount"), 4u);
+                    glProgramUniform1ui(m_transform, glGetUniformLocation(m_transform, "vertexOffset"),
+                                        (GLuint)(4 * b));
+                    glProgramUniform1ui(m_transform, glGetUniformLocation(m_transform, "varyingOffset"),
+                                        (GLuint)(4 * b));
+                    glDispatchCompute(1, 1, 1);
+                }
+                glUseProgram(0);
+                glMemoryBarrier(GL_SHADER_STORAGE_BARRIER_BIT);
+                glUseProgram((GLuint)currentProgram);
+                Restore(GL_SHADER_STORAGE_BUFFER, ssbo);
+                Restore(GL_ATOMIC_COUNTER_BUFFER, counters);
+
+                if (f.regrow) {
+                    RegrowOutputBuffer();
+                    m_laidOut = false; // isResized() forces the relayout
+                }
+
+                // --- drawBuffers() ----------------------------------------------------
+                glMemoryBarrier(GL_VERTEX_ATTRIB_ARRAY_BARRIER_BIT | GL_ELEMENT_ARRAY_BARRIER_BIT |
+                                GL_COMMAND_BARRIER_BIT);
+                glBindVertexArray(m_vao);
+                if (!m_laidOut || !f.skipRelayout) {
+                    glBindBuffer(GL_ARRAY_BUFFER, m_outVertices);
+                    SetupAttributes();
+                    m_laidOut = true;
+                }
+
+                if (f.blendAndDepth) {
+                    glEnable(GL_BLEND);
+                    glBlendFuncSeparate(GL_SRC_ALPHA, GL_ONE_MINUS_SRC_ALPHA, GL_ONE, GL_ONE_MINUS_SRC_ALPHA);
+                    glEnable(GL_DEPTH_TEST);
+                    glDepthFunc(GL_LEQUAL);
+                }
+
+                const GLenum indexType = f.shortIndices ? GL_UNSIGNED_SHORT : GL_UNSIGNED_INT;
+                const GLuint indexBuffer = f.shortIndices ? m_narrowIndices : m_wideIndices;
+                const GLsizei indexStride = f.shortIndices ? 2 : 4;
+
+                for (int b = 0; b < builders; ++b) {
+                    glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, indexBuffer); // AutoStorageIndexBuffer.bind
+                    glUseProgram(m_blit);
+                    glActiveTexture(GL_TEXTURE0);
+                    glBindTexture(GL_TEXTURE_2D, m_texture);
+                    glUniform1i(glGetUniformLocation(m_blit, "Sampler0"), 0);
+                    UploadIdentityMatrices();
+                    if (f.baseVertex) {
+                        // The mod's BASEVERTEX path: every builder reads the SAME first six
+                        // indices and offsets the vertices with a base vertex.
+                        glDrawElementsBaseVertex(GL_TRIANGLES, 6, indexType, (const void*)0, 4 * b);
+                    } else {
+                        glDrawElements(GL_TRIANGLES, 6, indexType, (const void*)(intptr_t)(b * 6 * indexStride));
+                    }
+                    glUseProgram(0);
+                }
+
+                if (f.blendAndDepth) {
+                    glDisable(GL_BLEND);
+                    glDisable(GL_DEPTH_TEST);
+                }
+                glBindVertexArray(0);
+            }
+
+            void UploadIdentityMatrices() {
+                static const float identity[16] = {1, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1};
+                glUniformMatrix4fv(glGetUniformLocation(m_blit, "ModelViewMat"), 1, GL_FALSE, identity);
+                glUniformMatrix4fv(glGetUniformLocation(m_blit, "ProjMat"), 1, GL_FALSE, identity);
+            }
+
+            void ExpectQuads(const Image& image, int builders, const char* when) {
+                EXPECT_TRUE(RegionIsMostly(image, 6, 26, 19, 45, "blue", 0.0,
+                                           std::string("left accelerated quad, ") + when));
+                if (builders > 1) {
+                    EXPECT_TRUE(RegionIsMostly(image, 38, 58, 19, 45, "blue", 0.0,
+                                               std::string("right accelerated quad, ") + when));
+                }
+            }
+
+            void RunFrames(const Fidelity& f, int builders, int frames, const char* when) {
+                for (int i = 0; i < frames; ++i) {
+                    BindFbo(m_target);
+                    ClearTo(0.0f, 0.0f, 0.0f, 1.0f);
+                    Frame(f, builders);
+                    ASSERT_EQ(FirstGLError(), 0u) << "frame " << i << " raised a GL error (" << when << ")";
+                    const Image image = ReadPixels(kFboSize, kFboSize);
+                    ExpectQuads(image, builders, (std::string(when) + ", frame " + std::to_string(i)).c_str());
+                    Gl().EndFrame();
+                }
+            }
+
+            unsigned int m_transform = 0;
+            unsigned int m_blit = 0;
+            ColorFbo m_target{};
+            GLuint m_texture = 0;
+            GLuint m_wideIndices = 0;
+            GLuint m_narrowIndices = 0;
+            GLuint m_vao = 0;
+            GLuint m_inVertices[kBuilders] = {};
+            GLuint m_inVaryings[kBuilders] = {};
+            void* m_inVertexMap[kBuilders] = {};
+            void* m_inVaryingMap[kBuilders] = {};
+            GLuint m_sharings = 0;
+            GLuint m_outVertices = 0;
+            GLuint m_outVaryings = 0;
+            GLuint m_meshes = 0;
+            GLsizeiptr m_outSize = 0;
+            bool m_laidOut = false;
+            std::vector<GLuint> m_buffers;
+            std::vector<GLuint> m_vaos;
+            std::string m_buildLog;
+        };
+
+    } // namespace
+
+    TEST_F(GuiBatchScenario, Baseline) {
+        if (!Ready() || IsSkipped()) return;
+        RunFrames(Fidelity{}, 1, 1, "baseline");
+    }
+
+    TEST_F(GuiBatchScenario, ShortIndices) {
+        if (!Ready() || IsSkipped()) return;
+        Fidelity f;
+        f.shortIndices = true;
+        RunFrames(f, 1, 1, "short indices");
+    }
+
+    TEST_F(GuiBatchScenario, TwoBuildersWideIndices) {
+        if (!Ready() || IsSkipped()) return;
+        RunFrames(Fidelity{}, 2, 1, "two builders, wide indices");
+    }
+
+    TEST_F(GuiBatchScenario, TwoBuildersBaseVertex) {
+        if (!Ready() || IsSkipped()) return;
+        Fidelity f;
+        f.baseVertex = true;
+        RunFrames(f, 2, 1, "two builders, base vertex");
+    }
+
+    TEST_F(GuiBatchScenario, TwoBuildersShortIndicesBaseVertex) {
+        if (!Ready() || IsSkipped()) return;
+        Fidelity f;
+        f.baseVertex = true;
+        f.shortIndices = true;
+        RunFrames(f, 2, 1, "two builders, short indices, base vertex");
+    }
+
+    TEST_F(GuiBatchScenario, BlendAndDepth) {
+        if (!Ready() || IsSkipped()) return;
+        Fidelity f;
+        f.baseVertex = true;
+        f.shortIndices = true;
+        f.blendAndDepth = true;
+        RunFrames(f, 2, 1, "blend and depth");
+    }
+
+    TEST_F(GuiBatchScenario, ThreeFramesWithoutRelayout) {
+        if (!Ready() || IsSkipped()) return;
+        Fidelity f;
+        f.baseVertex = true;
+        f.shortIndices = true;
+        f.blendAndDepth = true;
+        f.skipRelayout = true;
+        f.rewriteMapEachFrame = true;
+        RunFrames(f, 2, 3, "three frames without relayout");
+    }
+
+    TEST_F(GuiBatchScenario, MeshesBlockLeftUnbound) {
+        if (!Ready() || IsSkipped()) return;
+        Fidelity f;
+        f.baseVertex = true;
+        f.shortIndices = true;
+        f.blendAndDepth = true;
+        f.leaveMeshesUnbound = true;
+        RunFrames(f, 2, 1, "Meshes block left unbound");
+    }
+
+    TEST_F(GuiBatchScenario, FullFidelityWithRegrowth) {
+        if (!Ready() || IsSkipped()) return;
+        Fidelity f;
+        f.baseVertex = true;
+        f.shortIndices = true;
+        f.blendAndDepth = true;
+        f.skipRelayout = true;
+        f.rewriteMapEachFrame = true;
+        RunFrames(f, 2, 1, "full fidelity, pre-growth");
+        f.regrow = true;
+        RunFrames(f, 2, 1, "full fidelity, growth frame");
+        f.regrow = false;
+        RunFrames(f, 2, 2, "full fidelity, post-growth");
+    }
+
+    namespace {
+
+        // Atomic counter blocks resolve through the SAME descriptor path as shader storage
+        // blocks (glslang rewrites every atomic_uint into a synthesized storage block), so an
+        // unbound GL_ATOMIC_COUNTER_BUFFER point loses the dispatch for exactly the same reason
+        // an unbound SSBO did. The mod reaches this with its INDIRECT draw method, whose culling
+        // shaders carry a counter the BASEVERTEX default never binds.
+        //
+        // The counter is INCREMENTED, not merely declared: an unreferenced one is optimised out
+        // before it ever reaches a descriptor, so a shader that only declares it proves nothing.
+        constexpr const char* kCounterComputeSource = R"(#version 430 core
+layout(local_size_x = 1) in;
+layout(binding = 0, offset = 0) uniform atomic_uint g_unbound;
+layout(std430, binding = 0) buffer Output { uint g_data[]; };
+void main() {
+    atomicCounterIncrement(g_unbound);
+    g_data[gl_GlobalInvocationID.x] = gl_GlobalInvocationID.x + 1u;
+}
+)";
+
+        class UnboundCounterBlockScenario : public ScenarioTest {
+        protected:
+            void SetUp() override {
+                ScenarioTest::SetUp();
+                if (!Ready()) return;
+                GLint counters = 0;
+                glGetIntegerv(GL_MAX_COMPUTE_ATOMIC_COUNTERS, &counters);
+                if (counters < 1) {
+                    GTEST_SKIP() << "GL_MAX_COMPUTE_ATOMIC_COUNTERS is " << counters;
+                }
+                const GLuint shader = glCreateShader(GL_COMPUTE_SHADER);
+                glShaderSource(shader, 1, &kCounterComputeSource, nullptr);
+                glCompileShader(shader);
+                GLint compiled = 0;
+                glGetShaderiv(shader, GL_COMPILE_STATUS, &compiled);
+                ASSERT_NE(compiled, GL_FALSE);
+                m_program = glCreateProgram();
+                glAttachShader(m_program, shader);
+                glLinkProgram(m_program);
+                glDeleteShader(shader);
+                GLint linked = 0;
+                glGetProgramiv(m_program, GL_LINK_STATUS, &linked);
+                ASSERT_NE(linked, GL_FALSE);
+                glGenBuffers(1, &m_buffer);
+            }
+
+            void TearDown() override {
+                if (!Ready()) return;
+                glUseProgram(0);
+                if (m_buffer) glDeleteBuffers(1, &m_buffer);
+                if (m_program) glDeleteProgram(m_program);
+            }
+
+            unsigned int m_program = 0;
+            GLuint m_buffer = 0;
+        };
+
+    } // namespace
+
+    TEST_F(UnboundCounterBlockScenario, ADeclaredButUnboundCounterDoesNotLoseTheDispatch) {
+        if (!Ready() || IsSkipped()) return;
+
+        constexpr int kElements = 4;
+        const std::vector<unsigned int> zeros((std::size_t)kElements, 0u);
+        glBindBuffer(GL_SHADER_STORAGE_BUFFER, m_buffer);
+        glBufferData(GL_SHADER_STORAGE_BUFFER, (GLsizeiptr)(zeros.size() * sizeof(unsigned int)), zeros.data(),
+                     GL_DYNAMIC_COPY);
+        glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 0, m_buffer);
+        // Nothing is bound at GL_ATOMIC_COUNTER_BUFFER point 0 on purpose.
+        glBindBufferBase(GL_ATOMIC_COUNTER_BUFFER, 0, 0);
+        ASSERT_EQ(FirstGLError(), 0u);
+
+        glUseProgram(m_program);
+        glDispatchCompute(kElements, 1, 1);
+        glMemoryBarrier(GL_BUFFER_UPDATE_BARRIER_BIT);
+        EXPECT_EQ(FirstGLError(), 0u) << "the dispatch raised a GL error";
+
+        std::vector<unsigned int> values((std::size_t)kElements, 0xDEADBEEFu);
+        glBindBuffer(GL_SHADER_STORAGE_BUFFER, m_buffer);
+        glGetBufferSubData(GL_SHADER_STORAGE_BUFFER, 0, (GLsizeiptr)(values.size() * sizeof(unsigned int)),
+                           values.data());
+        for (int i = 0; i < kElements; ++i) {
+            EXPECT_EQ(values[(std::size_t)i], (unsigned int)(i + 1))
+                << "element " << i << " came back as " << values[(std::size_t)i]
+                << "; zero everywhere means the whole dispatch was dropped over the unbound counter block";
+        }
+        glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 0, 0);
+    }
+} // namespace MGITest

MobileGL/MG_IntegrationTest/Scenarios/IntegerBorderColorScenario.cpp

@@ -0,0 +1,391 @@
+// MobileGL - MobileGL/MG_IntegrationTest/Scenarios/IntegerBorderColorScenario.cpp
+// Copyright (c) 2025-2026 MobileGL-Dev
+// Licensed under the GNU Lesser General Public License v3.0:
+//   https://www.gnu.org/licenses/gpl-3.0.txt
+//   https://www.gnu.org/licenses/lgpl-3.0.txt
+// SPDX-License-Identifier: LGPL-3.0-only
+// End of Source File Header
+//
+// Scenario - AN INTEGER GL_TEXTURE_BORDER_COLOR REACHES AN isampler2D AS AN INTEGER.
+//
+// KHR-GL46.texture_border_clamp.Texture2D{R32I,R32UI} (and the 2DArray/3D siblings) set the border
+// colour with glSamplerParameterIiv/Iuiv, sample outside the texture through an integer sampler and
+// expect the value back. MobileGL returned 1132396544 on Espryt - which is 0x437F0000, the IEEE-754
+// bits of 255.0f, i.e. the float border-colour register read through an integer sampler - and 0 on
+// Magma, where the border fell through to VK_BORDER_COLOR_FLOAT_TRANSPARENT_BLACK.
+//
+// Two independent halves, and this scenario covers both because it goes through the frontend:
+//
+//   * the STATE had no record of which entry point wrote the border colour. All three
+//     representations are kept numerically in step, so the value alone cannot say whether the
+//     application called glTexParameterfv or glTexParameterIiv.
+//   * each backend then had exactly one border-colour call site: glTexParameterfv /
+//     glSamplerParameterfv on DirectGLES, and a snap-to-one-of-four-predefined-values on
+//     DirectVulkan that never emitted the VK_BORDER_COLOR_INT_* family at all.
+//
+// The border value is deliberately outside every predefined VkBorderColor and outside anything a
+// float register could round-trip: (255, -1, 7, 3) is neither transparent black, nor opaque black,
+// nor opaque white, so on DirectVulkan it can only be delivered through VK_EXT_custom_border_color.
+// That makes the scenario a real test of the extension path on lavapipe rather than a palette hit.
+//
+// Both an integer image view and an integer border colour are involved, which is the other half of
+// the Vulkan rule: VK_BORDER_COLOR_FLOAT_* on an integer image view is undefined behaviour
+// regardless of the value, so even a border of (0,0,0,1) has to resolve to INT_OPAQUE_BLACK.
+// InsideTexelsAreUnaffected is what keeps that from being asserted vacuously.
+
+#include <array>
+#include <cstdint>
+#include <string>
+#include <vector>
+
+#include "../Harness/HeadlessGL.h"
+#include "../Harness/ScenarioFixture.h"
+
+#ifdef GLAPI
+#undef GLAPI
+#endif
+#define GL_GLEXT_PROTOTYPES
+#include <GL/gl.h>
+#include <GL/glcorearb.h>
+#undef GL_GLEXT_PROTOTYPES
+
+namespace MGITest {
+    namespace {
+
+        constexpr int kOutputWidth = 8;
+        constexpr int kOutputHeight = 8;
+
+        // The texture's own texel, and the border. Neither is a Vulkan palette entry, and the border
+        // is deliberately not derivable from the texel.
+        constexpr std::int32_t kInsideTexel[4] = {11, 22, 33, 44};
+        constexpr std::int32_t kBorderColor[4] = {255, -1, 7, 3};
+
+        constexpr const char* kVertexSource = R"(#version 330 core
+void main()
+{
+    switch (gl_VertexID)
+    {
+      case 0: gl_Position = vec4(-1.0, 1.0, 0.0, 1.0); break;
+      case 1: gl_Position = vec4( 1.0, 1.0, 0.0, 1.0); break;
+      case 2: gl_Position = vec4(-1.0,-1.0, 0.0, 1.0); break;
+      case 3: gl_Position = vec4( 1.0,-1.0, 0.0, 1.0); break;
+    }
+}
+)";
+
+        // One channel per draw, so a failure names the component that is wrong. The coordinate is a
+        // uniform rather than a literal so the same program serves the border sample and the inside
+        // sample and nothing can be constant-folded differently between them.
+        std::string FragmentSource(int channel) {
+            static const char* kChannels[4] = {"x", "y", "z", "w"};
+            return std::string("#version 330 core\n\nuniform isampler2D smp;\nuniform vec2 uCoord;\n\n"
+                               "out int out_color;\n\nvoid main()\n{\n    out_color = texture(smp, uCoord).") +
+                   kChannels[channel] + ";\n}\n";
+        }
+
+        class IntegerBorderColorScenario : public ScenarioTest {
+        protected:
+            void SetUp() override {
+                ScenarioTest::SetUp();
+                if (!Ready()) return;
+
+                // 2x2 RGBA32I. Integer textures are not filterable, so NEAREST is mandatory.
+                const std::int32_t texels[4][4] = {{kInsideTexel[0], kInsideTexel[1], kInsideTexel[2], kInsideTexel[3]},
+                                                  {kInsideTexel[0], kInsideTexel[1], kInsideTexel[2], kInsideTexel[3]},
+                                                  {kInsideTexel[0], kInsideTexel[1], kInsideTexel[2], kInsideTexel[3]},
+                                                  {kInsideTexel[0], kInsideTexel[1], kInsideTexel[2], kInsideTexel[3]}};
+                glGenTextures(1, &m_sourceTexture);
+                glBindTexture(GL_TEXTURE_2D, m_sourceTexture);
+                glTexStorage2D(GL_TEXTURE_2D, 1, GL_RGBA32I, 2, 2);
+                glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, 2, 2, GL_RGBA_INTEGER, GL_INT, texels);
+                glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
+                glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
+                glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_BORDER);
+                glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_BORDER);
+                ASSERT_EQ(FirstGLError(), 0u) << "source texture setup left a GL error behind";
+
+                // 8x8 R32I render target: an integer readback, so nothing is normalized on the way
+                // out and a wrong value is reported as the number it actually was.
+                glGenTextures(1, &m_outputTexture);
+                glBindTexture(GL_TEXTURE_2D, m_outputTexture);
+                glTexStorage2D(GL_TEXTURE_2D, 1, GL_R32I, kOutputWidth, kOutputHeight);
+                glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
+                glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
+                glGenFramebuffers(1, &m_fbo);
+                glBindFramebuffer(GL_FRAMEBUFFER, m_fbo);
+                glFramebufferTexture2D(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, m_outputTexture, 0);
+                ASSERT_EQ(glCheckFramebufferStatus(GL_FRAMEBUFFER), GLenum(GL_FRAMEBUFFER_COMPLETE));
+                glGenVertexArrays(1, &m_vao);
+                ASSERT_EQ(FirstGLError(), 0u) << "output framebuffer setup left a GL error behind";
+            }
+
+            void TearDown() override {
+                if (!Ready()) return;
+                if (m_sampler != 0) {
+                    glBindSampler(0, 0);
+                    glDeleteSamplers(1, &m_sampler);
+                    m_sampler = 0;
+                }
+                if (m_vao != 0) glDeleteVertexArrays(1, &m_vao);
+                if (m_fbo != 0) glDeleteFramebuffers(1, &m_fbo);
+                if (m_outputTexture != 0) glDeleteTextures(1, &m_outputTexture);
+                if (m_sourceTexture != 0) glDeleteTextures(1, &m_sourceTexture);
+                if (m_narrowTexture != 0) glDeleteTextures(1, &m_narrowTexture);
+                glBindFramebuffer(GL_FRAMEBUFFER, 0);
+            }
+
+            // Samples `coord` through the integer sampler and returns every texel the draw wrote.
+            std::vector<std::int32_t> RenderChannel(int channel, float coordX, float coordY) {
+                const std::string fragment = FragmentSource(channel);
+                std::string error;
+                const unsigned int program = CompileProgram(kVertexSource, fragment.c_str(), &error);
+                if (program == 0) {
+                    ADD_FAILURE() << "channel " << channel << ": program did not build: " << error;
+                    return {};
+                }
+
+                glBindFramebuffer(GL_FRAMEBUFFER, m_fbo);
+                glViewport(0, 0, kOutputWidth, kOutputHeight);
+                glDisable(GL_SCISSOR_TEST);
+                glDisable(GL_DEPTH_TEST);
+                // A clear value nothing under test can produce, so an undrawn target is not mistaken
+                // for a correct one.
+                const GLint clearValue[4] = {-559038737, 0, 0, 0};
+                glClearBufferiv(GL_COLOR, 0, clearValue);
+
+                glUseProgram(program);
+                glActiveTexture(GL_TEXTURE0);
+                glBindTexture(GL_TEXTURE_2D, m_sourceTexture);
+                glUniform1i(glGetUniformLocation(program, "smp"), 0);
+                glUniform2f(glGetUniformLocation(program, "uCoord"), coordX, coordY);
+                glBindVertexArray(m_vao);
+                glDrawArrays(GL_TRIANGLE_STRIP, 0, 4);
+                glBindVertexArray(0);
+
+                std::vector<std::int32_t> texels(static_cast<std::size_t>(kOutputWidth) * kOutputHeight, 0);
+                glReadPixels(0, 0, kOutputWidth, kOutputHeight, GL_RED_INTEGER, GL_INT, texels.data());
+                glUseProgram(0);
+                glDeleteProgram(program);
+                return texels;
+            }
+
+            void ExpectAllTexels(const char* what, int channel, std::int32_t expected,
+                                 const std::vector<std::int32_t>& texels) {
+                if (texels.empty()) return;
+                std::size_t offenders = 0;
+                std::int32_t firstBad = 0;
+                for (const std::int32_t texel : texels) {
+                    if (texel == expected) continue;
+                    if (offenders == 0) firstBad = texel;
+                    ++offenders;
+                }
+                EXPECT_EQ(offenders, 0u) << what << " component " << channel << " returned " << firstBad
+                                         << " instead of " << expected << " (" << offenders << " of " << texels.size()
+                                         << " texels wrong)";
+            }
+
+            // Every component of the border, in one place, so both the texture-object and the
+            // sampler-object case assert exactly the same thing.
+            void ExpectBorderIsDelivered(const char* what) {
+                for (int channel = 0; channel < 4; ++channel) {
+                    // (-0.5, -0.5) is a full texture width outside the image on both axes, so
+                    // CLAMP_TO_BORDER can only answer with the border colour.
+                    const std::vector<std::int32_t> texels = RenderChannel(channel, -0.5f, -0.5f);
+                    EXPECT_EQ(FirstGLError(), 0u) << what << ": the border draw left a GL error behind";
+                    ExpectAllTexels(what, channel, kBorderColor[channel], texels);
+                }
+            }
+
+            // A narrow-format source built on demand, for the clamp cases. Returns the texture, which
+            // the caller owns until TearDown deletes it through m_narrowTexture.
+            void MakeNarrowSource(GLenum internalFormat, GLenum clientFormat, const void* texels,
+                                  const GLint* border, bool borderIsUnsigned) {
+                glGenTextures(1, &m_narrowTexture);
+                glBindTexture(GL_TEXTURE_2D, m_narrowTexture);
+                glTexStorage2D(GL_TEXTURE_2D, 1, internalFormat, 2, 2);
+                glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, 2, 2, clientFormat,
+                                internalFormat == GL_R8UI ? GL_UNSIGNED_BYTE : GL_BYTE, texels);
+                glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
+                glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
+                glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_BORDER);
+                glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_BORDER);
+                if (borderIsUnsigned) {
+                    const GLuint asUnsigned[4] = {static_cast<GLuint>(border[0]), static_cast<GLuint>(border[1]),
+                                                  static_cast<GLuint>(border[2]), static_cast<GLuint>(border[3])};
+                    glTexParameterIuiv(GL_TEXTURE_2D, GL_TEXTURE_BORDER_COLOR, asUnsigned);
+                } else {
+                    glTexParameterIiv(GL_TEXTURE_2D, GL_TEXTURE_BORDER_COLOR, border);
+                }
+                ASSERT_EQ(FirstGLError(), 0u) << "narrow source setup left a GL error behind";
+            }
+
+            // The narrow sources are single-channel, so only component 0 carries anything, and the
+            // sampler declaration has to match the format's signedness.
+            std::vector<std::int32_t> RenderNarrowBorder(bool isUnsignedSampler) {
+                const std::string fragment =
+                    std::string("#version 330 core\n\nuniform ") + (isUnsignedSampler ? "usampler2D" : "isampler2D") +
+                    " smp;\nuniform vec2 uCoord;\n\nout int out_color;\n\nvoid main()\n{\n"
+                    "    out_color = int(texture(smp, uCoord).x);\n}\n";
+                std::string error;
+                const unsigned int program = CompileProgram(kVertexSource, fragment.c_str(), &error);
+                if (program == 0) {
+                    ADD_FAILURE() << "narrow-border program did not build: " << error;
+                    return {};
+                }
+                glBindFramebuffer(GL_FRAMEBUFFER, m_fbo);
+                glViewport(0, 0, kOutputWidth, kOutputHeight);
+                glDisable(GL_SCISSOR_TEST);
+                glDisable(GL_DEPTH_TEST);
+                const GLint clearValue[4] = {-559038737, 0, 0, 0};
+                glClearBufferiv(GL_COLOR, 0, clearValue);
+                glUseProgram(program);
+                glActiveTexture(GL_TEXTURE0);
+                glBindTexture(GL_TEXTURE_2D, m_narrowTexture);
+                glUniform1i(glGetUniformLocation(program, "smp"), 0);
+                glUniform2f(glGetUniformLocation(program, "uCoord"), -0.5f, -0.5f);
+                glBindVertexArray(m_vao);
+                glDrawArrays(GL_TRIANGLE_STRIP, 0, 4);
+                glBindVertexArray(0);
+                std::vector<std::int32_t> texels(static_cast<std::size_t>(kOutputWidth) * kOutputHeight, 0);
+                glReadPixels(0, 0, kOutputWidth, kOutputHeight, GL_RED_INTEGER, GL_INT, texels.data());
+                glUseProgram(0);
+                glDeleteProgram(program);
+                return texels;
+            }
+
+            GLuint m_sourceTexture = 0;
+            GLuint m_outputTexture = 0;
+            GLuint m_fbo = 0;
+            GLuint m_vao = 0;
+            GLuint m_sampler = 0;
+            GLuint m_narrowTexture = 0;
+        };
+
+    } // namespace
+
+    // The floor, and the control that keeps the two tests below from passing vacuously: an INSIDE
+    // sample has to fetch the texture's own texel. If this fails the sampler, the shader or the
+    // integer readback is broken and nothing about the border colour has been measured.
+    TEST_F(IntegerBorderColorScenario, InsideTexelsAreUnaffectedByTheBorderColour) {
+        if (!Ready()) GTEST_SKIP();
+
+        glBindTexture(GL_TEXTURE_2D, m_sourceTexture);
+        glTexParameterIiv(GL_TEXTURE_2D, GL_TEXTURE_BORDER_COLOR, kBorderColor);
+        ASSERT_EQ(FirstGLError(), 0u) << "glTexParameterIiv(GL_TEXTURE_BORDER_COLOR) was rejected";
+
+        for (int channel = 0; channel < 4; ++channel) {
+            const std::vector<std::int32_t> texels = RenderChannel(channel, 0.5f, 0.5f);
+            EXPECT_EQ(FirstGLError(), 0u) << "the inside draw left a GL error behind";
+            ExpectAllTexels("inside sample", channel, kInsideTexel[channel], texels);
+        }
+        Gl().EndFrame();
+    }
+
+    // The regression, texture-object spelling. glTexParameterIiv is the entry point the frontend
+    // already accepted and then flattened into the same FloatVec4 every other spelling wrote.
+    TEST_F(IntegerBorderColorScenario, TexParameterIivBorderColourSurvivesToAnIntegerSampler) {
+        if (!Ready()) GTEST_SKIP();
+
+        glBindTexture(GL_TEXTURE_2D, m_sourceTexture);
+        glTexParameterIiv(GL_TEXTURE_2D, GL_TEXTURE_BORDER_COLOR, kBorderColor);
+        ASSERT_EQ(FirstGLError(), 0u) << "glTexParameterIiv(GL_TEXTURE_BORDER_COLOR) was rejected";
+
+        ExpectBorderIsDelivered("glTexParameterIiv");
+        Gl().EndFrame();
+    }
+
+    // The regression, sampler-object spelling - which is the one the conformance cases actually use,
+    // and a separate code path in both backends (BackendSamplerObject::Sync on DirectGLES, and the
+    // sampler cache key on DirectVulkan, where a border colour that is not part of the key would
+    // alias two samplers that differ only in it).
+    TEST_F(IntegerBorderColorScenario, SamplerParameterIivBorderColourSurvivesToAnIntegerSampler) {
+        if (!Ready()) GTEST_SKIP();
+
+        glGenSamplers(1, &m_sampler);
+        ASSERT_NE(m_sampler, 0u);
+        glSamplerParameteri(m_sampler, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
+        glSamplerParameteri(m_sampler, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
+        glSamplerParameteri(m_sampler, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_BORDER);
+        glSamplerParameteri(m_sampler, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_BORDER);
+        glSamplerParameterIiv(m_sampler, GL_TEXTURE_BORDER_COLOR, kBorderColor);
+        ASSERT_EQ(FirstGLError(), 0u) << "sampler-object setup was rejected";
+
+        // The texture object carries a DIFFERENT border colour, so a pass here cannot come from the
+        // texture's own state leaking through: GL 4.6 core 8.10 says a bound sampler object's state
+        // wins over the texture's for every sampling parameter.
+        const std::int32_t decoyBorder[4] = {0, 0, 0, 0};
+        glBindTexture(GL_TEXTURE_2D, m_sourceTexture);
+        glTexParameterIiv(GL_TEXTURE_2D, GL_TEXTURE_BORDER_COLOR, decoyBorder);
+        glBindSampler(0, m_sampler);
+        ASSERT_EQ(FirstGLError(), 0u) << "binding the sampler object was rejected";
+
+        ExpectBorderIsDelivered("glSamplerParameterIiv");
+        glBindSampler(0, 0);
+        Gl().EndFrame();
+    }
+
+    // GL 4.6 core 8.14.2: "For floating-point and integer formats, border values are clamped to the
+    // representable range of the format." A border of 300 on a GL_R8I texture is 127, not 300 - and
+    // VK_BORDER_COLOR_INT_CUSTOM_EXT delivers whatever it is handed, with format VK_FORMAT_UNDEFINED
+    // there is nothing for the driver to clamp against, so the clamp has to happen before the value
+    // leaves MobileGL. DirectGLES gets it right for free (the ES driver knows the texture format),
+    // which is what makes this a cross-backend divergence and not only a spec one.
+    TEST_F(IntegerBorderColorScenario, ASignedIntegerBorderIsClampedToTheFormatsRepresentableRange) {
+        if (!Ready()) GTEST_SKIP();
+
+        const std::int8_t texels[4] = {1, 1, 1, 1};
+        const GLint border[4] = {300, 0, 0, 1};
+        MakeNarrowSource(GL_R8I, GL_RED_INTEGER, texels, border, /*borderIsUnsigned=*/false);
+
+        const std::vector<std::int32_t> sampled = RenderNarrowBorder(/*isUnsignedSampler=*/false);
+        EXPECT_EQ(FirstGLError(), 0u) << "the clamped-border draw left a GL error behind";
+        ExpectAllTexels("R8I border 300", 0, 127, sampled);
+        Gl().EndFrame();
+    }
+
+    // The reciprocal half, and the one that decides how the two integer forms relate: -1 written
+    // through glTexParameterIiv against an UNSIGNED format. GL 4.6 core 8.10 stores an "I"-form
+    // border unmodified with an integer internal data type and defines no sign conversion between
+    // the two integer forms, so the stored bits are reinterpreted in the sampled format's own
+    // signedness: 0xFFFFFFFF, clamped to the format's maximum of 255.
+    //
+    // That is the DRIVER's answer, established by running this case rather than by reading the spec:
+    // clamping to 0 is an equally defensible reading of the same paragraph, and DirectVulkan can be
+    // made to produce either - but DirectGLES forwards the value to the ES driver verbatim and cannot
+    // deviate, so choosing 0 would mean the same program sampling 0 on Magma and 255 on Espryt. The
+    // whole point of carrying the border colour's form is to stop that class of divergence, so the
+    // backends agree on the driver's answer.
+    //
+    // The clamp itself is still doing the work: without it the value reaches the driver as
+    // 0xFFFFFFFF against a format whose maximum is 255, with format VK_FORMAT_UNDEFINED and so
+    // nothing for the driver to clamp against.
+    TEST_F(IntegerBorderColorScenario, ANegativeBorderOnAnUnsignedFormatClampsToTheFormatsMaximum) {
+        if (!Ready()) GTEST_SKIP();
+
+        const std::uint8_t texels[4] = {1, 1, 1, 1};
+        const GLint border[4] = {-1, 0, 0, 1};
+        MakeNarrowSource(GL_R8UI, GL_RED_INTEGER, texels, border, /*borderIsUnsigned=*/false);
+
+        const std::vector<std::int32_t> sampled = RenderNarrowBorder(/*isUnsignedSampler=*/true);
+        EXPECT_EQ(FirstGLError(), 0u) << "the clamped-border draw left a GL error behind";
+        ExpectAllTexels("R8UI border -1", 0, 255, sampled);
+        Gl().EndFrame();
+    }
+
+    // The same clamp from the unambiguous side: a value written through the UNSIGNED form that is
+    // simply too large for the format. No sign reinterpretation is involved, so both backends and
+    // the spec agree that 5000 on a GL_R8UI texture is 255.
+    TEST_F(IntegerBorderColorScenario, AnOversizedUnsignedBorderIsClampedToTheFormatsMaximum) {
+        if (!Ready()) GTEST_SKIP();
+
+        const std::uint8_t texels[4] = {1, 1, 1, 1};
+        const GLint border[4] = {5000, 0, 0, 1};
+        MakeNarrowSource(GL_R8UI, GL_RED_INTEGER, texels, border, /*borderIsUnsigned=*/true);
+
+        const std::vector<std::int32_t> sampled = RenderNarrowBorder(/*isUnsignedSampler=*/true);
+        EXPECT_EQ(FirstGLError(), 0u) << "the clamped-border draw left a GL error behind";
+        ExpectAllTexels("R8UI border 5000", 0, 255, sampled);
+        Gl().EndFrame();
+    }
+
+} // namespace MGITest

MobileGL/MG_IntegrationTest/Scenarios/LayeredAttachmentShapeScenario.cpp

@@ -0,0 +1,962 @@
+// MobileGL - MobileGL/MG_IntegrationTest/Scenarios/LayeredAttachmentShapeScenario.cpp
+// Copyright (c) 2025-2026 MobileGL-Dev
+// Licensed under the GNU Lesser General Public License v3.0:
+//   https://www.gnu.org/licenses/gpl-3.0.txt
+//   https://www.gnu.org/licenses/lgpl-3.0.txt
+// SPDX-License-Identifier: LGPL-3.0-only
+// End of Source File Header
+//
+// Scenario - THE ATTACHMENT SHAPES A LAYERED FRAMEBUFFER CAN TAKE, AND THE ONE VIEW TYPE
+// VULKAN ACCEPTS FOR ALL OF THEM.
+//
+// glFramebufferTexture on a GL_TEXTURE_3D or a GL_TEXTURE_CUBE_MAP_ARRAY makes a LAYERED
+// framebuffer: one attachment that covers every slice / layer-face, addressed by a geometry
+// shader writing gl_Layer. Vulkan has exactly one legal spelling for that
+// (VUID-VkFramebufferCreateInfo-flags-04113: an attachment view must be VK_IMAGE_VIEW_TYPE_2D
+// or _2D_ARRAY), and DirectVulkan used to hand vkCreateFramebuffer the IMAGE's own view type
+// instead:
+//
+//   * GL_TEXTURE_3D  -> VK_IMAGE_VIEW_TYPE_3D. A 3D image has arrayLayers == 1 and keeps its
+//     layers on z, so the layer-span guard measured [0, depth) against 1, refused, and returned
+//     VK_NULL_HANDLE - which then went into pAttachments as a null handle.
+//   * GL_TEXTURE_CUBE_MAP_ARRAY -> VK_IMAGE_VIEW_TYPE_CUBE_ARRAY. A perfectly valid view, of a
+//     type no framebuffer may take. The driver dereferenced or rejected it inside
+//     vkCreateFramebuffer.
+//
+// Both exits were guarded only by MOBILEGL_ASSERT, which an INFO build (the production and CTS
+// default) compiles to nothing - so both were process kills, not wrong pixels: 51 lost QPA
+// records over 7 conformance bodies, one runner restart each.
+//
+// The same function is what routes a NON-layered slice of a 3D texture
+// (glFramebufferTextureLayer), and it had the mirror-image hole: it asked for a 3D view there
+// too, so the per-slice branch that exists for exactly this case was unreachable and every
+// slice above z = 0 came back VK_NULL_HANDLE.
+//
+// The seven cases below are those shapes - layered 3D, one 3D slice, layered cube-map array with
+// its depth and packed depth-stencil attachments, and (cases 6 and 7) a layered cube MAP and 1D
+// ARRAY whose queued glClear is consumed outside a render pass. Each one asserts LAYER ROUTING,
+// not merely survival: what a layer receives is a function of its own index, so an attachment that
+// collapsed onto layer 0, or attached one face of a cube, fails on the layers it did not reach
+// rather than passing quietly. Every texture is seeded with a poison value first, so "the draw
+// never landed here" reads differently from "the wrong layer landed here".
+//
+// One of them turned out not to be a DirectVulkan bug at all. glFramebufferTexture on
+// GL_DEPTH_STENCIL_ATTACHMENT is a shorthand the front end splits into a depth and a stencil
+// attachment, and the split dropped the call's `layered` flag - so a layered colour attachment
+// sat beside a non-layered depth/stencil one and BOTH backends silently lost the draw. That is
+// the shape texture_cube_map_array.stencil_attachments_*_layered and
+// geometry_shader.layered_framebuffer.stencil_support are built on, and it is why they fail on
+// Espryt as well as crashing on Magma. Case (5) is what found it.
+//
+// DirectGLES is the control: it hands the same GL calls to the driver, so a red on both backends
+// means the scenario is wrong - or the defect is in the shared front end, as it was above - and a
+// red on DirectVulkan alone means Magma is.
+
+#include <cstddef>
+#include <cstdlib>
+#include <string>
+#include <vector>
+
+#include "../Harness/HeadlessGL.h"
+#include "../Harness/ScenarioFixture.h"
+
+#ifdef GLAPI
+#undef GLAPI
+#endif
+#define GL_GLEXT_PROTOTYPES
+#include <GL/gl.h>
+#include <GL/glcorearb.h>
+#undef GL_GLEXT_PROTOTYPES
+
+namespace MGITest {
+    namespace {
+
+        constexpr int kExtent = 4;
+        // Four z slices: enough that "only slice 0 was written" and "the whole thing was written"
+        // are different answers, and small enough that the geometry shader stays well inside
+        // GL_MAX_GEOMETRY_OUTPUT_VERTICES.
+        constexpr int k3DSlices = 4;
+        // Two cubes. One cube would let "attached a single cube" pass; twelve layer-faces would
+        // not.
+        constexpr int kCubeLayerFaces = 12;
+        // The slice a non-layered 3D attachment names. Not 0: slice 0 is the one address that is
+        // right whether or not the slice is resolved at all.
+        constexpr int kSubjectSlice = 2;
+
+        // Layers of the 1D array whose clear the last case checks. Its layer count lives in the
+        // state-side HEIGHT, not in z, which is the whole reason it is here.
+        constexpr int kOneDArrayLayers = 4;
+
+        // A colour no pass paints, uploaded before every draw. A layer that reads it back was
+        // never rendered to.
+        constexpr GLubyte kPoison = 0xAB;
+
+        // The glClear colour the two materialise cases use. Chosen as exact 8-bit values and fed
+        // to glClearColor as n/255, so the round trip through a UNORM8 target is lossless and a
+        // mismatch means a real miss rather than rounding.
+        constexpr Rgba8 kClearColor{17, 68, 187, 255};
+
+        // What pass `pass` paints on layer `layer`. r and g name the LAYER (so a mis-routed write
+        // says which layer it came from) and b names the PASS (so "the second draw was not
+        // rejected" is distinguishable from "the first draw never happened").
+        Rgba8 ExpectedColor(int layer, int pass) {
+            return {static_cast<GLubyte>(10 + layer * 20), static_cast<GLubyte>(200 - layer * 10),
+                    static_cast<GLubyte>(3 + pass * 60), 255};
+        }
+
+        std::string Describe(const Rgba8& color) {
+            return "(" + std::to_string(color.r) + ", " + std::to_string(color.g) + ", " +
+                   std::to_string(color.b) + ", " + std::to_string(color.a) + ")";
+        }
+
+        // A full-viewport triangle built from gl_VertexID, so nothing here needs a vertex buffer
+        // and the draw cannot fail for a reason that has nothing to do with the attachment.
+        // u_depth is the NDC z the whole primitive sits at - the depth/stencil case needs two
+        // different ones.
+        const char* const kVertexSource = R"(#version 420 core
+uniform float u_depth;
+void main()
+{
+    vec2 corner = vec2((gl_VertexID == 1) ? 3.0 : -1.0, (gl_VertexID == 2) ? 3.0 : -1.0);
+    gl_Position = vec4(corner, u_depth, 1.0);
+}
+)";
+
+        // The layer count is baked in as a literal rather than passed as a uniform: a
+        // non-constant loop bound in a geometry shader is legal but is one more thing the
+        // ESSL transpile could get wrong, and this scenario is not about that.
+        std::string MakeGeometrySource(int layerCount) {
+            return "#version 420 core\n"
+                   "layout(triangles) in;\n"
+                   "layout(triangle_strip, max_vertices = " +
+                   std::to_string(layerCount * 3) +
+                   ") out;\n"
+                   "flat out int v_layer;\n"
+                   "void main()\n"
+                   "{\n"
+                   "    for (int layer = 0; layer < " +
+                   std::to_string(layerCount) +
+                   "; ++layer) {\n"
+                   "        for (int i = 0; i < 3; ++i) {\n"
+                   "            gl_Layer = layer;\n"
+                   "            v_layer = layer;\n"
+                   "            gl_Position = gl_in[i].gl_Position;\n"
+                   "            EmitVertex();\n"
+                   "        }\n"
+                   "        EndPrimitive();\n"
+                   "    }\n"
+                   "}\n";
+        }
+
+        const char* const kLayeredFragmentSource = R"(#version 420 core
+flat in int v_layer;
+uniform int u_pass;
+out vec4 o_color;
+void main()
+{
+    o_color = vec4(float(10 + v_layer * 20) / 255.0,
+                   float(200 - v_layer * 10) / 255.0,
+                   float(3 + u_pass * 60) / 255.0,
+                   1.0);
+}
+)";
+
+        // The two clear cases do not draw into the layered attachment at all - they SAMPLE it, so
+        // the queued clear is consumed by MaterializePendingClearForTexture rather than by a render
+        // pass's LOAD_OP_CLEAR. What the sample returns is irrelevant; being sampled is the point.
+        const char* const kCubeSampleFragmentSource = R"(#version 420 core
+uniform samplerCube u_source;
+out vec4 o_color;
+void main() { o_color = texture(u_source, vec3(1.0, 0.0, 0.0)); }
+)";
+
+        const char* const kOneDArraySampleFragmentSource = R"(#version 420 core
+uniform sampler1DArray u_source;
+out vec4 o_color;
+void main() { o_color = texture(u_source, vec2(0.5, 0.0)); }
+)";
+
+        // The non-layered case has no geometry stage at all - the slice comes from the
+        // attachment, not from gl_Layer - so it names its layer through a uniform.
+        const char* const kFlatFragmentSource = R"(#version 420 core
+uniform int u_layer;
+uniform int u_pass;
+out vec4 o_color;
+void main()
+{
+    o_color = vec4(float(10 + u_layer * 20) / 255.0,
+                   float(200 - u_layer * 10) / 255.0,
+                   float(3 + u_pass * 60) / 255.0,
+                   1.0);
+}
+)";
+
+        class LayeredAttachmentShapeScenario : public ScenarioTest {
+        protected:
+            void SetUp() override {
+                ScenarioTest::SetUp();
+                if (!Ready()) return;
+                glGenVertexArrays(1, &m_vao);
+                glBindVertexArray(m_vao);
+                DrainErrors();
+            }
+
+            void TearDown() override {
+                if (!Ready()) return;
+                glUseProgram(0);
+                glBindFramebuffer(GL_FRAMEBUFFER, 0);
+                for (const GLuint fbo : m_fbos) glDeleteFramebuffers(1, &fbo);
+                m_fbos.clear();
+                for (const GLuint texture : m_textures) glDeleteTextures(1, &texture);
+                m_textures.clear();
+                for (const GLuint program : m_programs) glDeleteProgram(program);
+                m_programs.clear();
+                glBindVertexArray(0);
+                if (m_vao != 0) glDeleteVertexArrays(1, &m_vao);
+                m_vao = 0;
+                glDisable(GL_DEPTH_TEST);
+                glDisable(GL_STENCIL_TEST);
+                DrainErrors();
+            }
+
+            static void DrainErrors() {
+                for (int i = 0; i < 16 && glGetError() != GL_NO_ERROR; ++i) {
+                }
+            }
+
+            // 0 on a DirectGLES driver without GL_EXT_geometry_shader and on a DirectVulkan
+            // device without the geometryShader feature. The same probe GeometryDrawModeScenario
+            // and IoBlockNameCollisionScenario use.
+            static bool BackendHostsGeometry() {
+                GLint maxGeometryOutputVertices = 0;
+                glGetIntegerv(GL_MAX_GEOMETRY_OUTPUT_VERTICES, &maxGeometryOutputVertices);
+                DrainErrors();
+                return maxGeometryOutputVertices >= kCubeLayerFaces * 3;
+            }
+
+            static std::string InfoLog(GLuint object, bool isShader) {
+                GLint length = 0;
+                if (isShader) {
+                    glGetShaderiv(object, GL_INFO_LOG_LENGTH, &length);
+                } else {
+                    glGetProgramiv(object, GL_INFO_LOG_LENGTH, &length);
+                }
+                std::vector<char> buffer(static_cast<std::size_t>(length) + 1, '\0');
+                if (isShader) {
+                    glGetShaderInfoLog(object, length + 1, nullptr, buffer.data());
+                } else {
+                    glGetProgramInfoLog(object, length + 1, nullptr, buffer.data());
+                }
+                return buffer.data();
+            }
+
+            // geometrySource may be null, which builds the no-geometry-stage program the
+            // non-layered case uses.
+            GLuint BuildProgram(const char* geometrySource, const char* fragmentSource) {
+                std::vector<GLuint> shaders;
+                const auto compile = [&](GLenum stage, const char* source) {
+                    const GLuint shader = glCreateShader(stage);
+                    glShaderSource(shader, 1, &source, nullptr);
+                    glCompileShader(shader);
+                    GLint compiled = 0;
+                    glGetShaderiv(shader, GL_COMPILE_STATUS, &compiled);
+                    shaders.push_back(shader);
+                    if (compiled == GL_FALSE) {
+                        ADD_FAILURE() << "stage 0x" << std::hex << stage << std::dec
+                                      << " did not compile: " << InfoLog(shader, true);
+                        return false;
+                    }
+                    return true;
+                };
+
+                bool ok = compile(GL_VERTEX_SHADER, kVertexSource);
+                if (ok && geometrySource != nullptr) ok = compile(GL_GEOMETRY_SHADER, geometrySource);
+                if (ok) ok = compile(GL_FRAGMENT_SHADER, fragmentSource);
+                if (!ok) {
+                    for (const GLuint shader : shaders) glDeleteShader(shader);
+                    return 0;
+                }
+
+                const GLuint program = glCreateProgram();
+                for (const GLuint shader : shaders) glAttachShader(program, shader);
+                glLinkProgram(program);
+                for (const GLuint shader : shaders) glDeleteShader(shader);
+                GLint linked = 0;
+                glGetProgramiv(program, GL_LINK_STATUS, &linked);
+                if (linked == GL_FALSE) {
+                    ADD_FAILURE() << "the program did not link: " << InfoLog(program, false);
+                    glDeleteProgram(program);
+                    return 0;
+                }
+                m_programs.push_back(program);
+                return program;
+            }
+
+            GLuint TrackTexture() {
+                GLuint texture = 0;
+                glGenTextures(1, &texture);
+                m_textures.push_back(texture);
+                return texture;
+            }
+
+            GLuint TrackFramebuffer() {
+                GLuint fbo = 0;
+                glGenFramebuffers(1, &fbo);
+                m_fbos.push_back(fbo);
+                return fbo;
+            }
+
+            // An RGBA8 3D texture, every texel poisoned.
+            GLuint MakePoisoned3DColor() {
+                const GLuint texture = TrackTexture();
+                glBindTexture(GL_TEXTURE_3D, texture);
+                glTexParameteri(GL_TEXTURE_3D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
+                glTexParameteri(GL_TEXTURE_3D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
+                glTexStorage3D(GL_TEXTURE_3D, 1, GL_RGBA8, kExtent, kExtent, k3DSlices);
+                const std::vector<GLubyte> seed(
+                    static_cast<std::size_t>(kExtent) * kExtent * k3DSlices * 4, kPoison);
+                glPixelStorei(GL_UNPACK_ALIGNMENT, 1);
+                glTexSubImage3D(GL_TEXTURE_3D, 0, 0, 0, 0, kExtent, kExtent, k3DSlices, GL_RGBA,
+                                GL_UNSIGNED_BYTE, seed.data());
+                glBindTexture(GL_TEXTURE_3D, 0);
+                return texture;
+            }
+
+            // An RGBA8 cube-map array of kCubeLayerFaces layer-faces, every texel poisoned.
+            GLuint MakePoisonedCubeArrayColor() {
+                const GLuint texture = TrackTexture();
+                glBindTexture(GL_TEXTURE_CUBE_MAP_ARRAY, texture);
+                glTexParameteri(GL_TEXTURE_CUBE_MAP_ARRAY, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
+                glTexParameteri(GL_TEXTURE_CUBE_MAP_ARRAY, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
+                glTexStorage3D(GL_TEXTURE_CUBE_MAP_ARRAY, 1, GL_RGBA8, kExtent, kExtent, kCubeLayerFaces);
+                const std::vector<GLubyte> seed(
+                    static_cast<std::size_t>(kExtent) * kExtent * kCubeLayerFaces * 4, kPoison);
+                glPixelStorei(GL_UNPACK_ALIGNMENT, 1);
+                glTexSubImage3D(GL_TEXTURE_CUBE_MAP_ARRAY, 0, 0, 0, 0, kExtent, kExtent, kCubeLayerFaces,
+                                GL_RGBA, GL_UNSIGNED_BYTE, seed.data());
+                glBindTexture(GL_TEXTURE_CUBE_MAP_ARRAY, 0);
+                return texture;
+            }
+
+            // A plain RGBA8 CUBE MAP (not an array), every face poisoned. This is the shape whose
+            // layered attachment records the +X face as its representative upload target, so its
+            // level size reads z = 1 - the reason a shared layer-count helper is needed at all.
+            GLuint MakePoisonedCubeMap() {
+                const GLuint texture = TrackTexture();
+                glBindTexture(GL_TEXTURE_CUBE_MAP, texture);
+                glTexParameteri(GL_TEXTURE_CUBE_MAP, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
+                glTexParameteri(GL_TEXTURE_CUBE_MAP, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
+                glTexStorage2D(GL_TEXTURE_CUBE_MAP, 1, GL_RGBA8, kExtent, kExtent);
+                const std::vector<GLubyte> seed(static_cast<std::size_t>(kExtent) * kExtent * 4, kPoison);
+                glPixelStorei(GL_UNPACK_ALIGNMENT, 1);
+                for (int face = 0; face < 6; ++face) {
+                    glTexSubImage2D(static_cast<GLenum>(GL_TEXTURE_CUBE_MAP_POSITIVE_X + face), 0, 0, 0, kExtent,
+                                    kExtent, GL_RGBA, GL_UNSIGNED_BYTE, seed.data());
+                }
+                glBindTexture(GL_TEXTURE_CUBE_MAP, 0);
+                return texture;
+            }
+
+            // An RGBA8 1D array, every layer poisoned. glTexImage2D's HEIGHT is the layer count -
+            // that is what GL_TEXTURE_1D_ARRAY means, and it is why reading the level size's z
+            // gives 1 however many layers there are.
+            GLuint MakePoisoned1DArray() {
+                const GLuint texture = TrackTexture();
+                glBindTexture(GL_TEXTURE_1D_ARRAY, texture);
+                glTexParameteri(GL_TEXTURE_1D_ARRAY, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
+                glTexParameteri(GL_TEXTURE_1D_ARRAY, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
+                const std::vector<GLubyte> seed(static_cast<std::size_t>(kExtent) * kOneDArrayLayers * 4, kPoison);
+                glPixelStorei(GL_UNPACK_ALIGNMENT, 1);
+                glTexImage2D(GL_TEXTURE_1D_ARRAY, 0, GL_RGBA8, kExtent, kOneDArrayLayers, 0, GL_RGBA,
+                             GL_UNSIGNED_BYTE, seed.data());
+                glBindTexture(GL_TEXTURE_1D_ARRAY, 0);
+                return texture;
+            }
+
+            // A scratch 2D colour target for the sampling draw. It exists only so the draw has
+            // somewhere to go that is NOT the layered attachment under test - a draw into that
+            // would open a render pass and consume the pending clear through LOAD_OP_CLEAR, which
+            // is the other consumer and the one that was already right.
+            GLuint MakeScratchColorFbo() {
+                const GLuint scratch = TrackTexture();
+                glBindTexture(GL_TEXTURE_2D, scratch);
+                glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
+                glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
+                glTexStorage2D(GL_TEXTURE_2D, 1, GL_RGBA8, kExtent, kExtent);
+                glBindTexture(GL_TEXTURE_2D, 0);
+                const GLuint fbo = TrackFramebuffer();
+                glBindFramebuffer(GL_FRAMEBUFFER, fbo);
+                glFramebufferTexture2D(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, scratch, 0);
+                glDrawBuffer(GL_COLOR_ATTACHMENT0);
+                return fbo;
+            }
+
+            // One draw that SAMPLES `texture`, into `intoFbo`. This is what drags the queued clear
+            // through MaterializePendingClearForTexture (VulkanRenderer's sampled-texture
+            // pre-pass), which is the consumer that used to write the clear key's layerCount
+            // straight into a VkImageSubresourceRange.
+            void DrawSampling(GLuint program, GLuint intoFbo, GLenum textureTarget, GLuint texture) {
+                glBindFramebuffer(GL_FRAMEBUFFER, intoFbo);
+                glViewport(0, 0, kExtent, kExtent);
+                glDisable(GL_SCISSOR_TEST);
+                glDisable(GL_DEPTH_TEST);
+                glDisable(GL_STENCIL_TEST);
+                glActiveTexture(GL_TEXTURE0);
+                glBindTexture(textureTarget, texture);
+                glUseProgram(program);
+                const GLint sourceLocation = glGetUniformLocation(program, "u_source");
+                ASSERT_GE(sourceLocation, 0) << "u_source was not reflected";
+                glUniform1i(sourceLocation, 0);
+                const GLint depthLocation = glGetUniformLocation(program, "u_depth");
+                ASSERT_GE(depthLocation, 0) << "u_depth was not reflected";
+                glUniform1f(depthLocation, 0.0f);
+                glDrawArrays(GL_TRIANGLES, 0, 3);
+                glBindTexture(textureTarget, 0);
+                glBindFramebuffer(GL_FRAMEBUFFER, 0);
+            }
+
+            // Every texel of `texels` is the clear colour. +/-1 per channel, which no rounding can
+            // exceed and which cannot be confused with the poison (0xAB) it replaced.
+            void ExpectAllCleared(const std::vector<Rgba8>& texels, int perTexelStride, const char* what) {
+                for (std::size_t i = 0; i < texels.size(); ++i) {
+                    const Rgba8& actual = texels[i];
+                    const bool ok = std::abs(static_cast<int>(actual.r) - kClearColor.r) <= 1 &&
+                                    std::abs(static_cast<int>(actual.g) - kClearColor.g) <= 1 &&
+                                    std::abs(static_cast<int>(actual.b) - kClearColor.b) <= 1;
+                    if (ok) continue;
+                    ADD_FAILURE() << what << ": unit " << (static_cast<int>(i) / perTexelStride) << " texel "
+                                  << (static_cast<int>(i) % perTexelStride) << " is " << Describe(actual)
+                                  << ", expected " << Describe(kClearColor)
+                                  << (actual.r == kPoison && actual.g == kPoison
+                                          ? " - the poison, so the clear never reached this one"
+                                          : "");
+                    // One message per unit is enough to say what happened.
+                    i = (static_cast<std::size_t>(i) / perTexelStride + 1) * perTexelStride - 1;
+                }
+            }
+
+            // A depth (or packed depth-stencil) cube-map array of the same shape. No upload: a
+            // depth array is filled by clearing through an attachment, which is the state the
+            // gating cases start from anyway.
+            GLuint MakeCubeArrayDepth(GLenum internalFormat) {
+                const GLuint texture = TrackTexture();
+                glBindTexture(GL_TEXTURE_CUBE_MAP_ARRAY, texture);
+                glTexParameteri(GL_TEXTURE_CUBE_MAP_ARRAY, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
+                glTexParameteri(GL_TEXTURE_CUBE_MAP_ARRAY, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
+                glTexStorage3D(GL_TEXTURE_CUBE_MAP_ARRAY, 1, internalFormat, kExtent, kExtent, kCubeLayerFaces);
+                glBindTexture(GL_TEXTURE_CUBE_MAP_ARRAY, 0);
+                return texture;
+            }
+
+            // glGetTexImage rather than a per-layer glReadPixels: a cube-map array has no
+            // per-layer attachment on every backend, and glGetTexImage is the readback both of
+            // them answer for whole-level layered targets (LayeredTextureReadbackScenario pins
+            // that contract). It is a real GPU readback on DirectVulkan - the texture manager
+            // copies the image into a staging buffer - so a stale CPU shadow cannot pass it.
+            std::vector<Rgba8> ReadLevel(GLenum target, GLuint texture, int layers) {
+                std::vector<Rgba8> texels(static_cast<std::size_t>(kExtent) * kExtent * layers, Rgba8{});
+                glBindTexture(target, texture);
+                glPixelStorei(GL_PACK_ALIGNMENT, 1);
+                glGetTexImage(target, 0, GL_RGBA, GL_UNSIGNED_BYTE, texels.data());
+                glBindTexture(target, 0);
+                return texels;
+            }
+
+            // Every texel of every layer must be that layer's expected colour. Reported per layer
+            // so a failure names which one, and the poison is called out by name.
+            void ExpectEveryLayer(const std::vector<Rgba8>& texels, int layers, int pass, const char* what) {
+                for (int layer = 0; layer < layers; ++layer) {
+                    const Rgba8 expected = ExpectedColor(layer, pass);
+                    for (int y = 0; y < kExtent; ++y) {
+                        for (int x = 0; x < kExtent; ++x) {
+                            const std::size_t index =
+                                (static_cast<std::size_t>(layer) * kExtent + y) * kExtent + x;
+                            const Rgba8 actual = texels[index];
+                            if (actual == expected) continue;
+                            ADD_FAILURE()
+                                << what << ": layer " << layer << " texel (" << x << ", " << y << ") is "
+                                << Describe(actual) << ", expected " << Describe(expected)
+                                << (actual.r == kPoison && actual.g == kPoison
+                                        ? " - the poison, so nothing was ever rendered into this layer"
+                                        : "");
+                            // One message per layer is enough to say what happened.
+                            y = kExtent;
+                            break;
+                        }
+                    }
+                }
+            }
+
+            ::testing::AssertionResult FramebufferIsComplete() {
+                const GLenum status = glCheckFramebufferStatus(GL_FRAMEBUFFER);
+                if (status == GL_FRAMEBUFFER_COMPLETE) return ::testing::AssertionSuccess();
+                return ::testing::AssertionFailure() << "framebuffer status 0x" << std::hex << status;
+            }
+
+            // One layered pass over the whole attachment.
+            void DrawLayered(GLuint program, int pass, float depth) {
+                glUseProgram(program);
+                const GLint passLocation = glGetUniformLocation(program, "u_pass");
+                ASSERT_GE(passLocation, 0) << "u_pass was not reflected";
+                glUniform1i(passLocation, pass);
+                const GLint depthLocation = glGetUniformLocation(program, "u_depth");
+                ASSERT_GE(depthLocation, 0) << "u_depth was not reflected";
+                glUniform1f(depthLocation, depth);
+                glDrawArrays(GL_TRIANGLES, 0, 3);
+            }
+
+            GLuint m_vao = 0;
+            std::vector<GLuint> m_textures;
+            std::vector<GLuint> m_fbos;
+            std::vector<GLuint> m_programs;
+        };
+
+        // (1) A LAYERED GL_TEXTURE_3D colour attachment. Pre-fix this is the null VkImageView:
+        // the attachment asked for a 3D view, whose [0, 4) layer span was measured against the
+        // image's arrayLayers == 1 and refused, and VK_NULL_HANDLE went to vkCreateFramebuffer.
+        TEST_F(LayeredAttachmentShapeScenario, LayeredThreeDColorAttachmentReachesEverySlice) {
+            if (!Ready()) return;
+            if (!BackendHostsGeometry()) GTEST_SKIP() << "no geometry stage: nothing can write gl_Layer";
+
+            const std::string geometrySource = MakeGeometrySource(k3DSlices);
+            const GLuint program = BuildProgram(geometrySource.c_str(), kLayeredFragmentSource);
+            if (program == 0) return;
+
+            const GLuint color = MakePoisoned3DColor();
+            ASSERT_EQ(FirstGLError(), 0u) << "creating the RGBA8 3D texture failed";
+
+            const GLuint fbo = TrackFramebuffer();
+            glBindFramebuffer(GL_FRAMEBUFFER, fbo);
+            glFramebufferTexture(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, color, 0);
+            glDrawBuffer(GL_COLOR_ATTACHMENT0);
+            ASSERT_EQ(FirstGLError(), 0u) << "attaching the 3D texture layered failed";
+            ASSERT_TRUE(FramebufferIsComplete());
+
+            glViewport(0, 0, kExtent, kExtent);
+            glDisable(GL_SCISSOR_TEST);
+            glDisable(GL_DEPTH_TEST);
+            glDisable(GL_STENCIL_TEST);
+            DrawLayered(program, /*pass=*/0, /*depth=*/0.0f);
+            EXPECT_EQ(FirstGLError(), 0u) << "the layered draw errored";
+
+            glBindFramebuffer(GL_FRAMEBUFFER, 0);
+            const std::vector<Rgba8> texels = ReadLevel(GL_TEXTURE_3D, color, k3DSlices);
+            EXPECT_EQ(FirstGLError(), 0u) << "reading the 3D level back errored";
+            ExpectEveryLayer(texels, k3DSlices, /*pass=*/0, "layered GL_TEXTURE_3D colour attachment");
+
+            Gl().EndFrame();
+        }
+
+        // (2) The same texture attached ONE SLICE at a time, which is the other half of the same
+        // view-type decision. Pre-fix a non-layered 3D attachment also asked for a 3D view, so
+        // the per-slice branch never ran and slice 2 resolved to VK_NULL_HANDLE. Needs no
+        // geometry stage - the slice comes from the attachment.
+        TEST_F(LayeredAttachmentShapeScenario, NonLayeredThreeDSliceAttachmentWritesOnlyThatSlice) {
+            if (!Ready()) return;
+
+            const GLuint program = BuildProgram(nullptr, kFlatFragmentSource);
+            if (program == 0) return;
+
+            const GLuint color = MakePoisoned3DColor();
+            ASSERT_EQ(FirstGLError(), 0u) << "creating the RGBA8 3D texture failed";
+
+            const GLuint fbo = TrackFramebuffer();
+            glBindFramebuffer(GL_FRAMEBUFFER, fbo);
+            glFramebufferTextureLayer(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, color, 0, kSubjectSlice);
+            glDrawBuffer(GL_COLOR_ATTACHMENT0);
+            ASSERT_EQ(FirstGLError(), 0u) << "attaching slice " << kSubjectSlice << " failed";
+            ASSERT_TRUE(FramebufferIsComplete());
+
+            glViewport(0, 0, kExtent, kExtent);
+            glDisable(GL_SCISSOR_TEST);
+            glDisable(GL_DEPTH_TEST);
+            glDisable(GL_STENCIL_TEST);
+            glUseProgram(program);
+            const GLint layerLocation = glGetUniformLocation(program, "u_layer");
+            const GLint passLocation = glGetUniformLocation(program, "u_pass");
+            const GLint depthLocation = glGetUniformLocation(program, "u_depth");
+            ASSERT_GE(layerLocation, 0);
+            ASSERT_GE(passLocation, 0);
+            ASSERT_GE(depthLocation, 0);
+            glUniform1i(layerLocation, kSubjectSlice);
+            glUniform1i(passLocation, 0);
+            glUniform1f(depthLocation, 0.0f);
+            glDrawArrays(GL_TRIANGLES, 0, 3);
+            EXPECT_EQ(FirstGLError(), 0u) << "the per-slice draw errored";
+
+            glBindFramebuffer(GL_FRAMEBUFFER, 0);
+            const std::vector<Rgba8> texels = ReadLevel(GL_TEXTURE_3D, color, k3DSlices);
+            EXPECT_EQ(FirstGLError(), 0u) << "reading the 3D level back errored";
+
+            const Rgba8 expected = ExpectedColor(kSubjectSlice, 0);
+            const Rgba8 poison{kPoison, kPoison, kPoison, kPoison};
+            for (int slice = 0; slice < k3DSlices; ++slice) {
+                const Rgba8& target = (slice == kSubjectSlice) ? expected : poison;
+                for (int y = 0; y < kExtent; ++y) {
+                    for (int x = 0; x < kExtent; ++x) {
+                        const std::size_t index =
+                            (static_cast<std::size_t>(slice) * kExtent + y) * kExtent + x;
+                        const Rgba8 actual = texels[index];
+                        if (actual == target) continue;
+                        ADD_FAILURE() << "slice " << slice << " texel (" << x << ", " << y << ") is "
+                                      << Describe(actual) << ", expected " << Describe(target)
+                                      << (slice == kSubjectSlice
+                                              ? " - the attached slice was not the one written"
+                                              : " - a slice the attachment did not name was written");
+                        y = kExtent;
+                        break;
+                    }
+                }
+            }
+
+            Gl().EndFrame();
+        }
+
+        // (3) A LAYERED GL_TEXTURE_CUBE_MAP_ARRAY colour attachment. Pre-fix this is the other
+        // exit: a valid CUBE_ARRAY view of a type no framebuffer may take, handed straight to
+        // vkCreateFramebuffer.
+        TEST_F(LayeredAttachmentShapeScenario, LayeredCubeMapArrayColorAttachmentReachesEveryLayerFace) {
+            if (!Ready()) return;
+            if (!BackendHostsGeometry()) GTEST_SKIP() << "no geometry stage: nothing can write gl_Layer";
+
+            const GLuint color = MakePoisonedCubeArrayColor();
+            if (const GLenum error = FirstGLError()) {
+                GTEST_SKIP() << "no usable GL_TEXTURE_CUBE_MAP_ARRAY on this backend: " << GLErrorName(error);
+            }
+
+            const std::string geometrySource = MakeGeometrySource(kCubeLayerFaces);
+            const GLuint program = BuildProgram(geometrySource.c_str(), kLayeredFragmentSource);
+            if (program == 0) return;
+
+            const GLuint fbo = TrackFramebuffer();
+            glBindFramebuffer(GL_FRAMEBUFFER, fbo);
+            glFramebufferTexture(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, color, 0);
+            glDrawBuffer(GL_COLOR_ATTACHMENT0);
+            ASSERT_EQ(FirstGLError(), 0u) << "attaching the cube-map array layered failed";
+            ASSERT_TRUE(FramebufferIsComplete());
+
+            glViewport(0, 0, kExtent, kExtent);
+            glDisable(GL_SCISSOR_TEST);
+            glDisable(GL_DEPTH_TEST);
+            glDisable(GL_STENCIL_TEST);
+            DrawLayered(program, /*pass=*/0, /*depth=*/0.0f);
+            EXPECT_EQ(FirstGLError(), 0u) << "the layered draw errored";
+
+            glBindFramebuffer(GL_FRAMEBUFFER, 0);
+            const std::vector<Rgba8> texels = ReadLevel(GL_TEXTURE_CUBE_MAP_ARRAY, color, kCubeLayerFaces);
+            EXPECT_EQ(FirstGLError(), 0u) << "reading the cube-map-array level back errored";
+            ExpectEveryLayer(texels, kCubeLayerFaces, /*pass=*/0,
+                             "layered GL_TEXTURE_CUBE_MAP_ARRAY colour attachment");
+
+            Gl().EndFrame();
+        }
+
+        // (4) A layered cube-map-array DEPTH attachment, proved to have covered every layer-face:
+        //
+        //   pass 0 paints at z = 0 against a depth buffer cleared to 1;
+        //   pass 1 paints at z = +0.5, which GL_LESS must reject.
+        //
+        // A layer that reads back pass 1's colour is a layer the depth attachment never covered -
+        // which is exactly what attaching one layer-face of it, or none, looks like. Runs on both
+        // backends: this is the cross-backend control for the packed case below.
+        TEST_F(LayeredAttachmentShapeScenario, LayeredCubeMapArrayDepthAttachmentGatesEveryLayerFace) {
+            if (!Ready()) return;
+            if (!BackendHostsGeometry()) GTEST_SKIP() << "no geometry stage: nothing can write gl_Layer";
+
+            const GLuint color = MakePoisonedCubeArrayColor();
+            if (const GLenum error = FirstGLError()) {
+                GTEST_SKIP() << "no usable GL_TEXTURE_CUBE_MAP_ARRAY on this backend: " << GLErrorName(error);
+            }
+
+            const GLuint depth = MakeCubeArrayDepth(GL_DEPTH_COMPONENT24);
+            if (const GLenum error = FirstGLError()) {
+                GTEST_SKIP() << "no depth GL_TEXTURE_CUBE_MAP_ARRAY on this backend: " << GLErrorName(error);
+            }
+
+            const std::string geometrySource = MakeGeometrySource(kCubeLayerFaces);
+            const GLuint program = BuildProgram(geometrySource.c_str(), kLayeredFragmentSource);
+            if (program == 0) return;
+
+            const GLuint fbo = TrackFramebuffer();
+            glBindFramebuffer(GL_FRAMEBUFFER, fbo);
+            glFramebufferTexture(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, color, 0);
+            glFramebufferTexture(GL_FRAMEBUFFER, GL_DEPTH_ATTACHMENT, depth, 0);
+            glDrawBuffer(GL_COLOR_ATTACHMENT0);
+            ASSERT_EQ(FirstGLError(), 0u) << "attaching the layered colour + depth pair failed";
+            ASSERT_TRUE(FramebufferIsComplete());
+
+            glViewport(0, 0, kExtent, kExtent);
+            glDisable(GL_SCISSOR_TEST);
+            glDisable(GL_STENCIL_TEST);
+            glDepthMask(GL_TRUE);
+            glClearDepth(1.0);
+            glClear(GL_DEPTH_BUFFER_BIT);
+            ASSERT_EQ(FirstGLError(), 0u) << "clearing the layered depth attachment errored";
+
+            glEnable(GL_DEPTH_TEST);
+            glDepthFunc(GL_LESS);
+            DrawLayered(program, /*pass=*/0, /*depth=*/0.0f);
+            DrawLayered(program, /*pass=*/1, /*depth=*/0.5f); // farther: GL_LESS must reject it
+            EXPECT_EQ(FirstGLError(), 0u) << "the two layered draws errored";
+
+            glDisable(GL_DEPTH_TEST);
+            glBindFramebuffer(GL_FRAMEBUFFER, 0);
+            const std::vector<Rgba8> texels = ReadLevel(GL_TEXTURE_CUBE_MAP_ARRAY, color, kCubeLayerFaces);
+            EXPECT_EQ(FirstGLError(), 0u) << "reading the cube-map-array level back errored";
+            ExpectEveryLayer(texels, kCubeLayerFaces, /*pass=*/0,
+                             "layered cube-map-array depth attachment (pass 1's colour on a layer means the "
+                             "depth test did not cover it)");
+
+            Gl().EndFrame();
+        }
+
+        // (5) The PACKED depth-stencil shape the conformance suite crashes on:
+        // texture_cube_map_array.stencil_attachments_*_layered attaches a cube-map array as COLOR0
+        // AND the same-shaped GL_DEPTH24_STENCIL8 array as GL_DEPTH_STENCIL_ATTACHMENT, both
+        // layered. Both aspects are proved to have covered every layer-face:
+        //
+        //   pass 0 paints at z = 0 with the stencil op writing 1;
+        //   pass 1 paints at z = +0.5, which the depth test must reject;
+        //   pass 2 paints with the depth test off but a stencil func of EQUAL 0, which the
+        //          stencil written by pass 0 must reject.
+        //
+        // The probe in front of the gating is where this scenario earned its keep. The attachment
+        // point ITSELF was broken: glFramebufferTexture(GL_DEPTH_STENCIL_ATTACHMENT) is a
+        // shorthand that the front end splits into a depth and a stencil attachment, and the split
+        // dropped the call's `layered` flag (GL_Framebuffer.cpp,
+        // AttachFramebufferTextureWithUploadTarget). A layered colour attachment therefore sat
+        // beside a NON-layered depth/stencil one, and both backends lost the draw entirely - with
+        // no GL error and glCheckFramebufferStatus answering COMPLETE. DirectVulkan built the
+        // depth/stencil view with layerCount 1 under a framebuffer declaring 12 layers
+        // (VUID-VkFramebufferCreateInfo-flags-04535, which the validation layers report on this
+        // exact case); DirectGLES attached one layer of it beside a layered colour target, which
+        // the driver answers with GL_FRAMEBUFFER_INCOMPLETE_LAYER_TARGETS. Case (4) above is what
+        // isolates it to the attachment point: the same cube-map array on GL_DEPTH_ATTACHMENT
+        // rendered and gated correctly throughout.
+        //
+        // So the probe stays, as an assertion rather than as scaffolding: it turns that regression
+        // back into ONE message about the shape instead of twelve about individual layers.
+        TEST_F(LayeredAttachmentShapeScenario, LayeredCubeMapArrayDepthStencilAttachmentGatesEveryLayerFace) {
+            if (!Ready()) return;
+            if (!BackendHostsGeometry()) GTEST_SKIP() << "no geometry stage: nothing can write gl_Layer";
+
+            const GLuint color = MakePoisonedCubeArrayColor();
+            if (const GLenum error = FirstGLError()) {
+                GTEST_SKIP() << "no usable GL_TEXTURE_CUBE_MAP_ARRAY on this backend: " << GLErrorName(error);
+            }
+
+            const GLuint depthStencil = MakeCubeArrayDepth(GL_DEPTH24_STENCIL8);
+            if (const GLenum error = FirstGLError()) {
+                GTEST_SKIP() << "no depth-stencil GL_TEXTURE_CUBE_MAP_ARRAY on this backend: "
+                             << GLErrorName(error);
+            }
+
+            const std::string geometrySource = MakeGeometrySource(kCubeLayerFaces);
+            const GLuint program = BuildProgram(geometrySource.c_str(), kLayeredFragmentSource);
+            if (program == 0) return;
+
+            glViewport(0, 0, kExtent, kExtent);
+            glDisable(GL_SCISSOR_TEST);
+
+            // The probe: its own colour attachment (so the subject texture keeps its poison), the
+            // same depth-stencil attachment, and both tests off - so every layer-face must come
+            // back painted, whatever the gating below then decides.
+            {
+                const GLuint probeColor = MakePoisonedCubeArrayColor();
+                const GLuint probeFbo = TrackFramebuffer();
+                glBindFramebuffer(GL_FRAMEBUFFER, probeFbo);
+                glFramebufferTexture(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, probeColor, 0);
+                glFramebufferTexture(GL_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, depthStencil, 0);
+                glDrawBuffer(GL_COLOR_ATTACHMENT0);
+                ASSERT_EQ(FirstGLError(), 0u) << "attaching the layered colour + depth-stencil pair failed";
+                ASSERT_TRUE(FramebufferIsComplete());
+                glDisable(GL_DEPTH_TEST);
+                glDisable(GL_STENCIL_TEST);
+                DrawLayered(program, /*pass=*/3, /*depth=*/0.0f);
+                glBindFramebuffer(GL_FRAMEBUFFER, 0);
+                const std::vector<Rgba8> probeTexels =
+                    ReadLevel(GL_TEXTURE_CUBE_MAP_ARRAY, probeColor, kCubeLayerFaces);
+                EXPECT_EQ(FirstGLError(), 0u) << "the probe draw or readback errored";
+                ExpectEveryLayer(probeTexels, kCubeLayerFaces, /*pass=*/3,
+                                 "a layered draw with the depth and stencil tests DISABLED, into a colour + "
+                                 "GL_DEPTH_STENCIL_ATTACHMENT cube-map-array pair (all poison means the "
+                                 "attachment pair lost the draw outright, which is what a non-layered "
+                                 "depth/stencil attachment beside a layered colour one looks like)");
+                // The gating assertions below can only add noise once the shape itself is broken.
+                if (::testing::Test::HasNonfatalFailure()) return;
+            }
+
+            const GLuint fbo = TrackFramebuffer();
+            glBindFramebuffer(GL_FRAMEBUFFER, fbo);
+            glFramebufferTexture(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, color, 0);
+            glFramebufferTexture(GL_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, depthStencil, 0);
+            glDrawBuffer(GL_COLOR_ATTACHMENT0);
+            ASSERT_EQ(FirstGLError(), 0u) << "attaching the layered colour + depth-stencil pair failed";
+            ASSERT_TRUE(FramebufferIsComplete());
+
+            glDepthMask(GL_TRUE);
+            glStencilMask(0xFFu);
+            glClearDepth(1.0);
+            glClearStencil(0);
+            glClear(GL_DEPTH_BUFFER_BIT | GL_STENCIL_BUFFER_BIT);
+            ASSERT_EQ(FirstGLError(), 0u) << "clearing the layered depth-stencil attachment errored";
+
+            glEnable(GL_DEPTH_TEST);
+            glDepthFunc(GL_LESS);
+            glEnable(GL_STENCIL_TEST);
+            glStencilFunc(GL_ALWAYS, 1, 0xFFu);
+            glStencilOp(GL_KEEP, GL_KEEP, GL_REPLACE);
+            DrawLayered(program, /*pass=*/0, /*depth=*/0.0f);
+
+            // Farther than pass 0, so GL_LESS must reject it on every layer.
+            glStencilFunc(GL_ALWAYS, 1, 0xFFu);
+            glStencilOp(GL_KEEP, GL_KEEP, GL_KEEP);
+            DrawLayered(program, /*pass=*/1, /*depth=*/0.5f);
+
+            // Depth out of the way; only the stencil pass 0 wrote can reject this one.
+            glDepthFunc(GL_ALWAYS);
+            glStencilFunc(GL_EQUAL, 0, 0xFFu);
+            DrawLayered(program, /*pass=*/2, /*depth=*/-0.5f);
+            EXPECT_EQ(FirstGLError(), 0u) << "the three layered draws errored";
+
+            glDisable(GL_DEPTH_TEST);
+            glDisable(GL_STENCIL_TEST);
+            glBindFramebuffer(GL_FRAMEBUFFER, 0);
+            const std::vector<Rgba8> texels = ReadLevel(GL_TEXTURE_CUBE_MAP_ARRAY, color, kCubeLayerFaces);
+            EXPECT_EQ(FirstGLError(), 0u) << "reading the cube-map-array level back errored";
+            ExpectEveryLayer(texels, kCubeLayerFaces, /*pass=*/0,
+                             "layered cube-map-array depth-stencil attachment (a later pass's colour means "
+                             "the depth or stencil test did not cover that layer)");
+
+            Gl().EndFrame();
+        }
+
+        // (6) and (7) leave the render pass alone entirely and pin the OTHER consumer of a layered
+        // attachment's layer count.
+        //
+        // A glClear on a texture-backed FBO with the scissor test off is not executed on the spot:
+        // it is queued (VkClearManager), and then exactly one of two things consumes it - the next
+        // render pass's LOAD_OP_CLEAR over the attachment view, or MaterializePendingClearForTexture
+        // if the texture is used outside a pass first (sampled, blitted, copied, read back). The
+        // second path writes the queued key's layerCount straight into a VkImageSubresourceRange
+        // and then POPS the entry, so whatever it misses is lost for good - the render pass never
+        // gets a second chance at it.
+        //
+        // Both consumers must therefore agree about how many layers a layered attachment spans, and
+        // they are now literally the same function (ResolveAttachmentLayerCount, VkTextureManager.h).
+        // These two cases are the shapes where a raw `size.z()` and the real answer differ, and
+        // neither is reachable through the cases above: a cube MAP records the +X face as its
+        // representative upload target (z = 1, six real faces) and a 1D ARRAY keeps its layer count
+        // in the state-side height (z = 1, N real layers). The cube-map-ARRAY and 3D shapes the
+        // earlier cases use both carry their count in z, so they agree either way and cannot see it.
+        //
+        // The draw goes into a scratch 2D target, never into the layered attachment, so the
+        // materialise path is the only consumer that can fire.
+        TEST_F(LayeredAttachmentShapeScenario, LayeredCubeMapClearMaterialisedBySamplingReachesEveryFace) {
+            if (!Ready()) return;
+
+            const GLuint program = BuildProgram(nullptr, kCubeSampleFragmentSource);
+            if (program == 0) return;
+
+            const GLuint cube = MakePoisonedCubeMap();
+            ASSERT_EQ(FirstGLError(), 0u) << "creating the RGBA8 cube map failed";
+
+            const GLuint layeredFbo = TrackFramebuffer();
+            glBindFramebuffer(GL_FRAMEBUFFER, layeredFbo);
+            glFramebufferTexture(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, cube, 0);
+            glDrawBuffer(GL_COLOR_ATTACHMENT0);
+            ASSERT_EQ(FirstGLError(), 0u) << "attaching the cube map layered failed";
+            ASSERT_TRUE(FramebufferIsComplete());
+
+            glViewport(0, 0, kExtent, kExtent);
+            glDisable(GL_SCISSOR_TEST);
+            glClearColor(kClearColor.r / 255.0f, kClearColor.g / 255.0f, kClearColor.b / 255.0f, 1.0f);
+            glClear(GL_COLOR_BUFFER_BIT);
+            ASSERT_EQ(FirstGLError(), 0u) << "clearing the layered cube-map attachment errored";
+
+            // Consume the queued clear through the sampled-texture path, with no draw into the
+            // layered FBO in between.
+            const GLuint scratchFbo = MakeScratchColorFbo();
+            ASSERT_TRUE(FramebufferIsComplete()) << "the scratch 2D target is not complete";
+            DrawSampling(program, scratchFbo, GL_TEXTURE_CUBE_MAP, cube);
+            EXPECT_EQ(FirstGLError(), 0u) << "the sampling draw errored";
+
+            // Every face, read back through an FBO that names THAT face.
+            //
+            // Not glGetTexImage(GL_TEXTURE_CUBE_MAP_POSITIVE_X + face): measured against a tree
+            // where only +X had been cleared, that spelling returned the cleared colour for all
+            // six faces, so it cannot see per-face state on DirectVulkan and the case built on it
+            // was unfalsifiable. glFramebufferTexture2D + glReadPixels names one face and nothing
+            // else, and the pending clear is long gone by now (materialised and popped above), so
+            // this readback cannot alter what it is measuring.
+            static const char* const kFaceNames[6] = {"+X", "-X", "+Y", "-Y", "+Z", "-Z"};
+            for (int face = 0; face < 6; ++face) {
+                const GLuint faceFbo = TrackFramebuffer();
+                glBindFramebuffer(GL_FRAMEBUFFER, faceFbo);
+                glFramebufferTexture2D(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0,
+                                       static_cast<GLenum>(GL_TEXTURE_CUBE_MAP_POSITIVE_X + face), cube, 0);
+                glReadBuffer(GL_COLOR_ATTACHMENT0);
+                ASSERT_TRUE(FramebufferIsComplete()) << "cube face " << kFaceNames[face] << " is not attachable";
+                std::vector<Rgba8> texels(static_cast<std::size_t>(kExtent) * kExtent, Rgba8{});
+                glPixelStorei(GL_PACK_ALIGNMENT, 1);
+                glReadPixels(0, 0, kExtent, kExtent, GL_RGBA, GL_UNSIGNED_BYTE, texels.data());
+                glBindFramebuffer(GL_FRAMEBUFFER, 0);
+                EXPECT_EQ(FirstGLError(), 0u) << "reading cube face " << kFaceNames[face] << " back errored";
+                ExpectAllCleared(texels, kExtent * kExtent,
+                                 (std::string("layered GL_TEXTURE_CUBE_MAP glClear materialised by sampling, "
+                                              "face ") +
+                                  kFaceNames[face])
+                                     .c_str());
+            }
+
+            Gl().EndFrame();
+        }
+
+        // The 1D-array half of the same divergence. Pre-existing rather than introduced by this
+        // branch (the clear copy never had ToVulkanLevelExtent), and fixed by the same hoist.
+        TEST_F(LayeredAttachmentShapeScenario, LayeredOneDArrayClearMaterialisedBySamplingReachesEveryLayer) {
+            if (!Ready()) return;
+
+            const GLuint program = BuildProgram(nullptr, kOneDArraySampleFragmentSource);
+            if (program == 0) return;
+
+            const GLuint array = MakePoisoned1DArray();
+            if (const GLenum error = FirstGLError()) {
+                GTEST_SKIP() << "no usable GL_TEXTURE_1D_ARRAY on this backend: " << GLErrorName(error);
+            }
+
+            const GLuint layeredFbo = TrackFramebuffer();
+            glBindFramebuffer(GL_FRAMEBUFFER, layeredFbo);
+            glFramebufferTexture(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, array, 0);
+            glDrawBuffer(GL_COLOR_ATTACHMENT0);
+            ASSERT_EQ(FirstGLError(), 0u) << "attaching the 1D array layered failed";
+            ASSERT_TRUE(FramebufferIsComplete());
+
+            // The viewport is the LEVEL's shape: a 1D array level is `kExtent` wide and one row
+            // tall, whatever its layer count.
+            glViewport(0, 0, kExtent, 1);
+            glDisable(GL_SCISSOR_TEST);
+            glClearColor(kClearColor.r / 255.0f, kClearColor.g / 255.0f, kClearColor.b / 255.0f, 1.0f);
+            glClear(GL_COLOR_BUFFER_BIT);
+            ASSERT_EQ(FirstGLError(), 0u) << "clearing the layered 1D-array attachment errored";
+
+            const GLuint scratchFbo = MakeScratchColorFbo();
+            ASSERT_TRUE(FramebufferIsComplete()) << "the scratch 2D target is not complete";
+            DrawSampling(program, scratchFbo, GL_TEXTURE_1D_ARRAY, array);
+            EXPECT_EQ(FirstGLError(), 0u) << "the sampling draw errored";
+
+            // GL hands a 1D array back as a two-dimensional image whose ROWS are the layers.
+            std::vector<Rgba8> texels(static_cast<std::size_t>(kExtent) * kOneDArrayLayers, Rgba8{});
+            glBindTexture(GL_TEXTURE_1D_ARRAY, array);
+            glPixelStorei(GL_PACK_ALIGNMENT, 1);
+            glGetTexImage(GL_TEXTURE_1D_ARRAY, 0, GL_RGBA, GL_UNSIGNED_BYTE, texels.data());
+            glBindTexture(GL_TEXTURE_1D_ARRAY, 0);
+            EXPECT_EQ(FirstGLError(), 0u) << "reading the 1D-array level back errored";
+            ExpectAllCleared(texels, kExtent,
+                             "layered GL_TEXTURE_1D_ARRAY glClear materialised by sampling (unit = layer)");
+
+            Gl().EndFrame();
+        }
+
+    } // namespace
+} // namespace MGITest

MobileGL/MG_IntegrationTest/Scenarios/MultiDrawScenario.cpp

@@ -518,5 +518,62 @@ void main() {
             ExpectSameImage(batched, unrolled, "a batch with zero-count sub-draws");
         }
 
+        // The base-vertex family's argument checks (GL 4.6 core 10.3.9). These are what
+        // KHR-GL4x.draw_elements_base_vertex_tests.invalid_* assert, and the reason the group sat
+        // NotSupported for so long hid the fact that the entry points forwarded any argument
+        // straight to the backend: a negative count reached the emulation as a huge unsigned
+        // size. Each case drains the error queue first so the assertion names the call it made.
+        TEST_F(MultiDrawScenario, BaseVertexDrawsRejectMalformedArguments) {
+            if (!Ready()) return;
+            constexpr int kPad = 0;
+            BuildScene(kPad, kQuadIndices, sizeof(kQuadIndices));
+            // A bound program and VAO are prerequisites, not decoration: the entry points check
+            // "is there something to execute" (GL_INVALID_OPERATION) before they look at any
+            // argument, so without these every case below would pass for the wrong reason.
+            glUseProgram(m_program);
+            glBindVertexArray(m_vao);
+            ASSERT_EQ(FirstGLError(), GLenum(GL_NO_ERROR)) << "scene setup left a GL error behind";
+
+            const auto expectError = [&](const char* what, GLenum expected) {
+                EXPECT_EQ(FirstGLError(), expected) << what;
+                // FirstGLError stops at the first one; make sure nothing else is queued so the
+                // next case starts clean.
+                while (glGetError() != GL_NO_ERROR) {
+                }
+            };
+
+            glDrawElementsBaseVertex(GL_TRIANGLES, -1, GL_UNSIGNED_INT, nullptr, 0);
+            expectError("glDrawElementsBaseVertex with a negative count", GL_INVALID_VALUE);
+
+            glDrawElementsBaseVertex(GL_TRIANGLES, 3, GL_NONE, nullptr, 0);
+            expectError("glDrawElementsBaseVertex with a non-index type", GL_INVALID_ENUM);
+
+            glDrawRangeElementsBaseVertex(GL_TRIANGLES, 3, 0, 3, GL_UNSIGNED_INT, nullptr, 0);
+            expectError("glDrawRangeElementsBaseVertex with end < start", GL_INVALID_VALUE);
+
+            // start = -1 arrives as 0xFFFFFFFF, so this is the same end < start rule seen from
+            // the other side - and it is the shape the CTS's invalid_count case actually uses.
+            glDrawRangeElementsBaseVertex(GL_TRIANGLES, static_cast<GLuint>(-1), 2, 1, GL_UNSIGNED_INT, nullptr, 0);
+            expectError("glDrawRangeElementsBaseVertex with a wrapped start", GL_INVALID_VALUE);
+
+            glDrawElementsInstancedBaseVertex(GL_TRIANGLES, 3, GL_UNSIGNED_INT, nullptr, -1, 0);
+            expectError("glDrawElementsInstancedBaseVertex with a negative instancecount", GL_INVALID_VALUE);
+
+            const GLsizei negativeCount = -1;
+            const void* offsets[1] = {reinterpret_cast<const void*>(0)};
+            const GLint baseVertices[1] = {0};
+            glMultiDrawElementsBaseVertex(GL_TRIANGLES, &negativeCount, GL_UNSIGNED_INT, offsets, 1, baseVertices);
+            expectError("glMultiDrawElementsBaseVertex with a negative element of count", GL_INVALID_VALUE);
+
+            const GLsizei validCount = 6;
+            glMultiDrawElementsBaseVertex(GL_TRIANGLES, &validCount, GL_UNSIGNED_INT, offsets, -1, baseVertices);
+            expectError("glMultiDrawElementsBaseVertex with a negative drawcount", GL_INVALID_VALUE);
+
+            // The well-formed call still has to go through, or the checks above would be
+            // indistinguishable from a blanket rejection.
+            glMultiDrawElementsBaseVertex(GL_TRIANGLES, &validCount, GL_UNSIGNED_INT, offsets, 1, baseVertices);
+            expectError("a well-formed glMultiDrawElementsBaseVertex", GL_NO_ERROR);
+        }
+
     } // namespace
 } // namespace MGITest

MobileGL/MG_IntegrationTest/Scenarios/PrimitiveRestartScenario.cpp

@@ -0,0 +1,404 @@
+// MobileGL - MobileGL/MG_IntegrationTest/Scenarios/PrimitiveRestartScenario.cpp
+// Copyright (c) 2025-2026 MobileGL-Dev
+// Licensed under the GNU Lesser General Public License v3.0:
+//   https://www.gnu.org/licenses/gpl-3.0.txt
+//   https://www.gnu.org/licenses/lgpl-3.0.txt
+// SPDX-License-Identifier: LGPL-3.0-only
+// End of Source File Header
+//
+// Scenario - DESKTOP GL_PRIMITIVE_RESTART WITH AN APPLICATION-CHOSEN INDEX.
+//
+// Desktop GL restarts on whatever glPrimitiveRestartIndex named; GLES and Vulkan both restart
+// only on the all-ones value of the index type. DirectGLES used to THROW_EXCEPTION on the
+// mismatch, and a throw out of a GL entry point unwinds a C++ exception through the C ABI and
+// kills the process - which is how KHR-GL4x.geometry_shader.primitive_counter.*_rp took the whole
+// conformance runner down, nine bodies at a time, losing every result in the chunk with it.
+//
+// So the first thing this asserts is simply that the process is still here. The second is that
+// the restart actually happened: the substitution rewrites the index data so the driver restarts
+// where the application asked, and the difference between "restart honoured" and "restart
+// silently dropped" is a triangle strip that welds its two halves together across the gap.
+//
+// Needs a real context on purpose. The GPU-free suite cannot reach a backend at all, and this is
+// entirely about what the backend does with the index buffer.
+
+#include <cstddef>
+#include <iterator>
+#include <string>
+#include <utility>
+#include <vector>
+
+#include "../Harness/HeadlessGL.h"
+#include "../Harness/ScenarioFixture.h"
+
+#ifdef GLAPI
+#undef GLAPI
+#endif
+#define GL_GLEXT_PROTOTYPES
+#include <GL/gl.h>
+#include <GL/glcorearb.h>
+#undef GL_GLEXT_PROTOTYPES
+
+namespace MGITest {
+    namespace {
+
+        constexpr GLsizei kSurface = 64;
+
+        const char* const kVertexSource = R"(#version 420 core
+layout(location = 0) in vec2 a_position;
+void main()
+{
+    gl_Position = vec4(a_position, 0.0, 1.0);
+}
+)";
+
+        const char* const kFragmentSource = R"(#version 420 core
+out vec4 fragColor;
+void main()
+{
+    fragColor = vec4(0.0, 1.0, 0.0, 1.0);
+}
+)";
+
+        // Two triangles with a gap down the middle, plus two spare vertices parked at the origin.
+        //
+        // The spares exist so the restart index is a LEGAL vertex index: if the restart were
+        // dropped the driver would still fetch a real vertex rather than read out of bounds, so
+        // the negative case is defined behaviour and the test measures the restart rather than
+        // whatever robust-buffer-access does.
+        constexpr GLfloat kVertices[] = {
+            -0.9f, -0.9f, // 0 - left triangle
+            -0.1f, -0.9f, // 1
+            -0.9f,  0.9f, // 2
+             0.1f, -0.9f, // 3 - right triangle
+             0.9f, -0.9f, // 4
+             0.9f,  0.9f, // 5
+             0.0f,  0.0f, // 6 - spare
+             0.0f,  0.0f, // 7 - spare, and the application's restart index
+        };
+        constexpr GLuint kRestartIndex = 7;
+
+        // A triangle STRIP, restarted in the middle: honoured, it is exactly the two triangles
+        // above. Dropped, the strip welds vertices 2, 7 and 3 into extra triangles that spill
+        // across the gap - which is what the middle probe below catches.
+        constexpr GLuint kIndices[] = {0, 1, 2, kRestartIndex, 3, 4, 5};
+
+        struct Pixel {
+            GLubyte r = 0, g = 0, b = 0, a = 0;
+        };
+
+        class PrimitiveRestartScenario : public ScenarioTest {
+        protected:
+            void SetUp() override {
+                ScenarioTest::SetUp();
+                if (!Ready()) return;
+
+                glGenVertexArrays(1, &m_vao);
+                glBindVertexArray(m_vao);
+
+                glGenBuffers(1, &m_vbo);
+                glBindBuffer(GL_ARRAY_BUFFER, m_vbo);
+                glBufferData(GL_ARRAY_BUFFER, sizeof(kVertices), kVertices, GL_STATIC_DRAW);
+                glVertexAttribPointer(0, 2, GL_FLOAT, GL_FALSE, 2 * sizeof(GLfloat), nullptr);
+                glEnableVertexAttribArray(0);
+
+                glGenBuffers(1, &m_ebo);
+                glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, m_ebo);
+                glBufferData(GL_ELEMENT_ARRAY_BUFFER, sizeof(kIndices), kIndices, GL_STATIC_DRAW);
+
+                glGenTextures(1, &m_colorTexture);
+                glBindTexture(GL_TEXTURE_2D, m_colorTexture);
+                glTexStorage2D(GL_TEXTURE_2D, 1, GL_RGBA8, kSurface, kSurface);
+                glGenFramebuffers(1, &m_fbo);
+                glBindFramebuffer(GL_FRAMEBUFFER, m_fbo);
+                glFramebufferTexture2D(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, m_colorTexture, 0);
+                ASSERT_EQ(glCheckFramebufferStatus(GL_FRAMEBUFFER),
+                          static_cast<GLenum>(GL_FRAMEBUFFER_COMPLETE));
+                glViewport(0, 0, kSurface, kSurface);
+
+                m_program = BuildProgram();
+                ASSERT_NE(m_program, 0u) << "the flat-colour program did not build: " << m_buildLog;
+                glUseProgram(m_program);
+                DrainErrors();
+            }
+
+            void TearDown() override {
+                if (!Ready()) return;
+                glDisable(GL_PRIMITIVE_RESTART);
+                glDisable(GL_PRIMITIVE_RESTART_FIXED_INDEX);
+                glPrimitiveRestartIndex(0);
+                glUseProgram(0);
+                if (m_program != 0) glDeleteProgram(m_program);
+                glBindFramebuffer(GL_FRAMEBUFFER, 0);
+                if (m_fbo != 0) glDeleteFramebuffers(1, &m_fbo);
+                if (m_colorTexture != 0) glDeleteTextures(1, &m_colorTexture);
+                glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, 0);
+                glBindBuffer(GL_ARRAY_BUFFER, 0);
+                if (m_ebo != 0) glDeleteBuffers(1, &m_ebo);
+                if (m_vbo != 0) glDeleteBuffers(1, &m_vbo);
+                glBindVertexArray(0);
+                if (m_vao != 0) glDeleteVertexArrays(1, &m_vao);
+                DrainErrors();
+            }
+
+            static void DrainErrors() {
+                for (int i = 0; i < 16 && glGetError() != GL_NO_ERROR; ++i) {
+                }
+            }
+
+            GLuint BuildProgram() {
+                const GLuint vs = glCreateShader(GL_VERTEX_SHADER);
+                glShaderSource(vs, 1, &kVertexSource, nullptr);
+                glCompileShader(vs);
+                const GLuint fs = glCreateShader(GL_FRAGMENT_SHADER);
+                glShaderSource(fs, 1, &kFragmentSource, nullptr);
+                glCompileShader(fs);
+                const GLuint program = glCreateProgram();
+                glAttachShader(program, vs);
+                glAttachShader(program, fs);
+                glLinkProgram(program);
+                GLint linked = 0;
+                glGetProgramiv(program, GL_LINK_STATUS, &linked);
+                glDeleteShader(vs);
+                glDeleteShader(fs);
+                if (!linked) {
+                    GLint length = 0;
+                    glGetProgramiv(program, GL_INFO_LOG_LENGTH, &length);
+                    std::vector<char> buffer(static_cast<std::size_t>(length) + 1, '\0');
+                    glGetProgramInfoLog(program, length + 1, nullptr, buffer.data());
+                    m_buildLog = buffer.data();
+                    glDeleteProgram(program);
+                    return 0;
+                }
+                return program;
+            }
+
+            // The whole surface, so a failure can report the three probes together rather than
+            // three separate readbacks that might disagree about which draw they saw.
+            std::vector<Pixel> DrawAndRead() {
+                glClearColor(0.0f, 0.0f, 0.0f, 1.0f);
+                glClear(GL_COLOR_BUFFER_BIT);
+                glDrawElements(GL_TRIANGLE_STRIP, static_cast<GLsizei>(std::size(kIndices)), GL_UNSIGNED_INT,
+                               nullptr);
+                std::vector<Pixel> pixels(static_cast<std::size_t>(kSurface) * kSurface);
+                glReadPixels(0, 0, kSurface, kSurface, GL_RGBA, GL_UNSIGNED_BYTE, pixels.data());
+                return pixels;
+            }
+
+            static const Pixel& At(const std::vector<Pixel>& pixels, int x, int y) {
+                return pixels[static_cast<std::size_t>(y) * kSurface + x];
+            }
+
+            static bool IsGreen(const Pixel& p) { return p.g > 128 && p.r < 128; }
+
+            // NDC (-0.5, -0.5): well inside the left triangle whichever way the restart went.
+            static constexpr int kLeftX = 16, kLeftY = 16;
+            // NDC (0.6, -0.5): well inside the right triangle, and outside every welded one.
+            static constexpr int kRightX = 51, kRightY = 16;
+            // NDC (0.2, -0.5): in the gap between the two triangles, and INSIDE the triangle the
+            // strip welds out of vertices 7, 3 and 4 when the restart is dropped. This is the
+            // probe that distinguishes a working restart from a silently ignored one.
+            static constexpr int kGapX = 38, kGapY = 16;
+
+            GLuint m_vao = 0;
+            GLuint m_vbo = 0;
+            GLuint m_ebo = 0;
+            GLuint m_fbo = 0;
+            GLuint m_colorTexture = 0;
+            GLuint m_program = 0;
+            std::string m_buildLog;
+        };
+
+        // THE crash regression. Before the fix this call never returned: DirectGLES threw
+        // std::runtime_error out of glDrawElements and the process died on the spot. Reaching the
+        // assertion at all is most of the point.
+        TEST_F(PrimitiveRestartScenario, AnArbitraryRestartIndexDrawsInsteadOfKillingTheProcess) {
+            if (!Ready()) GTEST_SKIP();
+
+            glEnable(GL_PRIMITIVE_RESTART);
+            glPrimitiveRestartIndex(kRestartIndex);
+            ASSERT_EQ(glGetError(), static_cast<GLenum>(GL_NO_ERROR));
+
+            const std::vector<Pixel> pixels = DrawAndRead();
+            EXPECT_EQ(glGetError(), static_cast<GLenum>(GL_NO_ERROR))
+                << "an arbitrary restart index is legal desktop GL and must raise no error";
+
+            EXPECT_TRUE(IsGreen(At(pixels, kLeftX, kLeftY))) << "the first strip half did not render";
+            EXPECT_TRUE(IsGreen(At(pixels, kRightX, kRightY))) << "the second strip half did not render";
+            EXPECT_FALSE(IsGreen(At(pixels, kGapX, kGapY)))
+                << "the gap between the two halves is covered, so the restart was dropped and the "
+                   "strip welded across it";
+        }
+
+        // The other half of the state: an application that sets the restart index TO the fixed
+        // all-ones value needs no rewriting at all, and the cap must map straight onto the
+        // driver's own fixed-index restart. Same picture, different path through the backend.
+        TEST_F(PrimitiveRestartScenario, TheFixedIndexValueTakesTheForwardingPath) {
+            if (!Ready()) GTEST_SKIP();
+
+            // Index 0xFFFFFFFF is not a vertex this draw uses, so the strip is the same shape.
+            const GLuint fixedIndices[] = {0, 1, 2, 0xFFFFFFFFu, 3, 4, 5};
+            glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, m_ebo);
+            glBufferSubData(GL_ELEMENT_ARRAY_BUFFER, 0, sizeof(fixedIndices), fixedIndices);
+
+            glEnable(GL_PRIMITIVE_RESTART);
+            glPrimitiveRestartIndex(0xFFFFFFFFu);
+            ASSERT_EQ(glGetError(), static_cast<GLenum>(GL_NO_ERROR));
+
+            const std::vector<Pixel> pixels = DrawAndRead();
+            EXPECT_EQ(glGetError(), static_cast<GLenum>(GL_NO_ERROR));
+            EXPECT_TRUE(IsGreen(At(pixels, kLeftX, kLeftY)));
+            EXPECT_TRUE(IsGreen(At(pixels, kRightX, kRightY)));
+            EXPECT_FALSE(IsGreen(At(pixels, kGapX, kGapY)));
+
+            // Put the buffer back for whatever runs next in this fixture.
+            glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, m_ebo);
+            glBufferSubData(GL_ELEMENT_ARRAY_BUFFER, 0, sizeof(kIndices), kIndices);
+            DrainErrors();
+        }
+
+        // With the cap off, the same index data is just data - nothing restarts, and the strip
+        // welds across the gap. The negative control for the probe above: without it, a backend
+        // that lost the whole draw would pass the test by rendering nothing in the gap.
+        TEST_F(PrimitiveRestartScenario, WithoutTheCapTheStripWeldsAcrossTheGap) {
+            if (!Ready()) GTEST_SKIP();
+
+            glDisable(GL_PRIMITIVE_RESTART);
+            glPrimitiveRestartIndex(kRestartIndex);
+            ASSERT_EQ(glGetError(), static_cast<GLenum>(GL_NO_ERROR));
+
+            const std::vector<Pixel> pixels = DrawAndRead();
+            EXPECT_EQ(glGetError(), static_cast<GLenum>(GL_NO_ERROR));
+            EXPECT_TRUE(IsGreen(At(pixels, kLeftX, kLeftY))) << "the draw itself must still happen";
+            EXPECT_TRUE(IsGreen(At(pixels, kGapX, kGapY)))
+                << "with restart disabled the strip is continuous, so the gap must be covered - if "
+                   "it is not, the probe above proves nothing";
+        }
+
+        // A second draw with a DIFFERENT restart index has to be rewritten again. The substitution
+        // stages through one scratch buffer, so a cached or half-restored element-array binding
+        // would show up here as the second draw reusing the first one's data.
+        TEST_F(PrimitiveRestartScenario, ChangingTheRestartIndexBetweenDrawsIsHonoured) {
+            if (!Ready()) GTEST_SKIP();
+
+            glEnable(GL_PRIMITIVE_RESTART);
+            glPrimitiveRestartIndex(kRestartIndex);
+            const std::vector<Pixel> restarted = DrawAndRead();
+            ASSERT_EQ(glGetError(), static_cast<GLenum>(GL_NO_ERROR));
+            EXPECT_FALSE(IsGreen(At(restarted, kGapX, kGapY)));
+
+            // 6 is the other spare vertex, and it appears nowhere in the index data - so nothing
+            // restarts and the strip is continuous again, from the very same buffer.
+            glPrimitiveRestartIndex(6);
+            const std::vector<Pixel> notRestarted = DrawAndRead();
+            ASSERT_EQ(glGetError(), static_cast<GLenum>(GL_NO_ERROR));
+            EXPECT_TRUE(IsGreen(At(notRestarted, kLeftX, kLeftY)));
+            EXPECT_TRUE(IsGreen(At(notRestarted, kGapX, kGapY)))
+                << "the second draw restarted on an index that is not in its data";
+        }
+
+        // A NON-indexed draw has no index stream, so GL primitive restart cannot affect it - and a
+        // list topology is the shape DirectVulkan has to refuse when the device lacks
+        // VK_EXT_primitive_topology_list_restart. Deriving the pipeline's primitiveRestartEnable
+        // from the capability bits alone conflated the two: an application that enables
+        // GL_PRIMITIVE_RESTART once at init and then draws its UI with glDrawArrays(GL_TRIANGLES)
+        // had every one of those draws silently dropped on such a device.
+        TEST_F(PrimitiveRestartScenario, ANonIndexedListTopologyDrawIsUnaffectedByTheCap) {
+            if (!Ready()) GTEST_SKIP();
+
+            glEnable(GL_PRIMITIVE_RESTART);
+            glPrimitiveRestartIndex(kRestartIndex);
+            ASSERT_EQ(glGetError(), static_cast<GLenum>(GL_NO_ERROR));
+
+            glClearColor(0.0f, 0.0f, 0.0f, 1.0f);
+            glClear(GL_COLOR_BUFFER_BIT);
+            // Vertices 0,1,2 are the left triangle; GL_TRIANGLES is a list topology.
+            glDrawArrays(GL_TRIANGLES, 0, 3);
+            EXPECT_EQ(glGetError(), static_cast<GLenum>(GL_NO_ERROR));
+
+            std::vector<Pixel> pixels(static_cast<std::size_t>(kSurface) * kSurface);
+            glReadPixels(0, 0, kSurface, kSurface, GL_RGBA, GL_UNSIGNED_BYTE, pixels.data());
+            EXPECT_TRUE(IsGreen(At(pixels, kLeftX, kLeftY)))
+                << "primitive restart has no meaning for glDrawArrays, so the draw must render "
+                   "normally whatever the device supports";
+            DrainErrors();
+        }
+
+        // GL 4.6 core 10.3.6 compares the fetched index, zero-extended, against the full 32-bit
+        // PRIMITIVE_RESTART_INDEX. A restart index the index type cannot hold therefore matches
+        // nothing and the draw restarts NOWHERE - it does not restart on the truncated value, and
+        // it does not restart on the type's all-ones value either, which is what the driver's own
+        // fixed-index restart would have done if it had been left enabled.
+        TEST_F(PrimitiveRestartScenario, ARestartIndexTooLargeForTheIndexTypeRestartsNowhere) {
+            if (!Ready()) GTEST_SKIP();
+
+            // 16-bit indices with a restart index of 0x10007: the low half (7) IS a real index in
+            // the data, so a truncating comparison would split the strip exactly where a correct
+            // one leaves it whole.
+            const GLushort shortIndices[] = {0, 1, 2, static_cast<GLushort>(kRestartIndex), 3, 4, 5};
+            glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, m_ebo);
+            glBufferData(GL_ELEMENT_ARRAY_BUFFER, sizeof(shortIndices), shortIndices, GL_STATIC_DRAW);
+
+            glEnable(GL_PRIMITIVE_RESTART);
+            glPrimitiveRestartIndex(0x10000u + kRestartIndex);
+            ASSERT_EQ(glGetError(), static_cast<GLenum>(GL_NO_ERROR));
+
+            glClearColor(0.0f, 0.0f, 0.0f, 1.0f);
+            glClear(GL_COLOR_BUFFER_BIT);
+            glDrawElements(GL_TRIANGLE_STRIP, static_cast<GLsizei>(std::size(shortIndices)), GL_UNSIGNED_SHORT,
+                           nullptr);
+            EXPECT_EQ(glGetError(), static_cast<GLenum>(GL_NO_ERROR));
+
+            std::vector<Pixel> pixels(static_cast<std::size_t>(kSurface) * kSurface);
+            glReadPixels(0, 0, kSurface, kSurface, GL_RGBA, GL_UNSIGNED_BYTE, pixels.data());
+            EXPECT_TRUE(IsGreen(At(pixels, kLeftX, kLeftY)));
+            EXPECT_TRUE(IsGreen(At(pixels, kGapX, kGapY)))
+                << "no 16-bit index can equal 0x10007, so nothing restarts and the strip is "
+                   "continuous - truncating the restart index to 7 would split it here";
+
+            glBufferData(GL_ELEMENT_ARRAY_BUFFER, sizeof(kIndices), kIndices, GL_STATIC_DRAW);
+            DrainErrors();
+        }
+
+        // The all-ones value of an index type is an ordinary vertex index whenever the array uses
+        // the type's full range, which is exactly why an application picks an arbitrary restart
+        // index in the first place. Substituting the sentinel in place would either steal that
+        // vertex or spuriously restart on it, so the copy widens instead - and the draw has to be
+        // issued with the widened type, which is the part that is easy to forget.
+        TEST_F(PrimitiveRestartScenario, AnAllOnesVertexIndexSurvivesTheSubstitution) {
+            if (!Ready()) GTEST_SKIP();
+
+            // The buffer carries the 16-bit all-ones value as an ordinary element. It sits past
+            // the seven indices this draw reads, because the vertex array has only eight entries
+            // and fetching index 65535 would be out of range - what is under test is that its
+            // mere PRESENCE forces the widened copy, and that the draw still finds its own
+            // indices at the right offsets in a copy whose element width has changed underneath
+            // it. Narrowly substituting in place instead would rewrite this element to 0xFFFE.
+            const GLushort shortIndices[] = {0, 1, 2, static_cast<GLushort>(kRestartIndex), 3, 4, 5, 0xFFFFu};
+            glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, m_ebo);
+            glBufferData(GL_ELEMENT_ARRAY_BUFFER, sizeof(shortIndices), shortIndices, GL_STATIC_DRAW);
+
+            glEnable(GL_PRIMITIVE_RESTART);
+            glPrimitiveRestartIndex(kRestartIndex);
+            ASSERT_EQ(glGetError(), static_cast<GLenum>(GL_NO_ERROR));
+
+            glClearColor(0.0f, 0.0f, 0.0f, 1.0f);
+            glClear(GL_COLOR_BUFFER_BIT);
+            // Only the first seven indices are drawn, so the 0xFFFF element is never fetched - what
+            // is under test is that its PRESENCE does not break the substitution or the offsets.
+            glDrawElements(GL_TRIANGLE_STRIP, 7, GL_UNSIGNED_SHORT, nullptr);
+            EXPECT_EQ(glGetError(), static_cast<GLenum>(GL_NO_ERROR));
+
+            std::vector<Pixel> pixels(static_cast<std::size_t>(kSurface) * kSurface);
+            glReadPixels(0, 0, kSurface, kSurface, GL_RGBA, GL_UNSIGNED_BYTE, pixels.data());
+            EXPECT_TRUE(IsGreen(At(pixels, kLeftX, kLeftY))) << "the first strip half did not render";
+            EXPECT_TRUE(IsGreen(At(pixels, kRightX, kRightY))) << "the second strip half did not render";
+            EXPECT_FALSE(IsGreen(At(pixels, kGapX, kGapY)))
+                << "the restart still has to happen once the copy has been widened";
+
+            glBufferData(GL_ELEMENT_ARRAY_BUFFER, sizeof(kIndices), kIndices, GL_STATIC_DRAW);
+            DrainErrors();
+        }
+
+    } // namespace
+} // namespace MGITest

MobileGL/MG_IntegrationTest/Scenarios/RenderbufferBlendFormatScenario.cpp

@@ -0,0 +1,227 @@
+// MobileGL - MobileGL/MG_IntegrationTest/Scenarios/RenderbufferBlendFormatScenario.cpp
+// Copyright (c) 2025-2026 MobileGL-Dev
+// Licensed under the GNU Lesser General Public License v3.0:
+//   https://www.gnu.org/licenses/gpl-3.0.txt
+//   https://www.gnu.org/licenses/lgpl-3.0.txt
+// SPDX-License-Identifier: LGPL-3.0-only
+// End of Source File Header
+//
+// Scenario - BLENDING WORKS ON A RENDERBUFFER WHOSE GL FORMAT HAS NO EXACT VkFormat.
+//
+// DirectVulkan force-disables blending on an attachment whose VkFormat lacks
+// VK_FORMAT_FEATURE_COLOR_ATTACHMENT_BLEND_BIT, which is the right thing to do - blending on such a
+// format is invalid pipeline state. The probe has to ask about the format the attachment ACTUALLY
+// has, and for renderbuffers it asked a different question from the one that created the image: the
+// image comes from ResolveTextureFormatInfo (which widens GL formats with no Vulkan twin onto a real
+// one) while the probe used the strict 1:1 converter, which answers VK_FORMAT_UNDEFINED for RGBA2,
+// RGBA12, RGB10, RGB12, RGB16 and the three-channel formats, and the 16-bit packed formats for RGBA4
+// and RGB5_A1.
+//
+// VkFormatProperties for VK_FORMAT_UNDEFINED are all zero, so the probe concluded "not blendable"
+// and every pipeline for that attachment was built with blendEnable = VK_FALSE - permanently, and
+// silently apart from one log line. The source colour then overwrites the destination instead of
+// blending with it, which is a wrong PICTURE, not a wrong error code.
+//
+// GL_RGB8 is the ordinary shape and is what this scenario leads with: it is a required
+// colour-renderable format, its image has been R8G8B8A8_UNORM all along, and the probe asked about
+// the 24-bit R8G8B8_UNORM that most drivers do not support at all. GL_RGBA4 covers the other half -
+// a format whose probe answered a real-but-different VkFormat.
+//
+// DirectGLES is the control: it forwards the renderbuffer to the ES driver and blends whatever the
+// driver blends, so a disagreement between the two backends is the defect.
+
+#include <cstdint>
+#include <string>
+#include <vector>
+
+#include "../Harness/HeadlessGL.h"
+#include "../Harness/ScenarioFixture.h"
+
+#ifdef GLAPI
+#undef GLAPI
+#endif
+#define GL_GLEXT_PROTOTYPES
+#include <GL/gl.h>
+#include <GL/glcorearb.h>
+#undef GL_GLEXT_PROTOTYPES
+
+namespace MGITest {
+    namespace {
+
+        constexpr int kExtent = 16;
+
+        constexpr const char* kVertexSource = R"(#version 330 core
+void main()
+{
+    switch (gl_VertexID)
+    {
+      case 0: gl_Position = vec4(-1.0, 1.0, 0.0, 1.0); break;
+      case 1: gl_Position = vec4( 1.0, 1.0, 0.0, 1.0); break;
+      case 2: gl_Position = vec4(-1.0,-1.0, 0.0, 1.0); break;
+      case 3: gl_Position = vec4( 1.0,-1.0, 0.0, 1.0); break;
+    }
+}
+)";
+
+        constexpr const char* kFragmentSource = R"(#version 330 core
+uniform vec4 uColor;
+out vec4 fragColor;
+void main()
+{
+    fragColor = uColor;
+}
+)";
+
+        class RenderbufferBlendFormatScenario : public ScenarioTest {
+        protected:
+            void SetUp() override {
+                ScenarioTest::SetUp();
+                if (!Ready()) return;
+                glGenVertexArrays(1, &m_vao);
+                std::string error;
+                m_program = CompileProgram(kVertexSource, kFragmentSource, &error);
+                ASSERT_NE(m_program, 0u) << "program did not build: " << error;
+                ASSERT_EQ(FirstGLError(), 0u);
+            }
+
+            void TearDown() override {
+                if (!Ready()) return;
+                Destroy();
+                if (m_program != 0) glDeleteProgram(m_program);
+                if (m_vao != 0) glDeleteVertexArrays(1, &m_vao);
+                glBindFramebuffer(GL_FRAMEBUFFER, 0);
+            }
+
+            void Destroy() {
+                if (m_fbo != 0) {
+                    glBindFramebuffer(GL_FRAMEBUFFER, 0);
+                    glDeleteFramebuffers(1, &m_fbo);
+                    m_fbo = 0;
+                }
+                if (m_renderbuffer != 0) {
+                    glDeleteRenderbuffers(1, &m_renderbuffer);
+                    m_renderbuffer = 0;
+                }
+            }
+
+            // Returns false (having skipped, not failed) when the driver will not give us a complete
+            // framebuffer for this format - GL only requires a subset of formats to be
+            // colour-renderable, and the point of the scenario is blending, not format support.
+            bool MakeTarget(GLenum internalFormat) {
+                Destroy();
+                glGenRenderbuffers(1, &m_renderbuffer);
+                glBindRenderbuffer(GL_RENDERBUFFER, m_renderbuffer);
+                glRenderbufferStorage(GL_RENDERBUFFER, internalFormat, kExtent, kExtent);
+                glGenFramebuffers(1, &m_fbo);
+                glBindFramebuffer(GL_FRAMEBUFFER, m_fbo);
+                glFramebufferRenderbuffer(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_RENDERBUFFER, m_renderbuffer);
+                const GLenum status = glCheckFramebufferStatus(GL_FRAMEBUFFER);
+                for (int i = 0; i < 16 && glGetError() != GL_NO_ERROR; ++i) {
+                }
+                return status == GL_FRAMEBUFFER_COMPLETE;
+            }
+
+            void DrawColor(float r, float g, float b, float a) {
+                glUseProgram(m_program);
+                glUniform4f(glGetUniformLocation(m_program, "uColor"), r, g, b, a);
+                glBindVertexArray(m_vao);
+                glDrawArrays(GL_TRIANGLE_STRIP, 0, 4);
+                glBindVertexArray(0);
+                glUseProgram(0);
+            }
+
+            GLuint m_renderbuffer = 0;
+            GLuint m_fbo = 0;
+            GLuint m_vao = 0;
+            unsigned int m_program = 0;
+        };
+
+        // One draw of opaque black, then a 50%-alpha white draw over it with the ordinary
+        // SRC_ALPHA / ONE_MINUS_SRC_ALPHA function. Blending gives mid-grey; a pipeline built with
+        // blendEnable = VK_FALSE gives white, because the source simply overwrites.
+        //
+        // The tolerance is wide on purpose: RGBA4 has four bits per channel, so "mid-grey" is one of
+        // a handful of representable values and the test must not become a quantisation test.
+        void ExpectBlendedRatherThanOverwritten(const char* what) {
+            const Image image = ReadPixels(kExtent, kExtent);
+            ASSERT_FALSE(image.Empty()) << what;
+            const Rgba8 centre = image.At(kExtent / 2, kExtent / 2);
+            EXPECT_GT(int(centre.r), 40) << what << ": got " << centre << ", which is darker than a blend of "
+                                                              "black and 50% white";
+            EXPECT_LT(int(centre.r), 215) << what << ": got " << centre
+                                          << ", which is the source colour - blending was disabled";
+        }
+
+    } // namespace
+
+    // The ordinary case, and the one broken today rather than only after the format table was
+    // unified: a three-channel colour renderbuffer. Its image has been R8G8B8A8_UNORM all along while
+    // the blend probe asked about R8G8B8_UNORM, which most drivers do not support at all.
+    TEST_F(RenderbufferBlendFormatScenario, BlendingWorksOnAThreeChannelRenderbuffer) {
+        if (!Ready()) GTEST_SKIP();
+        if (!MakeTarget(GL_RGB8)) GTEST_SKIP() << "GL_RGB8 renderbuffer is not framebuffer-complete here";
+
+        glViewport(0, 0, kExtent, kExtent);
+        glDisable(GL_SCISSOR_TEST);
+        glDisable(GL_DEPTH_TEST);
+        glDisable(GL_BLEND);
+        DrawColor(0.0f, 0.0f, 0.0f, 1.0f);
+
+        glEnable(GL_BLEND);
+        glBlendFunc(GL_SRC_ALPHA, GL_ONE_MINUS_SRC_ALPHA);
+        DrawColor(1.0f, 1.0f, 1.0f, 0.5f);
+        glDisable(GL_BLEND);
+        EXPECT_EQ(FirstGLError(), 0u) << "the blended draw left a GL error behind";
+
+        ExpectBlendedRatherThanOverwritten("GL_RGB8");
+        Gl().EndFrame();
+    }
+
+    // The other half: a format whose strict converter answers a real-but-different VkFormat
+    // (R4G4B4A4_UNORM_PACK16) while the image is R8G8B8A8_UNORM. Blend support for the packed 16-bit
+    // formats is optional in Vulkan, so the probe could legitimately answer "no" for a format the
+    // attachment does not have.
+    TEST_F(RenderbufferBlendFormatScenario, BlendingWorksOnALowBitPackedRenderbuffer) {
+        if (!Ready()) GTEST_SKIP();
+        if (!MakeTarget(GL_RGBA4)) GTEST_SKIP() << "GL_RGBA4 renderbuffer is not framebuffer-complete here";
+
+        glViewport(0, 0, kExtent, kExtent);
+        glDisable(GL_SCISSOR_TEST);
+        glDisable(GL_DEPTH_TEST);
+        glDisable(GL_BLEND);
+        DrawColor(0.0f, 0.0f, 0.0f, 1.0f);
+
+        glEnable(GL_BLEND);
+        glBlendFunc(GL_SRC_ALPHA, GL_ONE_MINUS_SRC_ALPHA);
+        DrawColor(1.0f, 1.0f, 1.0f, 0.5f);
+        glDisable(GL_BLEND);
+        EXPECT_EQ(FirstGLError(), 0u) << "the blended draw left a GL error behind";
+
+        ExpectBlendedRatherThanOverwritten("GL_RGBA4");
+        Gl().EndFrame();
+    }
+
+    // The control that keeps both of the above honest: the same sequence on the format whose probe
+    // and image always agreed. If this one ever fails, the scenario is measuring the blend setup
+    // rather than the format resolution.
+    TEST_F(RenderbufferBlendFormatScenario, BlendingWorksOnAnRgba8Renderbuffer) {
+        if (!Ready()) GTEST_SKIP();
+        if (!MakeTarget(GL_RGBA8)) GTEST_SKIP() << "GL_RGBA8 renderbuffer is not framebuffer-complete here";
+
+        glViewport(0, 0, kExtent, kExtent);
+        glDisable(GL_SCISSOR_TEST);
+        glDisable(GL_DEPTH_TEST);
+        glDisable(GL_BLEND);
+        DrawColor(0.0f, 0.0f, 0.0f, 1.0f);
+
+        glEnable(GL_BLEND);
+        glBlendFunc(GL_SRC_ALPHA, GL_ONE_MINUS_SRC_ALPHA);
+        DrawColor(1.0f, 1.0f, 1.0f, 0.5f);
+        glDisable(GL_BLEND);
+        EXPECT_EQ(FirstGLError(), 0u) << "the blended draw left a GL error behind";
+
+        ExpectBlendedRatherThanOverwritten("GL_RGBA8");
+        Gl().EndFrame();
+    }
+
+} // namespace MGITest

MobileGL/MG_IntegrationTest/Scenarios/SampleVariablesScenario.cpp

@@ -0,0 +1,274 @@
+// MobileGL - MobileGL/MG_IntegrationTest/Scenarios/SampleVariablesScenario.cpp
+// Copyright (c) 2025-2026 MobileGL-Dev
+// Licensed under the GNU Lesser General Public License v3.0:
+//   https://www.gnu.org/licenses/gpl-3.0.txt
+//   https://www.gnu.org/licenses/lgpl-3.0.txt
+// SPDX-License-Identifier: LGPL-3.0-only
+// End of Source File Header
+//
+// Scenario - gl_NumSamples REACHES THE SHADER, AND IT FOLLOWS THE DRAW FRAMEBUFFER.
+//
+// glslang declares gl_NumSamples only when it is NOT targeting SPIR-V - both the desktop and the
+// ES branch of Initialize.cpp wrap `uniform int gl_NumSamples;` in `if (spvVersion.spv == 0)`,
+// because SPIR-V has no NumSamples builtin to lower it to - and MobileGL always targets SPIR-V.
+// Every fragment shader that read the built-in therefore died at COMPILE time with
+// "'gl_NumSamples' : undeclared identifier", which is all 144 KHR-GL46.sample_variables.mask.*
+// bodies plus their es_31_compatibility twins.
+//
+// The source pipeline now lowers it onto a reserved default-block uniform and the draw path writes
+// the current draw framebuffer's sample count into it. Two claims, and the second is the one a
+// compile-only test cannot make: the value must be the DRAW FRAMEBUFFER's, so one program drawn
+// into a multisample target and then into a single-sample target has to report both counts. A
+// link-time bake would pass the first assertion and fail the second, which is exactly why the
+// write lives per draw.
+//
+// llvmpipe and lavapipe both offer 4x multisample RGBA8, so this runs for real in CI rather than
+// skipping; the skips below are for a driver that offers no multisample renderbuffer at all.
+
+#include <algorithm>
+#include <string>
+
+#include "../Harness/HeadlessGL.h"
+#include "../Harness/ScenarioFixture.h"
+
+#ifdef GLAPI
+#undef GLAPI
+#endif
+#define GL_GLEXT_PROTOTYPES
+#include <GL/gl.h>
+#include <GL/glcorearb.h>
+#undef GL_GLEXT_PROTOTYPES
+
+namespace MGITest {
+    namespace {
+
+        constexpr const char* kVS = R"(#version 400 core
+in vec2 aPos;
+void main() { gl_Position = vec4(aPos, 0.0, 1.0); }
+)";
+
+        // gl_NumSamples scaled so each count lands on its own well-separated 8-bit value: 1 -> 16,
+        // 2 -> 32, 4 -> 64. Every sample of the fragment gets the same colour, so the resolve blit
+        // averages identical values and the readback is exact rather than approximate.
+        constexpr const char* kFS = R"(#version 400 core
+out vec4 o_color;
+void main() { o_color = vec4(float(gl_NumSamples) * (16.0 / 255.0), 0.0, 0.0, 1.0); }
+)";
+
+        class SampleVariablesScenario : public ScenarioTest {};
+
+        void DrawFullViewportQuad(unsigned int program) {
+            static const float kQuad[] = {-1.0f, -1.0f, 1.0f, -1.0f, -1.0f, 1.0f, 1.0f, 1.0f};
+            GLuint vao = 0, vbo = 0;
+            glGenVertexArrays(1, &vao);
+            glBindVertexArray(vao);
+            glGenBuffers(1, &vbo);
+            glBindBuffer(GL_ARRAY_BUFFER, vbo);
+            glBufferData(GL_ARRAY_BUFFER, sizeof(kQuad), kQuad, GL_STATIC_DRAW);
+            glEnableVertexAttribArray(0);
+            glVertexAttribPointer(0, 2, GL_FLOAT, GL_FALSE, 2 * sizeof(float), nullptr);
+            glUseProgram(program);
+            glDrawArrays(GL_TRIANGLE_STRIP, 0, 4);
+            glBindVertexArray(0);
+            glDeleteBuffers(1, &vbo);
+            glDeleteVertexArrays(1, &vao);
+        }
+
+    } // namespace
+
+    TEST_F(SampleVariablesScenario, GlNumSamplesFollowsTheDrawFramebuffersSampleCount) {
+        if (!Ready()) return;
+        HeadlessGL& gl = Gl();
+        const int width = gl.Width();
+        const int height = gl.Height();
+        ASSERT_GE(width, 8);
+        ASSERT_GE(height, 8);
+
+        std::string error;
+        const unsigned int program = CompileProgram(kVS, kFS, &error);
+        // The compile failure this scenario exists for lands here, with glslang's own text.
+        ASSERT_NE(program, 0u) << error;
+
+        GLint maxSamples = 0;
+        glGetIntegerv(GL_MAX_SAMPLES, &maxSamples);
+        const GLint requestedSamples = std::min<GLint>(maxSamples, 4);
+        if (requestedSamples < 2) {
+            glDeleteProgram(program);
+            GTEST_SKIP() << "GL_MAX_SAMPLES is " << maxSamples << "; this needs a multisample renderbuffer";
+        }
+
+        glDisable(GL_SCISSOR_TEST);
+        glDisable(GL_DEPTH_TEST);
+
+        // ---- multisample target ----
+        GLuint msFbo = 0, msRbo = 0;
+        glGenFramebuffers(1, &msFbo);
+        glBindFramebuffer(GL_FRAMEBUFFER, msFbo);
+        glGenRenderbuffers(1, &msRbo);
+        glBindRenderbuffer(GL_RENDERBUFFER, msRbo);
+        glRenderbufferStorageMultisample(GL_RENDERBUFFER, requestedSamples, GL_RGBA8, width, height);
+        glFramebufferRenderbuffer(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_RENDERBUFFER, msRbo);
+        if (glCheckFramebufferStatus(GL_FRAMEBUFFER) != GL_FRAMEBUFFER_COMPLETE) {
+            glDeleteRenderbuffers(1, &msRbo);
+            glDeleteFramebuffers(1, &msFbo);
+            glBindFramebuffer(GL_FRAMEBUFFER, 0);
+            glDeleteProgram(program);
+            GTEST_SKIP() << "no complete " << requestedSamples << "x multisample RGBA8 renderbuffer on this driver";
+        }
+
+        // What the driver actually allocated - a request is a lower bound, and the shader has to
+        // agree with the query rather than with what was asked for.
+        GLint realizedSamples = 0;
+        glGetIntegerv(GL_SAMPLES, &realizedSamples);
+        ASSERT_GE(realizedSamples, 2) << "the multisample framebuffer reports GL_SAMPLES " << realizedSamples;
+
+        glViewport(0, 0, width, height);
+        ClearTo(0.0f, 0.0f, 0.0f, 1.0f);
+        DrawFullViewportQuad(program);
+        EXPECT_EQ(FirstGLError(), 0u);
+
+        // Resolve into the default framebuffer to read it back.
+        BindDefaultFramebuffer();
+        glViewport(0, 0, width, height);
+        ClearTo(0.0f, 0.0f, 0.0f, 1.0f);
+        glBindFramebuffer(GL_READ_FRAMEBUFFER, msFbo);
+        glBindFramebuffer(GL_DRAW_FRAMEBUFFER, 0);
+        glBlitFramebuffer(0, 0, width, height, 0, 0, width, height, GL_COLOR_BUFFER_BIT, GL_NEAREST);
+        glBindFramebuffer(GL_FRAMEBUFFER, 0);
+        EXPECT_EQ(FirstGLError(), 0u);
+
+        {
+            const Image resolved = ReadPixels(width, height);
+            const Rgba8 centre = resolved.At(width / 2, height / 2);
+            EXPECT_NEAR(centre.r, 16 * realizedSamples, 2)
+                << "gl_NumSamples read " << (centre.r / 16.0) << " into a " << realizedSamples
+                << "-sample framebuffer; 1 means the reserved uniform was never written, 0 means it was "
+                << "written but never uploaded";
+        }
+        gl.EndFrame();
+
+        // ---- the SAME program into a single-sample target ----
+        // A link-time bake of the sample count would keep reporting the multisample value here.
+        GLuint ssFbo = 0, ssRbo = 0;
+        glGenFramebuffers(1, &ssFbo);
+        glBindFramebuffer(GL_FRAMEBUFFER, ssFbo);
+        glGenRenderbuffers(1, &ssRbo);
+        glBindRenderbuffer(GL_RENDERBUFFER, ssRbo);
+        glRenderbufferStorage(GL_RENDERBUFFER, GL_RGBA8, width, height);
+        glFramebufferRenderbuffer(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_RENDERBUFFER, ssRbo);
+        ASSERT_EQ(glCheckFramebufferStatus(GL_FRAMEBUFFER), static_cast<GLenum>(GL_FRAMEBUFFER_COMPLETE));
+
+        glViewport(0, 0, width, height);
+        ClearTo(0.0f, 0.0f, 0.0f, 1.0f);
+        DrawFullViewportQuad(program);
+        EXPECT_EQ(FirstGLError(), 0u);
+
+        {
+            const Image single = ReadPixels(width, height);
+            const Rgba8 centre = single.At(width / 2, height / 2);
+            // GL 4.6 core 15.2.2: gl_NumSamples is ONE for a non-multisample framebuffer, where
+            // glGetIntegerv(GL_SAMPLES) answers zero.
+            EXPECT_NEAR(centre.r, 16, 2)
+                << "gl_NumSamples read " << (centre.r / 16.0)
+                << " into a single-sample framebuffer; the value is a property of the DRAW FRAMEBUFFER, "
+                << "so re-using the program must re-write it";
+        }
+
+        glBindFramebuffer(GL_FRAMEBUFFER, 0);
+        glDeleteRenderbuffers(1, &ssRbo);
+        glDeleteFramebuffers(1, &ssFbo);
+        glDeleteRenderbuffers(1, &msRbo);
+        glDeleteFramebuffers(1, &msFbo);
+        glDeleteProgram(program);
+        gl.EndFrame();
+    }
+
+    // ARB_sample_shading is advertised, and until now glMinSampleShading was a logging no-op while
+    // glEnable(GL_SAMPLE_SHADING) fell out of RenderState::SetCapability's default arm - so an
+    // application could ask for a shading rate and get silence from both halves.
+    //
+    // What this can and cannot assert. The RATE itself is not observable from a portable shader:
+    // GL 4.6 core 14.3.1 makes any use of gl_SampleID or gl_SamplePosition force per-sample
+    // evaluation on its own, so the very built-ins that would report the rate defeat the
+    // measurement. What IS worth pinning is that the state now reaches both backends without
+    // damage: DirectGLES forwards glEnable(GL_SAMPLE_SHADING) + glMinSampleShading to the ES
+    // driver (and must not, on a driver that has neither, push an INVALID_ENUM into the
+    // application's error queue), and DirectVulkan bakes sampleShadingEnable/minSampleShading into
+    // a NEW pipeline - which it may only do with the device's sampleRateShading feature enabled.
+    TEST_F(SampleVariablesScenario, SampleShadingStateReachesTheBackendWithoutDisturbingTheDraw) {
+        if (!Ready()) return;
+        HeadlessGL& gl = Gl();
+        const int width = gl.Width();
+        const int height = gl.Height();
+
+        std::string error;
+        const unsigned int program = CompileProgram(kVS, kFS, &error);
+        ASSERT_NE(program, 0u) << error;
+
+        GLint maxSamples = 0;
+        glGetIntegerv(GL_MAX_SAMPLES, &maxSamples);
+        const GLint requestedSamples = std::min<GLint>(maxSamples, 4);
+        if (requestedSamples < 2) {
+            glDeleteProgram(program);
+            GTEST_SKIP() << "GL_MAX_SAMPLES is " << maxSamples << "; sample shading needs a multisample target";
+        }
+
+        GLuint msFbo = 0, msRbo = 0;
+        glGenFramebuffers(1, &msFbo);
+        glBindFramebuffer(GL_FRAMEBUFFER, msFbo);
+        glGenRenderbuffers(1, &msRbo);
+        glBindRenderbuffer(GL_RENDERBUFFER, msRbo);
+        glRenderbufferStorageMultisample(GL_RENDERBUFFER, requestedSamples, GL_RGBA8, width, height);
+        glFramebufferRenderbuffer(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_RENDERBUFFER, msRbo);
+        if (glCheckFramebufferStatus(GL_FRAMEBUFFER) != GL_FRAMEBUFFER_COMPLETE) {
+            glDeleteRenderbuffers(1, &msRbo);
+            glDeleteFramebuffers(1, &msFbo);
+            glBindFramebuffer(GL_FRAMEBUFFER, 0);
+            glDeleteProgram(program);
+            GTEST_SKIP() << "no complete " << requestedSamples << "x multisample RGBA8 renderbuffer on this driver";
+        }
+
+        GLint realizedSamples = 0;
+        glGetIntegerv(GL_SAMPLES, &realizedSamples);
+        glDisable(GL_SCISSOR_TEST);
+        glDisable(GL_DEPTH_TEST);
+        glViewport(0, 0, width, height);
+
+        glEnable(GL_SAMPLE_SHADING);
+        glMinSampleShading(1.0f);
+        EXPECT_EQ(glIsEnabled(GL_SAMPLE_SHADING), static_cast<GLboolean>(GL_TRUE));
+        GLfloat rate = -1.0f;
+        glGetFloatv(GL_MIN_SAMPLE_SHADING_VALUE, &rate);
+        EXPECT_FLOAT_EQ(rate, 1.0f);
+        EXPECT_EQ(FirstGLError(), 0u) << "enabling sample shading raised a GL error";
+
+        ClearTo(0.0f, 0.0f, 0.0f, 1.0f);
+        DrawFullViewportQuad(program);
+        EXPECT_EQ(FirstGLError(), 0u) << "the sample-shading draw raised a GL error";
+
+        BindDefaultFramebuffer();
+        glViewport(0, 0, width, height);
+        ClearTo(0.0f, 0.0f, 0.0f, 1.0f);
+        glBindFramebuffer(GL_READ_FRAMEBUFFER, msFbo);
+        glBindFramebuffer(GL_DRAW_FRAMEBUFFER, 0);
+        glBlitFramebuffer(0, 0, width, height, 0, 0, width, height, GL_COLOR_BUFFER_BIT, GL_NEAREST);
+        glBindFramebuffer(GL_FRAMEBUFFER, 0);
+
+        const Image resolved = ReadPixels(width, height);
+        const Rgba8 centre = resolved.At(width / 2, height / 2);
+        // The rate changes how OFTEN the shader runs, never what it computes - so the same
+        // gl_NumSamples reading has to come back.
+        EXPECT_NEAR(centre.r, 16 * realizedSamples, 2)
+            << "the draw changed its result once sample shading was enabled";
+
+        glMinSampleShading(0.0f);
+        glDisable(GL_SAMPLE_SHADING);
+        EXPECT_EQ(FirstGLError(), 0u);
+
+        glDeleteRenderbuffers(1, &msRbo);
+        glDeleteFramebuffers(1, &msFbo);
+        glDeleteProgram(program);
+        gl.EndFrame();
+    }
+
+} // namespace MGITest

MobileGL/MG_IntegrationTest/Scenarios/SpirvShaderBinaryScenario.cpp

@@ -0,0 +1,339 @@
+// MobileGL - MobileGL/MG_IntegrationTest/Scenarios/SpirvShaderBinaryScenario.cpp
+// Copyright (c) 2025-2026 MobileGL-Dev
+// Licensed under the GNU Lesser General Public License v3.0:
+//   https://www.gnu.org/licenses/gpl-3.0.txt
+//   https://www.gnu.org/licenses/lgpl-3.0.txt
+// SPDX-License-Identifier: LGPL-3.0-only
+// End of Source File Header
+//
+// Scenario - AN APPLICATION-SUPPLIED SPIR-V MODULE RENDERS, END TO END.
+//
+// GL_ARB_gl_spirv is core in 4.6 and MobileGL advertises a 4.6 context, but glShaderBinary and
+// glSpecializeShader were DECLARE_GL_FUNCTION_STUB entry points: they took their arguments,
+// recorded no error and did nothing, and glGetShaderiv(GL_SPIR_V_BINARY) raised GL_INVALID_ENUM.
+// Every gl_spirv conformance body died on the first of those two calls.
+//
+// This scenario is the end-to-end proof that the path now WORKS rather than merely answers: two
+// modules that glslang compiled ahead of time (embedded below as words, so the test depends on
+// no toolchain at run time), handed to glShaderBinary, specialized with a scale and a channel
+// index, linked, drawn, and read back. It runs on both backends and, in CI, on llvmpipe/lavapipe.
+//
+// The two specialization constants are the load-bearing part. The vertex module scales its
+// position by constant id 3 and the fragment module writes 1.0 into the channel named by constant
+// id 7 - so a specialization that silently did nothing would leave the default scale of 1.0 (a
+// full-viewport quad instead of a quarter-sized one) and the default channel 0 (red instead of
+// green), and BOTH would show up in the readback. A "specialization" that merely stored the
+// values without folding them in is exactly the failure mode this shape is built to catch.
+//
+// The GLSL the modules came from:
+//     vertex:   layout(location = 0) in vec2 aPos;
+//               layout(constant_id = 3) const float uScale = 1.0;
+//               void main() { gl_Position = vec4(aPos * uScale, 0.0, 1.0); }
+//     fragment: layout(location = 0) out vec4 oColor;
+//               layout(constant_id = 7) const int uChannel = 0;
+//               void main() { vec4 c = vec4(0,0,0,1); c[uChannel] = 1.0; oColor = c; }
+// compiled with `glslangValidator -G --target-env opengl`.
+
+#include <cstring>
+#include <string>
+#include <vector>
+
+#include "../Harness/HeadlessGL.h"
+#include "../Harness/ScenarioFixture.h"
+
+#ifdef GLAPI
+#undef GLAPI
+#endif
+#define GL_GLEXT_PROTOTYPES
+#include <GL/gl.h>
+#include <GL/glcorearb.h>
+#undef GL_GLEXT_PROTOTYPES
+
+#ifndef GL_SHADER_BINARY_FORMAT_SPIR_V
+#define GL_SHADER_BINARY_FORMAT_SPIR_V 0x9551
+#endif
+#ifndef GL_SPIR_V_BINARY
+#define GL_SPIR_V_BINARY 0x9552
+#endif
+
+namespace MGITest {
+    namespace {
+
+        class SpirvShaderBinaryScenario : public ScenarioTest {};
+
+        // 255 words
+        const unsigned int kVertexModule[] = {
+            0x07230203u, 0x00010000u, 0x0008000bu, 0x00000020u, 0x00000000u, 0x00020011u, 0x00000001u, 0x0006000bu,
+            0x00000001u, 0x4c534c47u, 0x6474732eu, 0x3035342eu, 0x00000000u, 0x0003000eu, 0x00000000u, 0x00000001u,
+            0x0009000fu, 0x00000000u, 0x00000004u, 0x6e69616du, 0x00000000u, 0x0000000du, 0x00000012u, 0x0000001eu,
+            0x0000001fu, 0x00030003u, 0x00000002u, 0x000001c2u, 0x00040005u, 0x00000004u, 0x6e69616du, 0x00000000u,
+            0x00060005u, 0x0000000bu, 0x505f6c67u, 0x65567265u, 0x78657472u, 0x00000000u, 0x00060006u, 0x0000000bu,
+            0x00000000u, 0x505f6c67u, 0x7469736fu, 0x006e6f69u, 0x00070006u, 0x0000000bu, 0x00000001u, 0x505f6c67u,
+            0x746e696fu, 0x657a6953u, 0x00000000u, 0x00070006u, 0x0000000bu, 0x00000002u, 0x435f6c67u, 0x4470696cu,
+            0x61747369u, 0x0065636eu, 0x00070006u, 0x0000000bu, 0x00000003u, 0x435f6c67u, 0x446c6c75u, 0x61747369u,
+            0x0065636eu, 0x00030005u, 0x0000000du, 0x00000000u, 0x00040005u, 0x00000012u, 0x736f5061u, 0x00000000u,
+            0x00040005u, 0x00000014u, 0x61635375u, 0x0000656cu, 0x00050005u, 0x0000001eu, 0x565f6c67u, 0x65747265u,
+            0x00444978u, 0x00060005u, 0x0000001fu, 0x495f6c67u, 0x6174736eu, 0x4965636eu, 0x00000044u, 0x00030047u,
+            0x0000000bu, 0x00000002u, 0x00050048u, 0x0000000bu, 0x00000000u, 0x0000000bu, 0x00000000u, 0x00050048u,
+            0x0000000bu, 0x00000001u, 0x0000000bu, 0x00000001u, 0x00050048u, 0x0000000bu, 0x00000002u, 0x0000000bu,
+            0x00000003u, 0x00050048u, 0x0000000bu, 0x00000003u, 0x0000000bu, 0x00000004u, 0x00040047u, 0x00000012u,
+            0x0000001eu, 0x00000000u, 0x00040047u, 0x00000014u, 0x00000001u, 0x00000003u, 0x00040047u, 0x0000001eu,
+            0x0000000bu, 0x00000005u, 0x00040047u, 0x0000001fu, 0x0000000bu, 0x00000006u, 0x00020013u, 0x00000002u,
+            0x00030021u, 0x00000003u, 0x00000002u, 0x00030016u, 0x00000006u, 0x00000020u, 0x00040017u, 0x00000007u,
+            0x00000006u, 0x00000004u, 0x00040015u, 0x00000008u, 0x00000020u, 0x00000000u, 0x0004002bu, 0x00000008u,
+            0x00000009u, 0x00000001u, 0x0004001cu, 0x0000000au, 0x00000006u, 0x00000009u, 0x0006001eu, 0x0000000bu,
+            0x00000007u, 0x00000006u, 0x0000000au, 0x0000000au, 0x00040020u, 0x0000000cu, 0x00000003u, 0x0000000bu,
+            0x0004003bu, 0x0000000cu, 0x0000000du, 0x00000003u, 0x00040015u, 0x0000000eu, 0x00000020u, 0x00000001u,
+            0x0004002bu, 0x0000000eu, 0x0000000fu, 0x00000000u, 0x00040017u, 0x00000010u, 0x00000006u, 0x00000002u,
+            0x00040020u, 0x00000011u, 0x00000001u, 0x00000010u, 0x0004003bu, 0x00000011u, 0x00000012u, 0x00000001u,
+            0x00040032u, 0x00000006u, 0x00000014u, 0x3f800000u, 0x0004002bu, 0x00000006u, 0x00000016u, 0x00000000u,
+            0x0004002bu, 0x00000006u, 0x00000017u, 0x3f800000u, 0x00040020u, 0x0000001bu, 0x00000003u, 0x00000007u,
+            0x00040020u, 0x0000001du, 0x00000001u, 0x0000000eu, 0x0004003bu, 0x0000001du, 0x0000001eu, 0x00000001u,
+            0x0004003bu, 0x0000001du, 0x0000001fu, 0x00000001u, 0x00050036u, 0x00000002u, 0x00000004u, 0x00000000u,
+            0x00000003u, 0x000200f8u, 0x00000005u, 0x0004003du, 0x00000010u, 0x00000013u, 0x00000012u, 0x0005008eu,
+            0x00000010u, 0x00000015u, 0x00000013u, 0x00000014u, 0x00050051u, 0x00000006u, 0x00000018u, 0x00000015u,
+            0x00000000u, 0x00050051u, 0x00000006u, 0x00000019u, 0x00000015u, 0x00000001u, 0x00070050u, 0x00000007u,
+            0x0000001au, 0x00000018u, 0x00000019u, 0x00000016u, 0x00000017u, 0x00050041u, 0x0000001bu, 0x0000001cu,
+            0x0000000du, 0x0000000fu, 0x0003003eu, 0x0000001cu, 0x0000001au, 0x000100fdu, 0x00010038u,
+        };
+
+        // 134 words
+        const unsigned int kFragmentModule[] = {
+            0x07230203u, 0x00010000u, 0x0008000bu, 0x00000014u, 0x00000000u, 0x00020011u, 0x00000001u, 0x0006000bu,
+            0x00000001u, 0x4c534c47u, 0x6474732eu, 0x3035342eu, 0x00000000u, 0x0003000eu, 0x00000000u, 0x00000001u,
+            0x0006000fu, 0x00000004u, 0x00000004u, 0x6e69616du, 0x00000000u, 0x00000012u, 0x00030010u, 0x00000004u,
+            0x00000008u, 0x00030003u, 0x00000002u, 0x000001c2u, 0x00040005u, 0x00000004u, 0x6e69616du, 0x00000000u,
+            0x00030005u, 0x00000009u, 0x00000063u, 0x00050005u, 0x0000000eu, 0x61684375u, 0x6c656e6eu, 0x00000000u,
+            0x00040005u, 0x00000012u, 0x6c6f436fu, 0x0000726fu, 0x00040047u, 0x0000000eu, 0x00000001u, 0x00000007u,
+            0x00040047u, 0x00000012u, 0x0000001eu, 0x00000000u, 0x00020013u, 0x00000002u, 0x00030021u, 0x00000003u,
+            0x00000002u, 0x00030016u, 0x00000006u, 0x00000020u, 0x00040017u, 0x00000007u, 0x00000006u, 0x00000004u,
+            0x00040020u, 0x00000008u, 0x00000007u, 0x00000007u, 0x0004002bu, 0x00000006u, 0x0000000au, 0x00000000u,
+            0x0004002bu, 0x00000006u, 0x0000000bu, 0x3f800000u, 0x0007002cu, 0x00000007u, 0x0000000cu, 0x0000000au,
+            0x0000000au, 0x0000000au, 0x0000000bu, 0x00040015u, 0x0000000du, 0x00000020u, 0x00000001u, 0x00040032u,
+            0x0000000du, 0x0000000eu, 0x00000000u, 0x00040020u, 0x0000000fu, 0x00000007u, 0x00000006u, 0x00040020u,
+            0x00000011u, 0x00000003u, 0x00000007u, 0x0004003bu, 0x00000011u, 0x00000012u, 0x00000003u, 0x00050036u,
+            0x00000002u, 0x00000004u, 0x00000000u, 0x00000003u, 0x000200f8u, 0x00000005u, 0x0004003bu, 0x00000008u,
+            0x00000009u, 0x00000007u, 0x0003003eu, 0x00000009u, 0x0000000cu, 0x00050041u, 0x0000000fu, 0x00000010u,
+            0x00000009u, 0x0000000eu, 0x0003003eu, 0x00000010u, 0x0000000bu, 0x0004003du, 0x00000007u, 0x00000013u,
+            0x00000009u, 0x0003003eu, 0x00000012u, 0x00000013u, 0x000100fdu, 0x00010038u,
+        };
+
+
+        // The quad the vertex module transforms. Full-viewport before the scale, so a scale of
+        // 0.5 covers exactly the middle half of each axis and the corners stay background.
+        const float kQuad[] = {-1.0f, -1.0f, 1.0f, -1.0f, -1.0f, 1.0f, 1.0f, 1.0f};
+
+        // The specialization constant ids the two modules declare.
+        constexpr unsigned int kScaleConstantId = 3;
+        constexpr unsigned int kChannelConstantId = 7;
+
+        unsigned int MakeSpirvShader(GLenum type, const unsigned int* words, size_t wordCount,
+                                     unsigned int constantId, unsigned int constantValue, std::string* outLog) {
+            const GLuint shader = glCreateShader(type);
+            glShaderBinary(1, &shader, GL_SHADER_BINARY_FORMAT_SPIR_V, words,
+                           static_cast<GLsizei>(wordCount * sizeof(unsigned int)));
+            if (glGetError() != GL_NO_ERROR) {
+                if (outLog) *outLog = "glShaderBinary rejected the module";
+                glDeleteShader(shader);
+                return 0;
+            }
+
+            GLint isSpirv = GL_FALSE;
+            glGetShaderiv(shader, GL_SPIR_V_BINARY, &isSpirv);
+            if (glGetError() != GL_NO_ERROR || isSpirv != GL_TRUE) {
+                if (outLog) *outLog = "GL_SPIR_V_BINARY did not read TRUE after glShaderBinary";
+                glDeleteShader(shader);
+                return 0;
+            }
+
+            glSpecializeShader(shader, "main", 1, &constantId, &constantValue);
+            GLint compiled = GL_FALSE;
+            glGetShaderiv(shader, GL_COMPILE_STATUS, &compiled);
+            if (compiled != GL_TRUE) {
+                if (outLog) {
+                    GLint length = 0;
+                    glGetShaderiv(shader, GL_INFO_LOG_LENGTH, &length);
+                    std::vector<char> log(static_cast<size_t>(length > 0 ? length : 1), '\0');
+                    glGetShaderInfoLog(shader, static_cast<GLsizei>(log.size()), nullptr, log.data());
+                    *outLog = std::string(log.data());
+                }
+                glDeleteShader(shader);
+                return 0;
+            }
+            return shader;
+        }
+
+    } // namespace
+
+    TEST_F(SpirvShaderBinaryScenario, ShaderBinaryFormatIsAdvertisedExactlyOnce) {
+        if (!Ready()) return;
+
+        GLint formatCount = -1;
+        glGetIntegerv(GL_NUM_SHADER_BINARY_FORMATS, &formatCount);
+        EXPECT_EQ(FirstGLError(), 0u);
+        ASSERT_EQ(formatCount, 1) << "a 4.6 context supports exactly the SPIR-V shader binary format";
+
+        std::vector<GLint> formats(static_cast<size_t>(formatCount), 0);
+        glGetIntegerv(GL_SHADER_BINARY_FORMATS, formats.data());
+        EXPECT_EQ(FirstGLError(), 0u);
+        EXPECT_EQ(formats[0], static_cast<GLint>(GL_SHADER_BINARY_FORMAT_SPIR_V))
+            << "the count and the list have to describe the same thing";
+    }
+
+    TEST_F(SpirvShaderBinaryScenario, AnUnsupportedBinaryFormatIsRejectedInsteadOfSilentlyAccepted) {
+        if (!Ready()) return;
+
+        const GLuint shader = glCreateShader(GL_VERTEX_SHADER);
+        // 0x8DF9 is GL_SHADER_BINARY_FORMATS' neighbour, not a format: any value but
+        // GL_SHADER_BINARY_FORMAT_SPIR_V is GL_INVALID_ENUM. The stub used to return silently.
+        glShaderBinary(1, &shader, 0x8DF9, kVertexModule, sizeof(kVertexModule));
+        EXPECT_EQ(FirstGLError(), static_cast<unsigned int>(GL_INVALID_ENUM));
+
+        GLint isSpirv = GL_TRUE;
+        glGetShaderiv(shader, GL_SPIR_V_BINARY, &isSpirv);
+        EXPECT_EQ(FirstGLError(), 0u);
+        EXPECT_EQ(isSpirv, GL_FALSE) << "a rejected glShaderBinary must not have attached anything";
+
+        glDeleteShader(shader);
+    }
+
+    TEST_F(SpirvShaderBinaryScenario, CompileShaderOnASpirvShaderIsInvalidOperationAndShaderSourceTakesItBack) {
+        if (!Ready()) return;
+
+        const GLuint shader = glCreateShader(GL_VERTEX_SHADER);
+        glShaderBinary(1, &shader, GL_SHADER_BINARY_FORMAT_SPIR_V, kVertexModule, sizeof(kVertexModule));
+        ASSERT_EQ(FirstGLError(), 0u);
+
+        glCompileShader(shader);
+        EXPECT_EQ(FirstGLError(), static_cast<unsigned int>(GL_INVALID_OPERATION))
+            << "glSpecializeShader, not glCompileShader, is what compiles a SPIR-V shader";
+
+        // glShaderSource takes the object back to being a GLSL shader, and GL_SPIR_V_BINARY with
+        // it - the transition the conformance suite checks explicitly.
+        const char* source = "#version 450\nvoid main() { gl_Position = vec4(0.0); }\n";
+        glShaderSource(shader, 1, &source, nullptr);
+        ASSERT_EQ(FirstGLError(), 0u);
+        GLint isSpirv = GL_TRUE;
+        glGetShaderiv(shader, GL_SPIR_V_BINARY, &isSpirv);
+        EXPECT_EQ(isSpirv, GL_FALSE);
+        glCompileShader(shader);
+        EXPECT_EQ(FirstGLError(), 0u) << "the object is an ordinary GLSL shader again";
+
+        glDeleteShader(shader);
+    }
+
+    TEST_F(SpirvShaderBinaryScenario, SpecializeShaderErrorSurfaceMatchesTheExtension) {
+        if (!Ready()) return;
+
+        const GLuint shader = glCreateShader(GL_VERTEX_SHADER);
+        glShaderBinary(1, &shader, GL_SHADER_BINARY_FORMAT_SPIR_V, kVertexModule, sizeof(kVertexModule));
+        ASSERT_EQ(FirstGLError(), 0u);
+
+        // 4242 is not one of the module's constant ids. ARB_gl_spirv enumerates that as
+        // GL_INVALID_VALUE, and an erroring GL command has no other effect - so the shader is left
+        // untouched rather than pushed into a failed-compile state.
+        const unsigned int badId = 4242;
+        const unsigned int value = 0;
+        glSpecializeShader(shader, "main", 1, &badId, &value);
+        EXPECT_EQ(FirstGLError(), static_cast<unsigned int>(GL_INVALID_VALUE));
+
+        // Same for an entry point the module does not carry.
+        glSpecializeShader(shader, "notMain", 0, nullptr, nullptr);
+        EXPECT_EQ(FirstGLError(), static_cast<unsigned int>(GL_INVALID_VALUE));
+
+        // Neither refusal specialized the shader, so a well-formed call still works.
+        glSpecializeShader(shader, "main", 0, nullptr, nullptr);
+        EXPECT_EQ(FirstGLError(), 0u);
+        GLint compiled = GL_FALSE;
+        glGetShaderiv(shader, GL_COMPILE_STATUS, &compiled);
+        EXPECT_EQ(compiled, GL_TRUE);
+
+        // But a SECOND specialization of a shader that HAS been specialized is INVALID_OPERATION
+        // until glShaderBinary re-associates the module.
+        glSpecializeShader(shader, "main", 0, nullptr, nullptr);
+        EXPECT_EQ(FirstGLError(), static_cast<unsigned int>(GL_INVALID_OPERATION));
+        glShaderBinary(1, &shader, GL_SHADER_BINARY_FORMAT_SPIR_V, kVertexModule, sizeof(kVertexModule));
+        glSpecializeShader(shader, "main", 0, nullptr, nullptr);
+        EXPECT_EQ(FirstGLError(), 0u) << "re-associating the module makes specialization legal again";
+
+        glDeleteShader(shader);
+    }
+
+    TEST_F(SpirvShaderBinaryScenario, SpecializedModulesLinkAndRenderWithTheirConstantsApplied) {
+        if (!Ready()) return;
+        HeadlessGL& gl = Gl();
+        const int width = gl.Width();
+        const int height = gl.Height();
+        ASSERT_GE(width, 16);
+        ASSERT_GE(height, 16);
+
+        std::string log;
+        // Scale 0.5 as a float, handed over as the GLuint bit pattern the extension specifies.
+        unsigned int halfBits = 0;
+        const float half = 0.5f;
+        std::memcpy(&halfBits, &half, sizeof(halfBits));
+
+        const unsigned int vs = MakeSpirvShader(GL_VERTEX_SHADER, kVertexModule,
+                                                sizeof(kVertexModule) / sizeof(kVertexModule[0]),
+                                                kScaleConstantId, halfBits, &log);
+        ASSERT_NE(vs, 0u) << "vertex: " << log;
+        // Channel 1 is green; the module's own default is 0 (red), so a specialization that did
+        // nothing paints the wrong colour.
+        const unsigned int fs = MakeSpirvShader(GL_FRAGMENT_SHADER, kFragmentModule,
+                                                sizeof(kFragmentModule) / sizeof(kFragmentModule[0]),
+                                                kChannelConstantId, 1u, &log);
+        ASSERT_NE(fs, 0u) << "fragment: " << log;
+
+        const GLuint program = glCreateProgram();
+        glAttachShader(program, vs);
+        glAttachShader(program, fs);
+        glLinkProgram(program);
+        GLint linked = GL_FALSE;
+        glGetProgramiv(program, GL_LINK_STATUS, &linked);
+        if (linked != GL_TRUE) {
+            GLint length = 0;
+            glGetProgramiv(program, GL_INFO_LOG_LENGTH, &length);
+            std::vector<char> programLog(static_cast<size_t>(length > 0 ? length : 1), '\0');
+            glGetProgramInfoLog(program, static_cast<GLsizei>(programLog.size()), nullptr, programLog.data());
+            FAIL() << "linking two specialized SPIR-V modules failed: " << programLog.data();
+        }
+
+        BindDefaultFramebuffer();
+        glViewport(0, 0, width, height);
+        glDisable(GL_SCISSOR_TEST);
+        glDisable(GL_DEPTH_TEST);
+        ClearTo(0.0f, 0.0f, 0.0f, 1.0f);
+
+        GLuint vao = 0, vbo = 0;
+        glGenVertexArrays(1, &vao);
+        glBindVertexArray(vao);
+        glGenBuffers(1, &vbo);
+        glBindBuffer(GL_ARRAY_BUFFER, vbo);
+        glBufferData(GL_ARRAY_BUFFER, sizeof(kQuad), kQuad, GL_STATIC_DRAW);
+        glEnableVertexAttribArray(0);
+        glVertexAttribPointer(0, 2, GL_FLOAT, GL_FALSE, 2 * sizeof(float), nullptr);
+        glUseProgram(program);
+        glDrawArrays(GL_TRIANGLE_STRIP, 0, 4);
+        EXPECT_EQ(FirstGLError(), 0u);
+
+        const Image painted = ReadPixels(width, height);
+        const Rgba8 centre = painted.At(width / 2, height / 2);
+        EXPECT_LT(centre.r, 32) << "the fragment module wrote the wrong channel; constant id 7 was not applied";
+        EXPECT_GT(centre.g, 224) << "the centre of a 0.5-scaled quad must be painted";
+
+        // A pixel just inside the corner is OUTSIDE the 0.5-scaled quad and must still be the
+        // clear colour - which is what proves constant id 3 reached the vertex module. At the
+        // default scale of 1.0 the quad covers the whole viewport and this pixel would be green.
+        const Rgba8 corner = painted.At(1, 1);
+        EXPECT_LT(corner.g, 32) << "the quad was not scaled; the vertex specialization constant was not applied";
+
+        glBindVertexArray(0);
+        glDeleteBuffers(1, &vbo);
+        glDeleteVertexArrays(1, &vao);
+        glDeleteProgram(program);
+        glDeleteShader(vs);
+        glDeleteShader(fs);
+        gl.EndFrame();
+    }
+
+} // namespace MGITest

MobileGL/MG_IntegrationTest/Scenarios/TextureViewScenario.cpp

@@ -0,0 +1,908 @@
+// MobileGL - MobileGL/MG_IntegrationTest/Scenarios/TextureViewScenario.cpp
+// Copyright (c) 2025-2026 MobileGL-Dev
+// Licensed under the GNU Lesser General Public License v3.0:
+//   https://www.gnu.org/licenses/gpl-3.0.txt
+//   https://www.gnu.org/licenses/lgpl-3.0.txt
+// SPDX-License-Identifier: LGPL-3.0-only
+// End of Source File Header
+//
+// glTextureView (ARB_texture_view / GL 4.6 core 8.18) end to end on both backends.
+//
+// THE DEFECT. glTextureView was a stub that logged once and returned. That is worse than not
+// having the function: MobileGL advertises GL 4.6, so LWJGL resolves a non-null pointer, an
+// application's capability check passes, it takes the texture-view path, and the view texture it
+// then samples has no storage at all. Nothing errors; the picture is simply wrong. The Better
+// Clouds Minecraft mod is exactly this shape - its GLCompat gates `supportsTextureView` on
+// `caps.glTextureView != NULL`, which was already true, so it ran its FULL path against a view
+// that aliased nothing.
+//
+// WHAT A VIEW IS, and why a copy cannot stand in for one. A view is a second texture NAME over
+// the SAME storage. Two consequences the tests below pin, both of which a copy fails:
+//   * writes through either name are visible through the other (CoherencyIsBidirectional), and
+//   * the two names carry INDEPENDENT per-texture parameters at the same time - which is the
+//     entire point for Better Clouds: one D24S8 image, sampled in ONE shading pass through the
+//     parent with DEPTH_STENCIL_TEXTURE_MODE = GL_STENCIL_INDEX and through the view with
+//     GL_DEPTH_COMPONENT (BetterCloudsCoveragePipeline below).
+//
+// MECHANISM PER BACKEND. DirectVulkan: the view resolves to the storage texture's ONE
+// TextureResource - one VkImage, one tracked layout, one upload path - and its own VkImageViews
+// (sub-range, reinterpreted VkFormat, its own aspect) are cached in alternateSampledViews /
+// attachmentViews keyed by the whole window. DirectGLES: the view gets its own ES name minted by
+// EXT/OES_texture_view over the storage texture's name, so the driver supplies the aliasing and
+// per-name parameters come for free. Without that extension the frontend refuses glTextureView
+// with GL_INVALID_OPERATION and withholds GL_ARB_texture_view rather than emulate by copying -
+// see NoExtensionSupportIsRefusedRatherThanFaked.
+//
+// CONTROLS. Every case here would pass on a stub for at least one of its assertions, so each one
+// also asserts something the stub cannot produce: a non-zero sampled value, a DIFFERENT value
+// through the two names, or a value that changed after a write through the other name.
+
+#include <array>
+#include <cstdint>
+#include <string>
+#include <vector>
+
+#include "../Harness/HeadlessGL.h"
+#include "../Harness/ScenarioFixture.h"
+
+#ifdef GLAPI
+#undef GLAPI
+#endif
+#define GL_GLEXT_PROTOTYPES
+#include <GL/gl.h>
+#include <GL/glcorearb.h>
+#undef GL_GLEXT_PROTOTYPES
+
+namespace MGITest {
+    namespace {
+        constexpr int kSize = 64;
+        // The lower strip no cloud quad covers, so coverage 0 / depth 0 is asserted too - a
+        // uniform image would otherwise pass a test that only ever looked at covered texels.
+        constexpr int kUncoveredTop = 16;
+
+        constexpr const char* kQuadVertexSource = R"(#version 330 core
+in vec2 aPos;
+uniform vec4 uRect;   // x0, y0, x1, y1 in NDC
+uniform float uDepth; // NDC z
+void main() {
+    vec2 p = mix(uRect.xy, uRect.zw, aPos);
+    gl_Position = vec4(p, uDepth, 1.0);
+}
+)";
+
+        // Mirrors betterclouds_coverage.fsh's shape: a second fragment output at location 1 whose
+        // draw buffer is GL_NONE. The mod declares and writes it while glDrawBuffers names only
+        // COLOR_ATTACHMENT0, so a layer that mishandles a write to a NONE draw buffer would either
+        // error or clobber attachment 0.
+        constexpr const char* kCoverageFragmentSource = R"(#version 330 core
+layout (location = 0) out vec4 outColor;
+layout (location = 1) out float outUnused;
+void main() {
+    outColor = vec4(1.0, 0.0, 0.0, 1.0);
+    outUnused = 1.0 / 255.0;
+}
+)";
+
+        // The Better Clouds shading pass, reduced to its sampling. Both fetches name the SAME
+        // D24S8 image through two GL texture names bound to two units in this one invocation.
+        // `ivec2(gl_FragCoord)` (a truncating vec4 -> ivec2 constructor) is the mod's own spelling
+        // at betterclouds_shading.fsh:56, kept verbatim because a strict GLSL front end can reject
+        // it; the depth fetch uses the conventional `.xy` form the mod uses at line 117.
+        constexpr const char* kShadingFragmentSource = R"(#version 330 core
+uniform usampler2D uCoverage;  // the PARENT, DEPTH_STENCIL_TEXTURE_MODE = GL_STENCIL_INDEX
+uniform sampler2D uDepthView;  // the VIEW,   DEPTH_STENCIL_TEXTURE_MODE = GL_DEPTH_COMPONENT
+out vec4 outColor;
+void main() {
+    uint coverage = texelFetch(uCoverage, ivec2(gl_FragCoord), 0).r;
+    float depth = texelFetch(uDepthView, ivec2(gl_FragCoord.xy), 0).r;
+    outColor = vec4(float(coverage) * 0.25, depth, 0.0, 1.0);
+    gl_FragDepth = depth;
+}
+)";
+
+        // Reads a reinterpreting view (GL_R32UI over GL_RGBA8 storage - both VIEW_CLASS_32_BITS)
+        // and unpacks the word back into the four bytes it was written as.
+        constexpr const char* kDecodeWordFragmentSource = R"(#version 330 core
+uniform usampler2D uWords;
+out vec4 outColor;
+void main() {
+    uint word = texelFetch(uWords, ivec2(gl_FragCoord.xy), 0).r;
+    outColor = vec4(float((word      ) & 0xFFu) / 255.0,
+                    float((word >>  8) & 0xFFu) / 255.0,
+                    float((word >> 16) & 0xFFu) / 255.0,
+                    float((word >> 24) & 0xFFu) / 255.0);
+}
+)";
+
+        constexpr const char* kSampleFragmentSource = R"(#version 330 core
+uniform sampler2D uTexture;
+uniform float uLod;
+out vec4 outColor;
+void main() {
+    outColor = textureLod(uTexture, gl_FragCoord.xy / 64.0, uLod);
+}
+)";
+
+        std::string Describe(const Rgba8& c) {
+            return "rgba(" + std::to_string(c.r) + "," + std::to_string(c.g) + "," + std::to_string(c.b) + "," +
+                   std::to_string(c.a) + ")";
+        }
+
+        class TextureViewScenario : public ScenarioTest {
+        protected:
+            void SetUp() override {
+                ScenarioTest::SetUp();
+                if (!Ready()) return;
+                if (!TextureViewUsable()) {
+                    GTEST_SKIP() << "glTextureView is unavailable on backend " << Gl().BackendName()
+                                 << " (GL_ARB_texture_view not advertised)";
+                }
+            }
+
+            void TearDown() override {
+                if (!Ready()) return;
+                for (const GLuint texture : m_textures) {
+                    glDeleteTextures(1, &texture);
+                }
+                m_textures.clear();
+                for (const GLuint fbo : m_fbos) {
+                    glBindFramebuffer(GL_FRAMEBUFFER, 0);
+                    glDeleteFramebuffers(1, &fbo);
+                }
+                m_fbos.clear();
+                for (const GLuint rbo : m_rbos) {
+                    glDeleteRenderbuffers(1, &rbo);
+                }
+                m_rbos.clear();
+                for (const GLuint program : m_programs) {
+                    glDeleteProgram(program);
+                }
+                m_programs.clear();
+                if (m_vao != 0) {
+                    glBindVertexArray(0);
+                    glDeleteVertexArrays(1, &m_vao);
+                    m_vao = 0;
+                }
+                if (m_vbo != 0) {
+                    glDeleteBuffers(1, &m_vbo);
+                    m_vbo = 0;
+                }
+            }
+
+            // A trivial same-format full-range view. It exercises nothing the cases below test,
+            // so a backend that simply does not have the feature skips instead of failing every
+            // one of them - the same shape CopyImageLayeredScenario uses for glCopyImageSubData.
+            bool TextureViewUsable() {
+                GLuint storage = 0;
+                glGenTextures(1, &storage);
+                glBindTexture(GL_TEXTURE_2D, storage);
+                glTexStorage2D(GL_TEXTURE_2D, 1, GL_RGBA8, 1, 1);
+                glBindTexture(GL_TEXTURE_2D, 0);
+                GLuint view = 0;
+                glGenTextures(1, &view);
+                while (glGetError() != GL_NO_ERROR) {
+                }
+                glTextureView(view, GL_TEXTURE_2D, storage, GL_RGBA8, 0, 1, 0, 1);
+                const bool usable = glGetError() == GL_NO_ERROR;
+                glDeleteTextures(1, &view);
+                glDeleteTextures(1, &storage);
+                return usable;
+            }
+
+            GLuint MakeVao() {
+                if (m_vao != 0) return m_vao;
+                // A unit quad; the vertex shader maps it onto whatever NDC rect uRect names, so
+                // one buffer serves every draw here.
+                static constexpr float kQuad[] = {0.0f, 0.0f, 1.0f, 0.0f, 0.0f, 1.0f,
+                                                  1.0f, 0.0f, 1.0f, 1.0f, 0.0f, 1.0f};
+                glGenVertexArrays(1, &m_vao);
+                glBindVertexArray(m_vao);
+                glGenBuffers(1, &m_vbo);
+                glBindBuffer(GL_ARRAY_BUFFER, m_vbo);
+                glBufferData(GL_ARRAY_BUFFER, sizeof(kQuad), kQuad, GL_STATIC_DRAW);
+                glEnableVertexAttribArray(0);
+                glVertexAttribPointer(0, 2, GL_FLOAT, GL_FALSE, 2 * sizeof(float), nullptr);
+                return m_vao;
+            }
+
+            GLuint MakeProgram(const char* vertexSource, const char* fragmentSource) {
+                std::string error;
+                const GLuint program = CompileProgram(vertexSource, fragmentSource, &error);
+                EXPECT_NE(program, 0u) << "program failed to build: " << error;
+                if (program != 0) m_programs.push_back(program);
+                return program;
+            }
+
+            GLuint MakeTexture() {
+                GLuint texture = 0;
+                glGenTextures(1, &texture);
+                m_textures.push_back(texture);
+                return texture;
+            }
+
+            GLuint MakeFbo() {
+                GLuint fbo = 0;
+                glGenFramebuffers(1, &fbo);
+                m_fbos.push_back(fbo);
+                return fbo;
+            }
+
+            // A 2D texture with immutable storage and NEAREST filtering, i.e. what every case
+            // here views. Levels beyond 1 stay undefined until a caller fills them.
+            GLuint MakeImmutable2D(GLenum internalFormat, int levels, int width, int height) {
+                const GLuint texture = MakeTexture();
+                glBindTexture(GL_TEXTURE_2D, texture);
+                glTexStorage2D(GL_TEXTURE_2D, levels, internalFormat, width, height);
+                glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_NEAREST_MIPMAP_NEAREST);
+                glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
+                glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
+                glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
+                return texture;
+            }
+
+            void DrawQuad(GLuint program, float x0, float y0, float x1, float y1, float depth) {
+                glUseProgram(program);
+                glUniform4f(glGetUniformLocation(program, "uRect"), x0, y0, x1, y1);
+                const GLint depthLocation = glGetUniformLocation(program, "uDepth");
+                if (depthLocation >= 0) glUniform1f(depthLocation, depth);
+                glBindVertexArray(MakeVao());
+                glDrawArrays(GL_TRIANGLES, 0, 6);
+            }
+
+            // Reads the colour texture currently attached to `fbo` as COLOR_ATTACHMENT0.
+            Image ReadFbo(GLuint fbo, int width, int height) {
+                glBindFramebuffer(GL_FRAMEBUFFER, fbo);
+                glReadBuffer(GL_COLOR_ATTACHMENT0);
+                return ReadPixels(width, height);
+            }
+
+            // Every pixel of the inclusive region must match `expected` within `tolerance` per
+            // channel. Whole-region rather than a spot check, for the reason HeadlessGL.h gives:
+            // three of four vertices carrying stale data still paints a correct centre pixel.
+            void ExpectRegion(const Image& image, int x0, int x1, int y0, int y1, Rgba8 expected, int tolerance,
+                              const char* what) {
+                int offenders = 0;
+                Rgba8 firstOffender{};
+                int firstX = -1;
+                int firstY = -1;
+                for (int y = y0; y <= y1; ++y) {
+                    for (int x = x0; x <= x1; ++x) {
+                        const Rgba8 actual = image.At(x, y);
+                        const bool ok = std::abs(int(actual.r) - int(expected.r)) <= tolerance &&
+                                        std::abs(int(actual.g) - int(expected.g)) <= tolerance &&
+                                        std::abs(int(actual.b) - int(expected.b)) <= tolerance &&
+                                        std::abs(int(actual.a) - int(expected.a)) <= tolerance;
+                        if (!ok) {
+                            if (offenders == 0) {
+                                firstOffender = actual;
+                                firstX = x;
+                                firstY = y;
+                            }
+                            ++offenders;
+                        }
+                    }
+                }
+                EXPECT_EQ(offenders, 0) << what << ": " << offenders << " of "
+                                        << (x1 - x0 + 1) * (y1 - y0 + 1) << " pixels disagree; first at (" << firstX
+                                        << ", " << firstY << ") is " << Describe(firstOffender) << ", expected "
+                                        << Describe(expected) << " +/- " << tolerance;
+            }
+
+            std::vector<GLuint> m_textures;
+            std::vector<GLuint> m_fbos;
+            std::vector<GLuint> m_rbos;
+            std::vector<GLuint> m_programs;
+            GLuint m_vao = 0;
+            GLuint m_vbo = 0;
+        };
+
+        // ------------------------------------------------------------------------------------
+        // The driving case: the Better Clouds full-mode pipeline, in its real order.
+        // ------------------------------------------------------------------------------------
+        TEST_F(TextureViewScenario, BetterCloudsCoveragePipeline) {
+            if (!Ready() || IsSkipped()) return;
+
+            // --- Resources.java:230-251, in order ---------------------------------------------
+            const GLuint coverageColor = MakeImmutable2D(GL_RGBA8, 1, kSize, kSize);
+            const GLuint coverage = MakeTexture();
+            glBindTexture(GL_TEXTURE_2D, coverage);
+            glTexStorage2D(GL_TEXTURE_2D, 1, GL_DEPTH24_STENCIL8, kSize, kSize);
+            glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
+            glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
+            glTexParameteri(GL_TEXTURE_2D, GL_DEPTH_STENCIL_TEXTURE_MODE, GL_STENCIL_INDEX);
+
+            const GLuint coverageFbo = MakeFbo();
+            glBindFramebuffer(GL_DRAW_FRAMEBUFFER, coverageFbo);
+            glFramebufferTexture2D(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, coverageColor, 0);
+            glFramebufferTexture2D(GL_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D, coverage, 0);
+            const GLenum drawBuffers[] = {GL_COLOR_ATTACHMENT0};
+            glDrawBuffers(1, drawBuffers);
+            ASSERT_EQ(glCheckFramebufferStatus(GL_FRAMEBUFFER), static_cast<GLenum>(GL_FRAMEBUFFER_COMPLETE))
+                << "the coverage framebuffer is incomplete; the mod would silently demote to its "
+                   "fallback configuration here (Resources.java:187-209)";
+
+            // The view is made from a name glGenTextures has only RESERVED - it has never been
+            // bound, so glTextureView has to instantiate the texture object itself.
+            const GLuint coverageDepthView = MakeTexture();
+            glTextureView(coverageDepthView, GL_TEXTURE_2D, coverage, GL_DEPTH24_STENCIL8, 0, 1, 0, 1);
+            ASSERT_EQ(glGetError(), static_cast<GLenum>(GL_NO_ERROR)) << "glTextureView raised an error";
+            glBindTexture(GL_TEXTURE_2D, coverageDepthView);
+            glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
+            glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
+            glTexParameteri(GL_TEXTURE_2D, GL_DEPTH_STENCIL_TEXTURE_MODE, GL_DEPTH_COMPONENT);
+            glBindTexture(GL_TEXTURE_2D, 0);
+            ASSERT_EQ(glGetError(), static_cast<GLenum>(GL_NO_ERROR)) << "setting up the view raised an error";
+
+            // The two names must be distinguishable through the queries, or nothing below proves
+            // which one produced a sample.
+            GLint parentMode = 0;
+            GLint viewMode = 0;
+            glBindTexture(GL_TEXTURE_2D, coverage);
+            glGetTexParameteriv(GL_TEXTURE_2D, GL_DEPTH_STENCIL_TEXTURE_MODE, &parentMode);
+            glBindTexture(GL_TEXTURE_2D, coverageDepthView);
+            glGetTexParameteriv(GL_TEXTURE_2D, GL_DEPTH_STENCIL_TEXTURE_MODE, &viewMode);
+            glBindTexture(GL_TEXTURE_2D, 0);
+            EXPECT_EQ(parentMode, GL_STENCIL_INDEX) << "the parent must keep the stencil aspect";
+            EXPECT_EQ(viewMode, GL_DEPTH_COMPONENT)
+                << "the view must carry its OWN depth-stencil mode; sharing one parameter set with "
+                   "the parent is precisely what a texture view exists to avoid";
+
+            // --- OpenGLRenderer.java:244-344, the coverage pass -------------------------------
+            const GLuint coverageProgram = MakeProgram(kQuadVertexSource, kCoverageFragmentSource);
+            ASSERT_NE(coverageProgram, 0u);
+
+            glBindFramebuffer(GL_DRAW_FRAMEBUFFER, coverageFbo);
+            glViewport(0, 0, kSize, kSize);
+            glEnable(GL_DEPTH_TEST);
+            glDepthMask(GL_TRUE);
+            glColorMask(GL_TRUE, GL_TRUE, GL_TRUE, GL_TRUE);
+            // Reverse-Z, as the mod runs it (OpenGLRenderer.java:236/241).
+            glClearDepth(0.0);
+            glDepthFunc(GL_GEQUAL);
+            glDisable(GL_BLEND);
+            glEnable(GL_STENCIL_TEST);
+            glStencilMask(0xff);
+            glClearStencil(0);
+            // The coverage COUNT: one increment per depth-passing cloud fragment.
+            glStencilOp(GL_KEEP, GL_INCR, GL_INCR);
+            glStencilFunc(GL_ALWAYS, 0xff, 0xff);
+            glClearColor(0.0f, 0.0f, 0.0f, 0.0f);
+            glClear(GL_STENCIL_BUFFER_BIT | GL_COLOR_BUFFER_BIT | GL_DEPTH_BUFFER_BIT);
+
+            // Quad 1 covers everything above the uncovered strip, at window depth 0.25.
+            const float stripTop = 2.0f * (float(kUncoveredTop) / float(kSize)) - 1.0f;
+            DrawQuad(coverageProgram, -1.0f, stripTop, 1.0f, 1.0f, -0.5f);
+            // Quad 2 covers the right half of that, at window depth 0.75 - nearer under GEQUAL,
+            // so it both passes the depth test and increments the stencil a second time.
+            DrawQuad(coverageProgram, 0.0f, stripTop, 1.0f, 1.0f, 0.5f);
+            ASSERT_EQ(glGetError(), static_cast<GLenum>(GL_NO_ERROR)) << "the coverage pass raised an error";
+
+            // --- OpenGLRenderer.java:393-464, the shading pass --------------------------------
+            // A different draw framebuffer, exactly as the mod does (it hands the frame back to
+            // Blaze3D before shading). The coverage texture stays ATTACHED to coverageFbo while
+            // being sampled here, which is the shape a lazy/deferred FBO binding gets wrong.
+            ColorFbo destination = MakeColorFbo(kSize, kSize);
+            ASSERT_NE(destination.fbo, 0u);
+            GLuint destinationDepth = 0;
+            glGenRenderbuffers(1, &destinationDepth);
+            m_rbos.push_back(destinationDepth);
+            glBindRenderbuffer(GL_RENDERBUFFER, destinationDepth);
+            glRenderbufferStorage(GL_RENDERBUFFER, GL_DEPTH24_STENCIL8, kSize, kSize);
+            glBindFramebuffer(GL_FRAMEBUFFER, destination.fbo);
+            glFramebufferRenderbuffer(GL_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_RENDERBUFFER, destinationDepth);
+            ASSERT_EQ(glCheckFramebufferStatus(GL_FRAMEBUFFER), static_cast<GLenum>(GL_FRAMEBUFFER_COMPLETE));
+
+            glViewport(0, 0, kSize, kSize);
+            glClearColor(0.0f, 0.0f, 0.0f, 1.0f);
+            glClearDepth(0.0);
+            glClear(GL_COLOR_BUFFER_BIT | GL_DEPTH_BUFFER_BIT);
+            glDepthFunc(GL_GEQUAL);
+            glDepthMask(GL_TRUE);
+            glEnable(GL_DEPTH_TEST);
+            glDisable(GL_STENCIL_TEST);
+            // The mod's own indexed/non-indexed colour-mask pair (OpenGLRenderer.java:411-412).
+            glColorMask(GL_FALSE, GL_FALSE, GL_FALSE, GL_FALSE);
+            glColorMaski(0, GL_TRUE, GL_TRUE, GL_TRUE, GL_TRUE);
+
+            const GLuint shadingProgram = MakeProgram(kQuadVertexSource, kShadingFragmentSource);
+            ASSERT_NE(shadingProgram, 0u);
+            glUseProgram(shadingProgram);
+            // Unit 1 = the view (depth aspect), unit 3 = the parent (stencil aspect), the mod's
+            // own unit assignment (Resources.java:309/311).
+            glActiveTexture(GL_TEXTURE1);
+            glBindTexture(GL_TEXTURE_2D, coverageDepthView);
+            glActiveTexture(GL_TEXTURE3);
+            glBindTexture(GL_TEXTURE_2D, coverage);
+            glUniform1i(glGetUniformLocation(shadingProgram, "uDepthView"), 1);
+            glUniform1i(glGetUniformLocation(shadingProgram, "uCoverage"), 3);
+            glActiveTexture(GL_TEXTURE0);
+
+            DrawQuad(shadingProgram, -1.0f, -1.0f, 1.0f, 1.0f, 0.0f);
+            ASSERT_EQ(glGetError(), static_cast<GLenum>(GL_NO_ERROR)) << "the shading pass raised an error";
+
+            const Image shaded = ReadFbo(destination.fbo, kSize, kSize);
+            glColorMask(GL_TRUE, GL_TRUE, GL_TRUE, GL_TRUE);
+
+            // R = coverage * 0.25 (so 1 -> 64, 2 -> 128), G = the depth read THROUGH THE VIEW.
+            // A stub view samples (0,0,0,1), which fails the green channel of both covered
+            // regions; a view that inherited the parent's stencil aspect fails them too.
+            constexpr int kTolerance = 3;
+            ExpectRegion(shaded, 1, kSize - 2, 1, kUncoveredTop - 2, Rgba8{0, 0, 0, 255}, kTolerance,
+                         "the uncovered strip must read coverage 0 and cleared depth 0");
+            ExpectRegion(shaded, 1, kSize / 2 - 2, kUncoveredTop + 1, kSize - 2, Rgba8{64, 64, 0, 255}, kTolerance,
+                         "one cloud quad: stencil 1 through the parent, window depth 0.25 through the view");
+            ExpectRegion(shaded, kSize / 2 + 1, kSize - 2, kUncoveredTop + 1, kSize - 2, Rgba8{128, 191, 0, 255},
+                         kTolerance,
+                         "two overlapping cloud quads: stencil 2 through the parent, window depth 0.75 "
+                         "through the view");
+
+            DestroyColorFbo(destination);
+        }
+
+        // ------------------------------------------------------------------------------------
+        // Storage sharing, in both directions. This is the assertion a copy-based emulation
+        // fails, and the reason the no-EXT path refuses rather than emulates.
+        // ------------------------------------------------------------------------------------
+        TEST_F(TextureViewScenario, CoherencyIsBidirectional) {
+            if (!Ready() || IsSkipped()) return;
+
+            const GLuint storage = MakeImmutable2D(GL_RGBA8, 1, kSize, kSize);
+            const GLuint view = MakeTexture();
+            glTextureView(view, GL_TEXTURE_2D, storage, GL_RGBA8, 0, 1, 0, 1);
+            ASSERT_EQ(glGetError(), static_cast<GLenum>(GL_NO_ERROR));
+
+            // Render red through the PARENT's name...
+            const GLuint fbo = MakeFbo();
+            glBindFramebuffer(GL_FRAMEBUFFER, fbo);
+            glFramebufferTexture2D(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, storage, 0);
+            ASSERT_EQ(glCheckFramebufferStatus(GL_FRAMEBUFFER), static_cast<GLenum>(GL_FRAMEBUFFER_COMPLETE));
+            glViewport(0, 0, kSize, kSize);
+            glDisable(GL_DEPTH_TEST);
+            glDisable(GL_STENCIL_TEST);
+            glClearColor(1.0f, 0.0f, 0.0f, 1.0f);
+            glClear(GL_COLOR_BUFFER_BIT);
+
+            // ...and read it back through the VIEW's.
+            const GLuint viewFbo = MakeFbo();
+            glBindFramebuffer(GL_FRAMEBUFFER, viewFbo);
+            glFramebufferTexture2D(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, view, 0);
+            ASSERT_EQ(glCheckFramebufferStatus(GL_FRAMEBUFFER), static_cast<GLenum>(GL_FRAMEBUFFER_COMPLETE))
+                << "a texture view must be attachable like any other texture";
+            Image throughView = ReadFbo(viewFbo, kSize, kSize);
+            ExpectRegion(throughView, 0, kSize - 1, 0, kSize - 1, Rgba8{255, 0, 0, 255}, 1,
+                         "a write through the parent must be visible through the view");
+
+            // Now the other direction: write green through the VIEW, read through the PARENT.
+            glBindFramebuffer(GL_FRAMEBUFFER, viewFbo);
+            glClearColor(0.0f, 1.0f, 0.0f, 1.0f);
+            glClear(GL_COLOR_BUFFER_BIT);
+            const Image throughParent = ReadFbo(fbo, kSize, kSize);
+            ExpectRegion(throughParent, 0, kSize - 1, 0, kSize - 1, Rgba8{0, 255, 0, 255}, 1,
+                         "a write through the view must be visible through the parent - they are one "
+                         "storage, not two");
+        }
+
+        // ------------------------------------------------------------------------------------
+        // Format reinterpretation within a view class (GL 4.6 core table 8.21).
+        // ------------------------------------------------------------------------------------
+        TEST_F(TextureViewScenario, ReinterpretingViewReadsTheSameBitsThroughAnotherFormat) {
+            if (!Ready() || IsSkipped()) return;
+
+            // GL_RGBA8 and GL_R32UI are both VIEW_CLASS_32_BITS, so one may be viewed as the
+            // other. Filling the RGBA8 storage with a known byte pattern makes the R32UI view's
+            // answer a fact about the BITS rather than about the colour.
+            const GLuint storage = MakeImmutable2D(GL_RGBA8, 1, kSize, kSize);
+            std::vector<std::uint8_t> texels(static_cast<std::size_t>(kSize) * kSize * 4);
+            for (std::size_t i = 0; i < texels.size(); i += 4) {
+                texels[i + 0] = 0x40;
+                texels[i + 1] = 0x80;
+                texels[i + 2] = 0xC0;
+                texels[i + 3] = 0xFF;
+            }
+            glBindTexture(GL_TEXTURE_2D, storage);
+            glPixelStorei(GL_UNPACK_ALIGNMENT, 1);
+            glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, kSize, kSize, GL_RGBA, GL_UNSIGNED_BYTE, texels.data());
+            glBindTexture(GL_TEXTURE_2D, 0);
+            ASSERT_EQ(glGetError(), static_cast<GLenum>(GL_NO_ERROR)) << "seeding the storage raised an error";
+
+            // NEGATIVE CONTROL. Everything below reads the storage through a REINTERPRETING view,
+            // so a test that only asserted the view's answer could not tell "the reinterpret is
+            // wrong" from "the seed never reached the GPU at all". Read the same texels through
+            // the parent's own format first.
+            const GLuint parentFbo = MakeFbo();
+            glBindFramebuffer(GL_FRAMEBUFFER, parentFbo);
+            glFramebufferTexture2D(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, storage, 0);
+            ASSERT_EQ(glCheckFramebufferStatus(GL_FRAMEBUFFER), static_cast<GLenum>(GL_FRAMEBUFFER_COMPLETE));
+            const Image seeded = ReadFbo(parentFbo, kSize, kSize);
+            ExpectRegion(seeded, 0, kSize - 1, 0, kSize - 1, Rgba8{0x40, 0x80, 0xC0, 0xFF}, 1,
+                         "control: the storage must hold the seeded byte pattern before any view reads it");
+
+            const GLuint view = MakeTexture();
+            glTextureView(view, GL_TEXTURE_2D, storage, GL_R32UI, 0, 1, 0, 1);
+            ASSERT_EQ(glGetError(), static_cast<GLenum>(GL_NO_ERROR)) << "an in-class reinterpret must be accepted";
+            glBindTexture(GL_TEXTURE_2D, view);
+            glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
+            glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
+
+            GLint viewFormat = 0;
+            glGetTexLevelParameteriv(GL_TEXTURE_2D, 0, GL_TEXTURE_INTERNAL_FORMAT, &viewFormat);
+            glBindTexture(GL_TEXTURE_2D, 0);
+            EXPECT_EQ(viewFormat, GL_R32UI) << "the view must report its OWN internal format";
+
+            // A REAL GL_R32UI texture holding the very word the storage's bytes spell. The
+            // assertion below is that the view and this texture sample IDENTICALLY.
+            //
+            // Comparing against a reference texture rather than against a hard-coded colour is
+            // deliberate. Sampling a 32-bit integer texture is not itself what this scenario is
+            // about, and llvmpipe's ES driver does it inconsistently (verified outside MobileGL,
+            // with a raw-EGL program that reproduces the same wrong decode with NO view in play).
+            // Holding both sides to the same driver factors that out completely: whatever the
+            // driver makes of a usampler2D fetch, the view has to make the same thing of it, or
+            // it is not delivering the storage's bits. A view that samples zero, that lands on
+            // the wrong texels, or that lost its format still fails.
+            constexpr std::uint32_t kExpectedWord = 0xFFC08040u;  // little-endian A,B,G,R
+            const GLuint reference = MakeImmutable2D(GL_R32UI, 1, kSize, kSize);
+            std::vector<std::uint32_t> words(static_cast<std::size_t>(kSize) * kSize, kExpectedWord);
+            glBindTexture(GL_TEXTURE_2D, reference);
+            glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, kSize, kSize, GL_RED_INTEGER, GL_UNSIGNED_INT, words.data());
+            glBindTexture(GL_TEXTURE_2D, 0);
+            ASSERT_EQ(glGetError(), static_cast<GLenum>(GL_NO_ERROR)) << "seeding the reference texture failed";
+
+            const GLuint program = MakeProgram(kQuadVertexSource, kDecodeWordFragmentSource);
+            ASSERT_NE(program, 0u);
+
+            ColorFbo destination = MakeColorFbo(kSize, kSize);
+            ASSERT_NE(destination.fbo, 0u);
+            const auto decodeThrough = [&](GLuint texture) {
+                glBindFramebuffer(GL_FRAMEBUFFER, destination.fbo);
+                glViewport(0, 0, kSize, kSize);
+                glDisable(GL_DEPTH_TEST);
+                glDisable(GL_STENCIL_TEST);
+                glClearColor(0.0f, 0.0f, 0.0f, 0.0f);
+                glClear(GL_COLOR_BUFFER_BIT);
+                glUseProgram(program);
+                glActiveTexture(GL_TEXTURE0);
+                glBindTexture(GL_TEXTURE_2D, texture);
+                glUniform1i(glGetUniformLocation(program, "uWords"), 0);
+                DrawQuad(program, -1.0f, -1.0f, 1.0f, 1.0f, 0.0f);
+                EXPECT_EQ(glGetError(), static_cast<GLenum>(GL_NO_ERROR)) << "sampling raised an error";
+                return ReadFbo(destination.fbo, kSize, kSize);
+            };
+
+            const Image throughReference = decodeThrough(reference);
+            const Image throughView = decodeThrough(view);
+
+            // Guard against the degenerate agreement of two black images: the reference must
+            // itself carry something, or "identical" would prove nothing.
+            const Rgba8 referenceTexel = throughReference.At(kSize / 2, kSize / 2);
+            ASSERT_FALSE(referenceTexel == (Rgba8{0, 0, 0, 0}))
+                << "the reference GL_R32UI texture sampled as nothing, so the comparison below is vacuous";
+
+            std::size_t mismatches = 0;
+            for (int y = 0; y < kSize; ++y) {
+                for (int x = 0; x < kSize; ++x) {
+                    if (!(throughView.At(x, y) == throughReference.At(x, y))) ++mismatches;
+                }
+            }
+            EXPECT_EQ(mismatches, 0u)
+                << "the GL_R32UI view of GL_RGBA8 storage must sample exactly what a real GL_R32UI texture "
+                   "holding the same word does; view centre is " << Describe(throughView.At(kSize / 2, kSize / 2))
+                << ", reference centre is " << Describe(referenceTexel);
+            DestroyColorFbo(destination);
+        }
+
+        // ------------------------------------------------------------------------------------
+        // Sub-ranges: one mip level of two, and one layer of an array.
+        // ------------------------------------------------------------------------------------
+        TEST_F(TextureViewScenario, ViewOfOneMipLevelAddressesThatLevelAsItsOwnLevelZero) {
+            if (!Ready() || IsSkipped()) return;
+
+            const GLuint storage = MakeImmutable2D(GL_RGBA8, 2, kSize, kSize);
+            // Level 0 red, level 1 blue, so the view's answer names the level it opened onto.
+            const GLuint seedFbo = MakeFbo();
+            glBindFramebuffer(GL_FRAMEBUFFER, seedFbo);
+            glDisable(GL_DEPTH_TEST);
+            glDisable(GL_STENCIL_TEST);
+            glFramebufferTexture2D(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, storage, 0);
+            glViewport(0, 0, kSize, kSize);
+            glClearColor(1.0f, 0.0f, 0.0f, 1.0f);
+            glClear(GL_COLOR_BUFFER_BIT);
+            glFramebufferTexture2D(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, storage, 1);
+            glViewport(0, 0, kSize / 2, kSize / 2);
+            glClearColor(0.0f, 0.0f, 1.0f, 1.0f);
+            glClear(GL_COLOR_BUFFER_BIT);
+            ASSERT_EQ(glGetError(), static_cast<GLenum>(GL_NO_ERROR)) << "seeding the mip chain raised an error";
+
+            const GLuint view = MakeTexture();
+            glTextureView(view, GL_TEXTURE_2D, storage, GL_RGBA8, 1, 1, 0, 1);
+            ASSERT_EQ(glGetError(), static_cast<GLenum>(GL_NO_ERROR));
+
+            GLint minLevel = -1;
+            GLint numLevels = -1;
+            GLint immutableLevels = -1;
+            glBindTexture(GL_TEXTURE_2D, view);
+            glGetTexParameteriv(GL_TEXTURE_2D, GL_TEXTURE_VIEW_MIN_LEVEL, &minLevel);
+            glGetTexParameteriv(GL_TEXTURE_2D, GL_TEXTURE_VIEW_NUM_LEVELS, &numLevels);
+            glGetTexParameteriv(GL_TEXTURE_2D, GL_TEXTURE_IMMUTABLE_LEVELS, &immutableLevels);
+            glBindTexture(GL_TEXTURE_2D, 0);
+            EXPECT_EQ(minLevel, 1);
+            EXPECT_EQ(numLevels, 1);
+            // GL 4.6 core 8.18: inherited from the ORIGINAL, not set to <numlevels>.
+            EXPECT_EQ(immutableLevels, 2) << "TEXTURE_IMMUTABLE_LEVELS is the original texture's value";
+
+            // The view's level 0 IS the parent's level 1: attaching level 0 of the view must find
+            // the blue half-size image.
+            const GLuint viewFbo = MakeFbo();
+            glBindFramebuffer(GL_FRAMEBUFFER, viewFbo);
+            glFramebufferTexture2D(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, view, 0);
+            ASSERT_EQ(glCheckFramebufferStatus(GL_FRAMEBUFFER), static_cast<GLenum>(GL_FRAMEBUFFER_COMPLETE));
+            const Image levelOne = ReadFbo(viewFbo, kSize / 2, kSize / 2);
+            ExpectRegion(levelOne, 0, kSize / 2 - 1, 0, kSize / 2 - 1, Rgba8{0, 0, 255, 255}, 1,
+                         "the view's level 0 must be the parent's level 1 (blue), not its level 0 (red)");
+        }
+
+        TEST_F(TextureViewScenario, ViewOfOneArrayLayerAddressesThatLayer) {
+            if (!Ready() || IsSkipped()) return;
+
+            constexpr int kLayers = 4;
+            constexpr int kChosenLayer = 2;
+            const GLuint storage = MakeTexture();
+            glBindTexture(GL_TEXTURE_2D_ARRAY, storage);
+            glTexStorage3D(GL_TEXTURE_2D_ARRAY, 1, GL_RGBA8, kSize, kSize, kLayers);
+            glTexParameteri(GL_TEXTURE_2D_ARRAY, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
+            glTexParameteri(GL_TEXTURE_2D_ARRAY, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
+            ASSERT_EQ(glGetError(), static_cast<GLenum>(GL_NO_ERROR));
+
+            // A different colour per layer, so a view that lost its layer offset reads the wrong
+            // one rather than merely reading nothing.
+            const GLuint seedFbo = MakeFbo();
+            glBindFramebuffer(GL_FRAMEBUFFER, seedFbo);
+            glDisable(GL_DEPTH_TEST);
+            glDisable(GL_STENCIL_TEST);
+            glViewport(0, 0, kSize, kSize);
+            for (int layer = 0; layer < kLayers; ++layer) {
+                glFramebufferTextureLayer(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, storage, 0, layer);
+                glClearColor(float(layer) / 8.0f, 1.0f - float(layer) / 8.0f, 0.5f, 1.0f);
+                glClear(GL_COLOR_BUFFER_BIT);
+            }
+            ASSERT_EQ(glGetError(), static_cast<GLenum>(GL_NO_ERROR)) << "seeding the array layers raised an error";
+
+            const GLuint view = MakeTexture();
+            glTextureView(view, GL_TEXTURE_2D, storage, GL_RGBA8, 0, 1, kChosenLayer, 1);
+            ASSERT_EQ(glGetError(), static_cast<GLenum>(GL_NO_ERROR)) << "2D_ARRAY -> 2D is a legal view pair";
+
+            GLint minLayer = -1;
+            GLint numLayers = -1;
+            glBindTexture(GL_TEXTURE_2D, view);
+            glGetTexParameteriv(GL_TEXTURE_2D, GL_TEXTURE_VIEW_MIN_LAYER, &minLayer);
+            glGetTexParameteriv(GL_TEXTURE_2D, GL_TEXTURE_VIEW_NUM_LAYERS, &numLayers);
+            glBindTexture(GL_TEXTURE_2D, 0);
+            EXPECT_EQ(minLayer, kChosenLayer);
+            EXPECT_EQ(numLayers, 1);
+
+            const GLuint viewFbo = MakeFbo();
+            glBindFramebuffer(GL_FRAMEBUFFER, viewFbo);
+            glFramebufferTexture2D(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, view, 0);
+            ASSERT_EQ(glCheckFramebufferStatus(GL_FRAMEBUFFER), static_cast<GLenum>(GL_FRAMEBUFFER_COMPLETE));
+            const Image sliced = ReadFbo(viewFbo, kSize, kSize);
+            const Rgba8 expected{static_cast<std::uint8_t>(kChosenLayer * 255 / 8),
+                                 static_cast<std::uint8_t>(255 - kChosenLayer * 255 / 8), 128, 255};
+            ExpectRegion(sliced, 0, kSize - 1, 0, kSize - 1, expected, 2,
+                         "a single-layer 2D view of an array must address the layer it named");
+        }
+
+        // ------------------------------------------------------------------------------------
+        // Writing THROUGH a layer-sliced view. The read direction is covered above; this is the
+        // write direction, and it is the one that can corrupt the parent rather than merely
+        // return the wrong pixels - a view whose texel path forgot its layer origin writes over
+        // the parent's layer 0 while the application believes it addressed layer minLayer.
+        // ------------------------------------------------------------------------------------
+        TEST_F(TextureViewScenario, WritingThroughALayerSlicedViewLandsOnItsOwnLayers) {
+            if (!Ready() || IsSkipped()) return;
+
+            constexpr int kLayers = 4;
+            constexpr int kViewMinLayer = 2;
+            const GLuint storage = MakeTexture();
+            glBindTexture(GL_TEXTURE_2D_ARRAY, storage);
+            glTexStorage3D(GL_TEXTURE_2D_ARRAY, 1, GL_RGBA8, kSize, kSize, kLayers);
+            glTexParameteri(GL_TEXTURE_2D_ARRAY, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
+            glTexParameteri(GL_TEXTURE_2D_ARRAY, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
+            ASSERT_EQ(glGetError(), static_cast<GLenum>(GL_NO_ERROR));
+
+            const auto layerFill = [](int layer) {
+                return Rgba8{static_cast<std::uint8_t>(10 + layer * 20),
+                             static_cast<std::uint8_t>(200 - layer * 20), 30, 255};
+            };
+            // Seeded by CPU sub-image rather than by rendering, deliberately: this scenario is
+            // about the view's LAYER ORIGIN, and seeding through the GPU would additionally
+            // depend on a CPU sub-image reaching a layer whose content the GPU wrote - which
+            // DirectVulkan does not currently do even for a plain array texture (no view
+            // involved), and which would make a failure here unattributable.
+            const auto uploadLayer = [&](GLuint texture, int layer, Rgba8 colour) {
+                std::vector<std::uint8_t> texels(static_cast<std::size_t>(kSize) * kSize * 4);
+                for (std::size_t i = 0; i < texels.size(); i += 4) {
+                    texels[i + 0] = colour.r;
+                    texels[i + 1] = colour.g;
+                    texels[i + 2] = colour.b;
+                    texels[i + 3] = colour.a;
+                }
+                glBindTexture(GL_TEXTURE_2D_ARRAY, texture);
+                glPixelStorei(GL_UNPACK_ALIGNMENT, 1);
+                glTexSubImage3D(GL_TEXTURE_2D_ARRAY, 0, 0, 0, layer, kSize, kSize, 1, GL_RGBA, GL_UNSIGNED_BYTE,
+                                texels.data());
+                glBindTexture(GL_TEXTURE_2D_ARRAY, 0);
+            };
+            for (int layer = 0; layer < kLayers; ++layer) {
+                uploadLayer(storage, layer, layerFill(layer));
+            }
+            const GLuint fbo = MakeFbo();
+            glDisable(GL_DEPTH_TEST);
+            glDisable(GL_STENCIL_TEST);
+            ASSERT_EQ(glGetError(), static_cast<GLenum>(GL_NO_ERROR)) << "seeding the layers raised an error";
+
+            // A two-layer window starting at layer 2, so a lost offset lands on layer 0 - which
+            // the assertions below would see as an untouched layer that moved.
+            const GLuint view = MakeTexture();
+            glTextureView(view, GL_TEXTURE_2D_ARRAY, storage, GL_RGBA8, 0, 1, kViewMinLayer, 2);
+            ASSERT_EQ(glGetError(), static_cast<GLenum>(GL_NO_ERROR));
+
+            // Write the view's OWN layer 0, i.e. the storage's layer 2.
+            constexpr Rgba8 kPainted{255, 0, 255, 255};
+            std::vector<std::uint8_t> texels(static_cast<std::size_t>(kSize) * kSize * 4);
+            for (std::size_t i = 0; i < texels.size(); i += 4) {
+                texels[i + 0] = kPainted.r;
+                texels[i + 1] = kPainted.g;
+                texels[i + 2] = kPainted.b;
+                texels[i + 3] = kPainted.a;
+            }
+            glBindTexture(GL_TEXTURE_2D_ARRAY, view);
+            glPixelStorei(GL_UNPACK_ALIGNMENT, 1);
+            glTexSubImage3D(GL_TEXTURE_2D_ARRAY, 0, 0, 0, 0, kSize, kSize, 1, GL_RGBA, GL_UNSIGNED_BYTE,
+                            texels.data());
+            glBindTexture(GL_TEXTURE_2D_ARRAY, 0);
+            ASSERT_EQ(glGetError(), static_cast<GLenum>(GL_NO_ERROR)) << "writing through the view raised an error";
+
+            // POSITIVE CONTROL, through the parent's own name and into a layer outside the view's
+            // window. It makes the assertions below able to tell "the view lost its layer origin"
+            // from "a CPU sub-image into this array does not reach the GPU at all", which is a
+            // different question and not one a texture view can answer.
+            constexpr Rgba8 kControl{0, 0, 255, 255};
+            std::vector<std::uint8_t> controlTexels(texels.size());
+            for (std::size_t i = 0; i < controlTexels.size(); i += 4) {
+                controlTexels[i + 0] = kControl.r;
+                controlTexels[i + 1] = kControl.g;
+                controlTexels[i + 2] = kControl.b;
+                controlTexels[i + 3] = kControl.a;
+            }
+            glBindTexture(GL_TEXTURE_2D_ARRAY, storage);
+            glTexSubImage3D(GL_TEXTURE_2D_ARRAY, 0, 0, 0, 1, kSize, kSize, 1, GL_RGBA, GL_UNSIGNED_BYTE,
+                            controlTexels.data());
+            glBindTexture(GL_TEXTURE_2D_ARRAY, 0);
+            ASSERT_EQ(glGetError(), static_cast<GLenum>(GL_NO_ERROR)) << "the control write raised an error";
+
+            // Read every layer of the PARENT back: only the one the view's layer 0 maps to may
+            // have changed.
+            for (int layer = 0; layer < kLayers; ++layer) {
+                glBindFramebuffer(GL_FRAMEBUFFER, fbo);
+                glFramebufferTextureLayer(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, storage, 0, layer);
+                glReadBuffer(GL_COLOR_ATTACHMENT0);
+                const Image image = ReadPixels(kSize, kSize);
+                Rgba8 expected = layerFill(layer);
+                const char* what = "a layer outside the view's window must not have been written";
+                if (layer == kViewMinLayer) {
+                    expected = kPainted;
+                    what = "the view's layer 0 must be the storage layer it named";
+                } else if (layer == 1) {
+                    expected = kControl;
+                    what = "control: a sub-image written through the PARENT must reach its layer";
+                }
+                ExpectRegion(image, 0, kSize - 1, 0, kSize - 1, expected, 2, what);
+            }
+        }
+
+        // ------------------------------------------------------------------------------------
+        // Views of views compose; the composed view still reaches the ROOT storage.
+        // ------------------------------------------------------------------------------------
+        TEST_F(TextureViewScenario, ViewOfAViewComposesTheLevelRanges) {
+            if (!Ready() || IsSkipped()) return;
+
+            constexpr int kLevels = 3;
+            const GLuint storage = MakeImmutable2D(GL_RGBA8, kLevels, kSize, kSize);
+            const GLuint seedFbo = MakeFbo();
+            glBindFramebuffer(GL_FRAMEBUFFER, seedFbo);
+            glDisable(GL_DEPTH_TEST);
+            glDisable(GL_STENCIL_TEST);
+            for (int level = 0; level < kLevels; ++level) {
+                glFramebufferTexture2D(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, storage, level);
+                glViewport(0, 0, kSize >> level, kSize >> level);
+                glClearColor(0.0f, 0.0f, float(level + 1) / 4.0f, 1.0f);
+                glClear(GL_COLOR_BUFFER_BIT);
+            }
+            ASSERT_EQ(glGetError(), static_cast<GLenum>(GL_NO_ERROR));
+
+            // First view opens onto levels [1, 3); the second takes level 1 OF THAT, which is the
+            // root's level 2. GL 4.6 core 8.18 makes the offsets add.
+            const GLuint firstView = MakeTexture();
+            glTextureView(firstView, GL_TEXTURE_2D, storage, GL_RGBA8, 1, 2, 0, 1);
+            ASSERT_EQ(glGetError(), static_cast<GLenum>(GL_NO_ERROR));
+            const GLuint secondView = MakeTexture();
+            glTextureView(secondView, GL_TEXTURE_2D, firstView, GL_RGBA8, 1, 1, 0, 1);
+            ASSERT_EQ(glGetError(), static_cast<GLenum>(GL_NO_ERROR)) << "origtexture may itself be a view";
+
+            GLint minLevel = -1;
+            GLint numLevels = -1;
+            glBindTexture(GL_TEXTURE_2D, secondView);
+            glGetTexParameteriv(GL_TEXTURE_2D, GL_TEXTURE_VIEW_MIN_LEVEL, &minLevel);
+            glGetTexParameteriv(GL_TEXTURE_2D, GL_TEXTURE_VIEW_NUM_LEVELS, &numLevels);
+            glBindTexture(GL_TEXTURE_2D, 0);
+            EXPECT_EQ(minLevel, 2) << "TEXTURE_VIEW_MIN_LEVEL adds the original's";
+            EXPECT_EQ(numLevels, 1);
+
+            const GLuint viewFbo = MakeFbo();
+            glBindFramebuffer(GL_FRAMEBUFFER, viewFbo);
+            glFramebufferTexture2D(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, secondView, 0);
+            ASSERT_EQ(glCheckFramebufferStatus(GL_FRAMEBUFFER), static_cast<GLenum>(GL_FRAMEBUFFER_COMPLETE));
+            const Image composed = ReadFbo(viewFbo, kSize >> 2, kSize >> 2);
+            ExpectRegion(composed, 0, (kSize >> 2) - 1, 0, (kSize >> 2) - 1, Rgba8{0, 0, 191, 255}, 2,
+                         "the composed view must land on the root's level 2");
+        }
+
+        // ------------------------------------------------------------------------------------
+        // GL name-deletion semantics: the storage outlives the original's NAME.
+        // ------------------------------------------------------------------------------------
+        TEST_F(TextureViewScenario, DeletingTheOriginalKeepsTheViewUsable) {
+            if (!Ready() || IsSkipped()) return;
+
+            GLuint storage = 0;
+            glGenTextures(1, &storage);
+            glBindTexture(GL_TEXTURE_2D, storage);
+            glTexStorage2D(GL_TEXTURE_2D, 1, GL_RGBA8, kSize, kSize);
+            glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
+            glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
+
+            const GLuint view = MakeTexture();
+            glTextureView(view, GL_TEXTURE_2D, storage, GL_RGBA8, 0, 1, 0, 1);
+            ASSERT_EQ(glGetError(), static_cast<GLenum>(GL_NO_ERROR));
+
+            const GLuint fbo = MakeFbo();
+            glBindFramebuffer(GL_FRAMEBUFFER, fbo);
+            glFramebufferTexture2D(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, storage, 0);
+            glViewport(0, 0, kSize, kSize);
+            glDisable(GL_DEPTH_TEST);
+            glDisable(GL_STENCIL_TEST);
+            glClearColor(0.0f, 1.0f, 1.0f, 1.0f);
+            glClear(GL_COLOR_BUFFER_BIT);
+            glBindFramebuffer(GL_FRAMEBUFFER, 0);
+
+            // The NAME goes; the storage may not, because a view still references it
+            // (GL 4.6 core 5.1.2 - an object is not deleted while anything still refers to it).
+            glDeleteTextures(1, &storage);
+            EXPECT_EQ(glIsTexture(storage), static_cast<GLboolean>(GL_FALSE));
+            ASSERT_EQ(glGetError(), static_cast<GLenum>(GL_NO_ERROR));
+
+            // Sample the view through a shader, so the answer comes from a live descriptor rather
+            // than from an attachment the frontend might have kept alive by other means.
+            ColorFbo destination = MakeColorFbo(kSize, kSize);
+            ASSERT_NE(destination.fbo, 0u);
+            glBindFramebuffer(GL_FRAMEBUFFER, destination.fbo);
+            glViewport(0, 0, kSize, kSize);
+            glClearColor(0.0f, 0.0f, 0.0f, 1.0f);
+            glClear(GL_COLOR_BUFFER_BIT);
+
+            const GLuint program = MakeProgram(kQuadVertexSource, kSampleFragmentSource);
+            ASSERT_NE(program, 0u);
+            glUseProgram(program);
+            glActiveTexture(GL_TEXTURE0);
+            glBindTexture(GL_TEXTURE_2D, view);
+            glUniform1i(glGetUniformLocation(program, "uTexture"), 0);
+            glUniform1f(glGetUniformLocation(program, "uLod"), 0.0f);
+            DrawQuad(program, -1.0f, -1.0f, 1.0f, 1.0f, 0.0f);
+            ASSERT_EQ(glGetError(), static_cast<GLenum>(GL_NO_ERROR));
+
+            const Image sampled = ReadFbo(destination.fbo, kSize, kSize);
+            ExpectRegion(sampled, 1, kSize - 2, 1, kSize - 2, Rgba8{0, 255, 255, 255}, 2,
+                         "the view must still reach its storage after the original's name was deleted");
+            DestroyColorFbo(destination);
+        }
+    } // namespace
+} // namespace MGITest

MobileGL/MG_IntegrationTest/Scenarios/UnboundImageDescriptorScenario.cpp

@@ -0,0 +1,380 @@
+// MobileGL - MobileGL/MG_IntegrationTest/Scenarios/UnboundImageDescriptorScenario.cpp
+// Copyright (c) 2026 MobileGL-Dev
+// Licensed under the GNU Lesser General Public License v3.0:
+//   https://www.gnu.org/licenses/gpl-3.0.txt
+//   https://www.gnu.org/licenses/lgpl-3.0.txt
+// SPDX-License-Identifier: LGPL-3.0-only
+// End of Source File Header
+//
+// Scenario - A PROGRAM DECLARES AN IMAGE-BACKED RESOURCE AND THE APPLICATION BINDS NOTHING.
+//
+// The sibling of GuiBatchScenario's MeshesBlockLeftUnbound, one descriptor kind further out.
+// That one pinned an unbound shader storage BLOCK; the same "nothing is bound, so lose the
+// whole draw" shape survived in the three image-backed kinds:
+//
+//   * `samplerBuffer`  - a texture unit with no buffer texture on it, and a buffer texture with
+//                        no GL buffer attached to it. Both make the sampler INCOMPLETE (GL 4.6
+//                        core 8.9, 8.24), and sampling an incomplete texture returns undefined
+//                        VALUES. It is not an error and it is not a lost draw.
+//   * `imageBuffer`    - an image unit with nothing on it. GL 4.6 core 8.26 is explicit: loads
+//                        return zero and stores are discarded.
+//   * `image2D`        - the same rule, through a VkImageView rather than a VkBufferView.
+//
+// Vulkan has no such thing as an unwritten descriptor, so DirectVulkan's descriptor resolution
+// used to answer "no valid descriptor" and both SetupDraw and DispatchCompute skip everything on
+// that answer - the draw or dispatch simply never happened, silently. Every test below asserts
+// on the OTHER work in the same shader: the pixels the fragment stage painted, or the buffer the
+// dispatch filled. All of it is unrelated to the unbound resource and all of it disappeared.
+//
+// The unbound resource is STATICALLY USED in every case, because an unreferenced one is
+// optimised out before it ever reaches a descriptor and would prove nothing. Where the use is a
+// read it sits behind a uniform-controlled branch that is false at runtime - the descriptor is
+// declared and must be written, but no undefined value reaches an assertion. Where it is a write
+// (the `writeonly` cases, which is how the real workloads spell it) it is unconditional: GL says
+// the store is discarded, so there is nothing to guard against.
+//
+// Reproduces on DirectVulkan only. DirectGLES forwards the unbound unit to the GLES driver,
+// which does what GL says, so it is the control - every test here must stay green on both.
+
+#include <cstdint>
+#include <string>
+#include <vector>
+
+#include "../Harness/HeadlessGL.h"
+#include "../Harness/ScenarioFixture.h"
+
+#ifdef GLAPI
+#undef GLAPI
+#endif
+#define GL_GLEXT_PROTOTYPES
+#include <GL/gl.h>
+#include <GL/glcorearb.h>
+#undef GL_GLEXT_PROTOTYPES
+
+namespace MGITest {
+    namespace {
+
+        constexpr int kFboSize = 32;
+        constexpr int kElements = 4;
+
+        // No vertex attributes: the quad's corners come from gl_VertexID, so nothing about the
+        // vertex fetch can be confused with the descriptor question under test.
+        constexpr const char* kQuadVertexSource = R"(#version 430 core
+void main() {
+    vec2 corner = vec2((gl_VertexID & 1) == 0 ? -1.0 : 1.0,
+                       (gl_VertexID & 2) == 0 ? -1.0 : 1.0);
+    gl_Position = vec4(corner, 0.0, 1.0);
+}
+)";
+
+        // The assertion in every draw case: opaque green everywhere. The unbound resource
+        // contributes nothing to it - u_readUnbound is 0, so the fetch never runs - but the
+        // descriptor for it still has to exist, which is the point.
+        constexpr const char* kSamplerBufferFragmentSource = R"(#version 430 core
+uniform samplerBuffer u_unbound;
+uniform int u_readUnbound;
+out vec4 o_color;
+void main() {
+    vec4 color = vec4(0.0, 1.0, 0.0, 1.0);
+    if (u_readUnbound != 0) {
+        color = texelFetch(u_unbound, 0);
+    }
+    o_color = color;
+}
+)";
+
+        constexpr const char* kSamplerBufferComputeSource = R"(#version 430 core
+layout(local_size_x = 1) in;
+layout(std430, binding = 0) buffer Output { uint g_data[]; };
+uniform samplerBuffer u_unbound;
+uniform int u_readUnbound;
+void main() {
+    uint index = gl_GlobalInvocationID.x;
+    uint value = index + 1u;
+    if (u_readUnbound != 0) {
+        value += uint(texelFetch(u_unbound, 0).r);
+    }
+    g_data[index] = value;
+}
+)";
+
+        // writeonly, and the store is unconditional: this is how AcceleratedRendering and the
+        // conformance cases spell an image the shader only produces into. GL discards the store
+        // when the unit is empty; nothing here reads it back.
+        constexpr const char* kImageBufferFragmentSource = R"(#version 430 core
+layout(binding = 0, r32ui) uniform writeonly uimageBuffer u_unbound;
+out vec4 o_color;
+void main() {
+    imageStore(u_unbound, 0, uvec4(7u));
+    o_color = vec4(0.0, 1.0, 0.0, 1.0);
+}
+)";
+
+        constexpr const char* kImageBufferComputeSource = R"(#version 430 core
+layout(local_size_x = 1) in;
+layout(std430, binding = 0) buffer Output { uint g_data[]; };
+layout(binding = 0, r32ui) uniform writeonly uimageBuffer u_unbound;
+void main() {
+    uint index = gl_GlobalInvocationID.x;
+    imageStore(u_unbound, int(index), uvec4(7u));
+    g_data[index] = index + 1u;
+}
+)";
+
+        constexpr const char* kImage2DFragmentSource = R"(#version 430 core
+layout(binding = 0, rgba8) uniform writeonly image2D u_unbound;
+out vec4 o_color;
+void main() {
+    imageStore(u_unbound, ivec2(0, 0), vec4(1.0));
+    o_color = vec4(0.0, 1.0, 0.0, 1.0);
+}
+)";
+
+        constexpr const char* kImage2DComputeSource = R"(#version 430 core
+layout(local_size_x = 1) in;
+layout(std430, binding = 0) buffer Output { uint g_data[]; };
+layout(binding = 0, rgba8) uniform writeonly image2D u_unbound;
+void main() {
+    uint index = gl_GlobalInvocationID.x;
+    imageStore(u_unbound, ivec2(int(index), 0), vec4(1.0));
+    g_data[index] = index + 1u;
+}
+)";
+
+        // No layout format at all, which GLSL 4.20 allows for a write-only image. The reflection
+        // then carries NO format for the binding, so the placeholder descriptor can only be
+        // constrained by the declaration's numeric class - a different route through the fix than
+        // every typed case above.
+        constexpr const char* kFormatlessImage2DComputeSource = R"(#version 430 core
+layout(local_size_x = 1) in;
+layout(std430, binding = 0) buffer Output { uint g_data[]; };
+layout(binding = 0) uniform writeonly image2D u_unbound;
+void main() {
+    uint index = gl_GlobalInvocationID.x;
+    imageStore(u_unbound, ivec2(int(index), 0), vec4(1.0));
+    g_data[index] = index + 1u;
+}
+)";
+
+        class UnboundImageDescriptorScenario : public ScenarioTest {
+        protected:
+            void SetUp() override {
+                ScenarioTest::SetUp();
+                if (!Ready()) return;
+                m_target = MakeColorFbo(kFboSize, kFboSize);
+                ASSERT_NE(m_target.fbo, 0u) << "could not create the render target";
+                glGenVertexArrays(1, &m_vao);
+                glGenBuffers(1, &m_storage);
+                // The harness shares one context across every scenario in the process, so an
+                // earlier one may well have left a texture on unit 0 or an image on unit 0. The
+                // whole subject here is that nothing is bound, so say so rather than assume it.
+                glActiveTexture(GL_TEXTURE0);
+                glBindTexture(GL_TEXTURE_BUFFER, 0);
+                glBindTexture(GL_TEXTURE_2D, 0);
+                glBindImageTexture(0, 0, 0, GL_FALSE, 0, GL_READ_WRITE, GL_RGBA8);
+                FirstGLError();
+            }
+
+            void TearDown() override {
+                if (!Ready()) return;
+                glUseProgram(0);
+                glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 0, 0);
+                if (m_program != 0) glDeleteProgram(m_program);
+                if (m_storage != 0) glDeleteBuffers(1, &m_storage);
+                if (m_vao != 0) glDeleteVertexArrays(1, &m_vao);
+                BindDefaultFramebuffer();
+                DestroyColorFbo(m_target);
+                glViewport(0, 0, Gl().Width(), Gl().Height());
+            }
+
+            // Each case needs exactly one kind of opaque uniform in one stage, and a host with
+            // none of that kind there would report a failure that is about the host, not the fix.
+            // Asked for by the limit that governs the kind under test and no other: a guard that
+            // over-asks turns into a silent skip of the very thing the case exists for.
+            static bool LimitIsAtLeastOne(GLenum limit) {
+                GLint value = 0;
+                glGetIntegerv(limit, &value);
+                while (glGetError() != GL_NO_ERROR) {
+                }
+                return value >= 1;
+            }
+
+            unsigned int MakeComputeProgram(const char* source) {
+                const GLuint shader = glCreateShader(GL_COMPUTE_SHADER);
+                glShaderSource(shader, 1, &source, nullptr);
+                glCompileShader(shader);
+                GLint compiled = GL_FALSE;
+                glGetShaderiv(shader, GL_COMPILE_STATUS, &compiled);
+                if (compiled == GL_FALSE) {
+                    char log[4096] = {};
+                    glGetShaderInfoLog(shader, sizeof(log) - 1, nullptr, log);
+                    ADD_FAILURE() << "the compute shader did not compile: " << log;
+                    glDeleteShader(shader);
+                    return 0;
+                }
+                const GLuint program = glCreateProgram();
+                glAttachShader(program, shader);
+                glLinkProgram(program);
+                glDeleteShader(shader);
+                GLint linked = GL_FALSE;
+                glGetProgramiv(program, GL_LINK_STATUS, &linked);
+                if (linked == GL_FALSE) {
+                    char log[4096] = {};
+                    glGetProgramInfoLog(program, sizeof(log) - 1, nullptr, log);
+                    ADD_FAILURE() << "the compute program did not link: " << log;
+                    glDeleteProgram(program);
+                    return 0;
+                }
+                return program;
+            }
+
+            // Fills a four-element SSBO with 1..4 while the unbound resource is declared and
+            // statically used. Zeros everywhere mean the dispatch never ran.
+            void ExpectDispatchStillRuns(const char* source, const char* what) {
+                m_program = MakeComputeProgram(source);
+                ASSERT_NE(m_program, 0u);
+
+                const std::vector<unsigned int> zeros(static_cast<std::size_t>(kElements), 0u);
+                glBindBuffer(GL_SHADER_STORAGE_BUFFER, m_storage);
+                glBufferData(GL_SHADER_STORAGE_BUFFER,
+                             static_cast<GLsizeiptr>(zeros.size() * sizeof(unsigned int)), zeros.data(),
+                             GL_DYNAMIC_COPY);
+                glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 0, m_storage);
+                ASSERT_EQ(FirstGLError(), 0u) << "setting up the output buffer raised a GL error";
+
+                glUseProgram(m_program);
+                const GLint readUnbound = glGetUniformLocation(m_program, "u_readUnbound");
+                if (readUnbound != -1) {
+                    glUniform1i(readUnbound, 0);
+                }
+                glDispatchCompute(kElements, 1, 1);
+                glMemoryBarrier(GL_BUFFER_UPDATE_BARRIER_BIT);
+                EXPECT_EQ(FirstGLError(), 0u) << "the dispatch raised a GL error (" << what << ")";
+
+                std::vector<unsigned int> values(static_cast<std::size_t>(kElements), 0xDEADBEEFu);
+                glBindBuffer(GL_SHADER_STORAGE_BUFFER, m_storage);
+                glGetBufferSubData(GL_SHADER_STORAGE_BUFFER, 0,
+                                   static_cast<GLsizeiptr>(values.size() * sizeof(unsigned int)), values.data());
+                for (int i = 0; i < kElements; ++i) {
+                    EXPECT_EQ(values[static_cast<std::size_t>(i)], static_cast<unsigned int>(i + 1))
+                        << "element " << i << " came back as " << values[static_cast<std::size_t>(i)]
+                        << "; zero everywhere means the whole dispatch was dropped over the unbound " << what;
+                }
+            }
+
+            // Paints the whole render target green while the unbound resource is declared and
+            // statically used. A black target means the draw never happened.
+            void ExpectDrawStillRuns(const char* fragmentSource, const char* what) {
+                std::string error;
+                m_program = CompileProgram(kQuadVertexSource, fragmentSource, &error);
+                ASSERT_NE(m_program, 0u) << error;
+
+                BindFbo(m_target);
+                ClearTo(0.0f, 0.0f, 0.0f, 1.0f);
+                glBindVertexArray(m_vao);
+                glUseProgram(m_program);
+                const GLint readUnbound = glGetUniformLocation(m_program, "u_readUnbound");
+                if (readUnbound != -1) {
+                    glUniform1i(readUnbound, 0);
+                }
+                glDrawArrays(GL_TRIANGLE_STRIP, 0, 4);
+                glBindVertexArray(0);
+                EXPECT_EQ(FirstGLError(), 0u) << "the draw raised a GL error (" << what << ")";
+
+                const Image image = ReadPixels(kFboSize, kFboSize);
+                ASSERT_FALSE(image.Empty()) << "the readback came back empty";
+                // Whole-region, not a centre pixel: the quad covers the target exactly, so
+                // anything short of all of it is a failure worth naming.
+                EXPECT_TRUE(RegionIsMostly(image, 0, kFboSize - 1, 0, kFboSize - 1, "green", 0.0,
+                                           std::string("the quad drawn with an unbound ") + what))
+                    << "an all-black target means the draw was dropped over the unbound " << what;
+            }
+
+            ColorFbo m_target{};
+            GLuint m_vao = 0;
+            GLuint m_storage = 0;
+            unsigned int m_program = 0;
+        };
+
+    } // namespace
+
+    // ---- uniform samplerBuffer (VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER) --------------------
+
+    TEST_F(UnboundImageDescriptorScenario, ADeclaredButUnboundSamplerBufferDoesNotLoseTheDispatch) {
+        if (!Ready() || IsSkipped()) return;
+        if (!LimitIsAtLeastOne(GL_MAX_COMPUTE_TEXTURE_IMAGE_UNITS)) {
+            GTEST_SKIP() << "the compute stage has no texture image units";
+        }
+        ExpectDispatchStillRuns(kSamplerBufferComputeSource, "samplerBuffer");
+    }
+
+    TEST_F(UnboundImageDescriptorScenario, ADeclaredButUnboundSamplerBufferDoesNotLoseTheDraw) {
+        if (!Ready() || IsSkipped()) return;
+        ExpectDrawStillRuns(kSamplerBufferFragmentSource, "samplerBuffer");
+    }
+
+    // The other way a texel-buffer descriptor comes out empty: the unit HAS a buffer texture, but
+    // no glTexBuffer ever attached a buffer object to it. GL calls that texture incomplete, which
+    // is undefined data and not a lost draw - a separate site in the resolve from the one above,
+    // and it used to return false too.
+    TEST_F(UnboundImageDescriptorScenario, ABufferTextureWithNoAttachedBufferDoesNotLoseTheDraw) {
+        if (!Ready() || IsSkipped()) return;
+
+        GLuint texture = 0;
+        glGenTextures(1, &texture);
+        glActiveTexture(GL_TEXTURE0);
+        glBindTexture(GL_TEXTURE_BUFFER, texture);
+        // Deliberately no glTexBuffer: the texture exists and is bound, and has no store.
+        ASSERT_EQ(FirstGLError(), 0u) << "binding an empty buffer texture raised a GL error";
+
+        ExpectDrawStillRuns(kSamplerBufferFragmentSource, "buffer texture with no attached buffer");
+
+        glBindTexture(GL_TEXTURE_BUFFER, 0);
+        glDeleteTextures(1, &texture);
+    }
+
+    // ---- writeonly imageBuffer (VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER) --------------------
+
+    TEST_F(UnboundImageDescriptorScenario, AWriteonlyImageBufferLeftUnboundDoesNotLoseTheDispatch) {
+        if (!Ready() || IsSkipped()) return;
+        if (!LimitIsAtLeastOne(GL_MAX_COMPUTE_IMAGE_UNIFORMS)) {
+            GTEST_SKIP() << "the compute stage has no image uniforms";
+        }
+        ExpectDispatchStillRuns(kImageBufferComputeSource, "imageBuffer");
+    }
+
+    TEST_F(UnboundImageDescriptorScenario, AWriteonlyImageBufferLeftUnboundDoesNotLoseTheDraw) {
+        if (!Ready() || IsSkipped()) return;
+        if (!LimitIsAtLeastOne(GL_MAX_FRAGMENT_IMAGE_UNIFORMS)) {
+            GTEST_SKIP() << "the fragment stage has no image uniforms";
+        }
+        ExpectDrawStillRuns(kImageBufferFragmentSource, "imageBuffer");
+    }
+
+    // ---- writeonly image2D (VK_DESCRIPTOR_TYPE_STORAGE_IMAGE) -------------------------------
+
+    TEST_F(UnboundImageDescriptorScenario, AWriteonlyImage2DLeftUnboundDoesNotLoseTheDispatch) {
+        if (!Ready() || IsSkipped()) return;
+        if (!LimitIsAtLeastOne(GL_MAX_COMPUTE_IMAGE_UNIFORMS)) {
+            GTEST_SKIP() << "the compute stage has no image uniforms";
+        }
+        ExpectDispatchStillRuns(kImage2DComputeSource, "image2D");
+    }
+
+    TEST_F(UnboundImageDescriptorScenario, AWriteonlyImage2DLeftUnboundDoesNotLoseTheDraw) {
+        if (!Ready() || IsSkipped()) return;
+        if (!LimitIsAtLeastOne(GL_MAX_FRAGMENT_IMAGE_UNIFORMS)) {
+            GTEST_SKIP() << "the fragment stage has no image uniforms";
+        }
+        ExpectDrawStillRuns(kImage2DFragmentSource, "image2D");
+    }
+
+    TEST_F(UnboundImageDescriptorScenario, AFormatlessWriteonlyImage2DLeftUnboundDoesNotLoseTheDispatch) {
+        if (!Ready() || IsSkipped()) return;
+        if (!LimitIsAtLeastOne(GL_MAX_COMPUTE_IMAGE_UNIFORMS)) {
+            GTEST_SKIP() << "the compute stage has no image uniforms";
+        }
+        ExpectDispatchStillRuns(kFormatlessImage2DComputeSource, "format-less image2D");
+    }
+
+} // namespace MGITest

MobileGL/MG_IntegrationTest/Scenarios/ViewportArrayScenario.cpp

@@ -55,10 +55,6 @@
 #include "../Harness/HeadlessGL.h"
 #include "../Harness/ScenarioFixture.h"
 
-// For the emulation switch the negative-control case below flips. Nothing else in this file needs
-// to know which backend it is running on.
-#include <Config.h>
-
 #ifdef GLAPI
 #undef GLAPI
 #endif
@@ -529,7 +525,8 @@ void main() { fragColor = vec4(float(gsIndex) * 16.0 / 255.0, 0.0, 0.0, 1.0); }
         //
         // Everything above is a claim about pixels, and a claim about pixels cannot tell an
         // emulation that works from a backend that was going to be right anyway. This case builds
-        // the SAME program with MOBILEGL_FORCE_VIEWPORT_ARRAY_EMULATION off and requires case 1's
+        // the SAME program in a process started with MOBILEGL_FORCE_VIEWPORT_ARRAY_EMULATION=0
+        // (the NoViewportArrayEmulation. ctest entry) and requires case 1's
         // result to COLLAPSE: with no routing, every geometry invocation rasterizes against
         // viewport 0's rectangle, so the last invocation paints the whole surface and every cell
         // reads 15 instead of its own index. That is the pre-emulation behaviour this backend had
@@ -544,25 +541,31 @@ void main() { fragColor = vec4(float(gsIndex) * 16.0 / 255.0, 0.0, 0.0, 1.0); }
                                 "gl_ViewportIndex natively and ignores it";
             }
 
-            // The feature table is a process-global and this fixture shares its context with every
-            // other scenario in the process, so the restore is not optional.
-            struct ScopedEmulationOff {
-                ScopedEmulationOff(): saved(MobileGL::MG_Config::Features.ViewportArrayEmulation) {
-                    MobileGL::MG_Config::Features.ViewportArrayEmulation =
-                        MobileGL::MG_Config::QuirkOverride::ForceOff;
-                }
-                ~ScopedEmulationOff() { MobileGL::MG_Config::Features.ViewportArrayEmulation = saved; }
-                MobileGL::MG_Config::QuirkOverride saved;
-            };
+            // The switch comes from the ENVIRONMENT, and this case runs only in a process that
+            // was started with it off. It used to write MG_Config::Features directly, which is
+            // not available to it any more: on Android this module links the shipping
+            // libMobileGL.so - so that the on-device run validates the real artifact - and that
+            // library exports no such symbol. The process-wide variable is also the more honest
+            // spelling of the control, since it is the one a developer chasing this failure
+            // would actually set. CMakeLists.txt registers the NoViewportArrayEmulation. ctest
+            // entry for it, so the control still runs in every ctest run; anywhere else - the
+            // ambient ctest entries, or the binary run straight from a device shell - the
+            // emulation is on and this case skips.
+            if (AmbientQuirkFromEnvironment("MOBILEGL_FORCE_VIEWPORT_ARRAY_EMULATION") != AmbientQuirk::Off) {
+                GTEST_SKIP() << "this is the negative control for the emulation and needs it off for the "
+                                "whole process; the NoViewportArrayEmulation. ctest entry runs it with "
+                                "MOBILEGL_FORCE_VIEWPORT_ARRAY_EMULATION=0";
+            }
 
             IntTarget target = MakeIntTarget(kSurfaceSide, kSurfaceSide);
             SetupGridViewports(kCellSize, kCellSize);
 
             GLuint unroutedProgram = 0;
             {
-                const ScopedEmulationOff scopedEmulationOff;
-                // A FRESH program: the emitted ESSL is decided at link time and memoized on a key
-                // that carries this flag, so reusing m_program would just replay the routed build.
+                // A program of its own rather than the fixture's, even though in this process
+                // the fixture's was built unrouted too: the emitted ESSL is decided at link
+                // time and memoized on a key that carries this flag, and building it here keeps
+                // what this case measures independent of when SetUp happened to link.
                 unroutedProgram = BuildProgram(kGridGeometrySource, kIntFragmentSource);
                 ASSERT_NE(unroutedProgram, 0u) << "unrouted program failed to build: " << m_buildLog;
                 glUseProgram(unroutedProgram);

MobileGL/MG_State/GLState/BufferState/BufferObject.cpp

@@ -118,7 +118,13 @@ namespace MobileGL::MG_State::GLState {
         // record that so backends skip uploading the stale shadow bytes.
         m_hasDefinedContent = (data != nullptr) || size == 0;
         m_isImmutableStorage = false;
-        m_storageFlags = 0;
+        // GL 4.6 core 6.2 defines glBufferData as glBufferStorage with
+        // DYNAMIC_STORAGE_BIT | MAP_READ_BIT | MAP_WRITE_BIT, so GL_BUFFER_STORAGE_FLAGS has to
+        // report those three afterwards. Reporting 0 - the value that belongs to a buffer whose
+        // store has never been specified - told an application that a perfectly writable
+        // glBufferData buffer accepted neither glBufferSubData nor a map. Only the IMMUTABLE flag
+        // distinguishes the two cases, and it is cleared just above.
+        m_storageFlags = GL_DYNAMIC_STORAGE_BIT | GL_MAP_READ_BIT | GL_MAP_WRITE_BIT;
         NotifyRespecify();
     }
 

MobileGL/MG_State/GLState/BufferState/BufferState.h

@@ -19,7 +19,14 @@ namespace MobileGL::MG_State::GLState {
                 BufferTarget::DrawIndirect, BufferTarget::Parameter, BufferTarget::ShaderStorage);
     constexpr const auto BufferBindPointTargets = ToArray(BufferTarget::Uniform, BufferTarget::TransformFeedback,
                                                           BufferTarget::AtomicCounter, BufferTarget::ShaderStorage);
-    constexpr SizeT BufferBindingPointCount = 36;
+    // How many indexed binding points each of BufferBindPointTargets gets. 84 is the GL 4.5 core
+    // minimum for GL_MAX_UNIFORM_BUFFER_BINDINGS (table 23.64) and this array is the capacity
+    // that limit is clamped against - at 36 the clamp in GL_Getter was degenerate (lo == hi) and
+    // no application could ever be told about, or bind to, a binding point past the 36th. The
+    // other three targets advertise their own, smaller ceilings out of
+    // GetIndexedBufferQueryPointCount, so widening this does not widen what they promise; it only
+    // costs the unused tail of three arrays.
+    constexpr SizeT BufferBindingPointCount = 84;
 
     class BufferState {
     public:

MobileGL/MG_State/GLState/Core.cpp

@@ -267,6 +267,13 @@ namespace MobileGL::MG_State {
             return m_textureState.CreateTextureObject(index, target);
         }
 
+        const SharedPtr<ITextureObject>& GLContext::CreateTextureViewObject(
+            Uint index, TextureTarget target, const SharedPtr<ITextureObject>& storageOwner, Uint minLevel,
+            Uint numLevels, Uint minLayer, Uint numLayers) {
+            return m_textureState.CreateTextureViewObject(index, target, storageOwner, minLevel, numLevels, minLayer,
+                                                          numLayers);
+        }
+
         void GLContext::MarkTextureObjectForDeletion(Uint index) {
             // GL 3.3 core 4.4.2: deleting a texture whose image is attached to the framebuffer
             // that is currently bound acts as if FramebufferTexture* had been called with texture
@@ -643,6 +650,12 @@ namespace MobileGL::MG_State {
             // a graphics program carrying a compute module, which Adreno 830 does not reject
             // from vkCreateGraphicsPipelines - it SIGSEGVs inside it.
             Bool anyStage = false;
+            // Which stages the composite ACTUALLY got a shader for. Not the same question as
+            // "which stages have a stage program bound": one program bound with
+            // GL_ALL_SHADER_BITS occupies every slot while contributing a shader to only the
+            // stages it was linked with. The transform-feedback capture stage is chosen off this,
+            // because it has to be the stage that will exist in the composite's own link.
+            Bool compositeHasStage[ProgramPipelineObject::kGraphicsStageCount] = {};
             for (SizeT stage = 0; stage < ProgramPipelineObject::kGraphicsStageCount; ++stage) {
                 const auto& stageProgram = pipeline->GetStageProgram(static_cast<ShaderStage>(stage));
                 if (!stageProgram) continue;
@@ -658,9 +671,57 @@ namespace MobileGL::MG_State {
                     if (!ref.shader || static_cast<SizeT>(ref.shader->GetShaderStage()) != stage) continue;
                     composite->AttachShaderWithPinnedLinkInput(ref);
                     anyStage = true;
+                    compositeHasStage[stage] = true;
                 }
             }
             if (!anyStage) return nullProgram;
+            // Transform feedback captures the output of the LAST vertex-processing stage
+            // (GL 4.6 core 11.1.2.1), and glTransformFeedbackVaryings is per-PROGRAM state that
+            // only the stage program carrying that stage can have been given. The composite is
+            // assembled out of the stage programs' shaders and inherits none of their
+            // GL-thread-owned state, so without this it links with an empty capture list and
+            // glBeginTransformFeedback rejects the draw with INVALID_OPERATION ("the program has
+            // no transform feedback varyings") even though glValidateProgramPipeline had passed.
+            //
+            // TWO RULES, both easy to get subtly wrong and both load-bearing:
+            //
+            // (1) THE LINKED LIST, NOT THE PENDING REQUEST. glTransformFeedbackVaryings does not
+            //     take effect until the program's next link (GL 4.6 core 7.3/11.1.2.1), and it
+            //     deliberately bumps no version - so a request written after the stage program's
+            //     last link is invisible to the composite cache's signature yet would be picked up
+            //     by the next rebuild, making the capture list depend on whether some unrelated
+            //     event happened to invalidate the cache. Worse, a name that is not an output of
+            //     the capture stage fails the composite's OWN link, and a failed composite makes
+            //     every draw through the pipeline report INVALID_OPERATION. Reading the LINKED
+            //     snapshot removes the whole class: linked state only moves at a link, and a link
+            //     is exactly what ComputeDrawProgramSignature's per-stage link version tracks, so
+            //     the existing cache key is sufficient by construction.
+            //     GetTransformFeedbackInterfaceNames() is the right accessor rather than the
+            //     resolved xfbVaryings: it is the request as that link consumed it, pseudo-varyings
+            //     (gl_NextBuffer / gl_SkipComponentsN) included, which is what re-issuing it needs.
+            //
+            // (2) THE FIRST STAGE THAT EXISTS, not the first with something to capture. This is
+            //     the rule ProgramLinkTask::ResolveTransformFeedbackVaryings applies (it breaks on
+            //     getIntermediate(stage) != nullptr), and the two MUST agree: this loop picks
+            //     WHOSE list, the link task picks WHICH stage's outputs the names resolve against.
+            //     Skipping a geometry stage that has no capture list and installing the vertex
+            //     stage's instead made them disagree, and the composite then resolved a vertex
+            //     program's names against the geometry intermediate - capturing where GL says it
+            //     must not, or failing the link and killing every draw. A capture stage with an
+            //     empty list is not a reason to look further down: it is the answer, and
+            //     glBeginTransformFeedback's INVALID_OPERATION is the correct consequence.
+            for (const ShaderStage captureStage:
+                 {ShaderStage::Geometry, ShaderStage::TessEval, ShaderStage::Vertex}) {
+                if (!compositeHasStage[static_cast<SizeT>(captureStage)]) continue;
+                const auto& captureProgram = pipeline->GetStageProgram(captureStage);
+                if (!captureProgram) continue;
+                const auto& linkedNames = captureProgram->GetTransformFeedbackInterfaceNames();
+                if (!linkedNames.empty()) {
+                    composite->SetTransformFeedbackVaryings(Vector<String>(linkedNames),
+                                                            captureProgram->GetTransformFeedbackBufferMode());
+                }
+                break;
+            }
             // A pipeline with no fragment stage still rasterises, so the default fragment
             // shader is wanted here even though the separable stage programs never get one.
             composite->Link(true);
@@ -805,6 +866,22 @@ namespace MobileGL::MG_State {
             m_renderState.SetPatchVertices(vertices);
         }
 
+        void GLContext::SetPatchDefaultOuterLevel(const FloatVec4& levels) {
+            m_renderState.SetPatchDefaultOuterLevel(levels);
+        }
+
+        const FloatVec4& GLContext::GetPatchDefaultOuterLevel() const {
+            return m_renderState.GetPatchDefaultOuterLevel();
+        }
+
+        void GLContext::SetPatchDefaultInnerLevel(const FloatVec2& levels) {
+            m_renderState.SetPatchDefaultInnerLevel(levels);
+        }
+
+        const FloatVec2& GLContext::GetPatchDefaultInnerLevel() const {
+            return m_renderState.GetPatchDefaultInnerLevel();
+        }
+
         Uint GLContext::GetPatchVertices() const {
             return m_renderState.GetPatchVertices();
         }
@@ -825,6 +902,26 @@ namespace MobileGL::MG_State {
             return m_renderState.GetPolygonOffsetUnits();
         }
 
+        void GLContext::SetPolygonOffsetClamped(Float factor, Float units, Float clamp) {
+            m_renderState.SetPolygonOffsetClamped(factor, units, clamp);
+        }
+
+        Float GLContext::GetPolygonOffsetClamp() const {
+            return m_renderState.GetPolygonOffsetClamp();
+        }
+
+        void GLContext::SetClipControl(GLenum origin, GLenum depth) {
+            m_renderState.SetClipControl(origin, depth);
+        }
+
+        GLenum GLContext::GetClipOrigin() const {
+            return m_renderState.GetClipOrigin();
+        }
+
+        GLenum GLContext::GetClipDepthMode() const {
+            return m_renderState.GetClipDepthMode();
+        }
+
         void GLContext::SetCapability(CapabilityInput cap, Bool enabled) {
             m_renderState.SetCapability(cap, enabled);
         }
@@ -1002,6 +1099,14 @@ namespace MobileGL::MG_State {
             return m_renderState.GetSampleMaskValue();
         }
 
+        void GLContext::SetMinSampleShadingValue(Float value) {
+            m_renderState.SetMinSampleShadingValue(value);
+        }
+
+        Float GLContext::GetMinSampleShadingValue() const {
+            return m_renderState.GetMinSampleShadingValue();
+        }
+
         void GLContext::SetPixelStoreParam(PixelStoreParam param, Int value) {
             m_renderState.SetPixelStoreParam(param, value);
         }

MobileGL/MG_State/GLState/Core.h

@@ -111,6 +111,11 @@ namespace MobileGL {
                 // Per-target default texture object (name 0); see TextureState::GetDefaultTextureObject.
                 const SharedPtr<ITextureObject>& GetDefaultTextureObject(TextureTarget target) const;
                 const SharedPtr<ITextureObject>& CreateTextureObject(Uint index, TextureTarget target);
+                // See TextureState::CreateTextureViewObject (glTextureView, GL 4.6 core 8.18).
+                const SharedPtr<ITextureObject>& CreateTextureViewObject(Uint index, TextureTarget target,
+                                                                         const SharedPtr<ITextureObject>& storageOwner,
+                                                                         Uint minLevel, Uint numLevels, Uint minLayer,
+                                                                         Uint numLayers);
                 void MarkTextureObjectForDeletion(Uint index);
                 TextureUnit& GetTextureUnitObject(Int unit);
                 ImageTextureBinding& GetImageTextureBinding(Int unit);
@@ -208,9 +213,18 @@ namespace MobileGL {
                 Float GetPointSize() const;
                 void SetPatchVertices(Uint vertices);
                 Uint GetPatchVertices() const;
+                void SetPatchDefaultOuterLevel(const FloatVec4& levels);
+                const FloatVec4& GetPatchDefaultOuterLevel() const;
+                void SetPatchDefaultInnerLevel(const FloatVec2& levels);
+                const FloatVec2& GetPatchDefaultInnerLevel() const;
                 void SetPolygonOffset(Float factor, Float units);
+                void SetPolygonOffsetClamped(Float factor, Float units, Float clamp);
                 Float GetPolygonOffsetFactor() const;
                 Float GetPolygonOffsetUnits() const;
+                Float GetPolygonOffsetClamp() const;
+                void SetClipControl(GLenum origin, GLenum depth);
+                GLenum GetClipOrigin() const;
+                GLenum GetClipDepthMode() const;
                 void SetHint(GLenum target, GLenum mode);
                 GLenum GetHint(GLenum target) const;
                 void SetPointFadeThresholdSize(Float size);
@@ -271,6 +285,8 @@ namespace MobileGL {
                 Bool GetSampleCoverageInvert() const;
                 void SetSampleMaskValue(Uint32 mask);
                 Uint32 GetSampleMaskValue() const;
+                void SetMinSampleShadingValue(Float value);
+                Float GetMinSampleShadingValue() const;
                 void SetPixelStoreParam(PixelStoreParam param, Int value);
                 Int GetPixelStoreParam(PixelStoreParam param) const;
                 PixelStoreParameters GetPixelStoreParameters(Bool isUnpack) const;

MobileGL/MG_State/GLState/ProgramState/ProgramLinkTask.cpp

@@ -10,6 +10,7 @@
 
 #include <MG_State/GLState/ProgramState/ProgramTranslationCache.h>
 
+#include <MG_State/GLState/BufferState/BufferState.h>
 #include <MG_State/GLState/VertexArrayState/VertexArrayObject.h>
 #include <MG_Util/Async/ShaderCompilePool.h>
 #include <MG_Util/Converters/GLToStr/GLEnumConverter.h>
@@ -29,13 +30,11 @@ namespace {
     // capacity, which is also the width of the Uint32 masks backends build from it.
     static MobileGL::Int GetReflectionVertexAttribLimit(
         const MobileGL::MG_Util::ShaderTranspiler::CompileEnv& env) {
-        constexpr MobileGL::Int capacity =
-            static_cast<MobileGL::Int>(MobileGL::MG_State::GLState::VertexArrayObject::MAX_VERTEX_ATTRIBS);
-        if (!env.HasBackend()) return capacity;
-
-        const MobileGL::Int backendLimit = env.params.MaxVertexAttribs;
-        if (backendLimit <= 0) return capacity;
-        return std::min(backendLimit, capacity);
+        // One shared definition with glGetIntegerv(GL_MAX_VERTEX_ATTRIBS) and with
+        // BuildTBuiltInResource's gl_MaxVertexAttribs - the three used to carry three copies of
+        // this formula and glslang's copy was a hardcoded 64.
+        return MobileGL::MG_Util::ShaderTranspiler::ResolveMaxVertexAttribs(env.HasBackend(),
+                                                                            env.params.MaxVertexAttribs);
     }
 
     // Everything the post-link query surface ever asks a glslang::TType, flattened into a
@@ -622,12 +621,21 @@ namespace MobileGL::MG_State::GLState {
         // mapper's collect callback is the last point at which a resource's qualifier still
         // says what the SHADER declared rather than what glslang assigned, so both captures
         // have to be taken from inside the link. See TMglGlslIoResolver::reserverResourceSlot.
+        // The binding-range rule (GLSL 4.30 4.4.5): its ceilings in, and the first violation the
+        // resolver finds out. Enforced at the link because mapIO's collect callback is the last
+        // point at which a resource's qualifier still says what the SHADER declared - see
+        // TMglGlslIoResolver::CheckDeclaredBindingRange.
+        String resourceBindingViolation;
         ProgramAttrib attrib{.shaders = Move(shaders),
                              .explicitVertexInLocations = in.explicitAttribLocations,
                              .explicitFragmentOutLocations = in.explicitFragDataLocation,
                              .explicitFragmentOutIndices = in.explicitFragDataIndex,
                              .explicitOpaqueUniformBindings = &artifacts.explicitOpaqueUniformBindings,
-                             .storageBlocksWithoutBinding = &artifacts.storageBlocksWithoutBinding};
+                             .storageBlocksWithoutBinding = &artifacts.storageBlocksWithoutBinding,
+                             .uniformBlocksWithoutBinding = &artifacts.uniformBlocksWithoutBinding,
+                             .resourceBindingLimits = in.env ? ResolveResourceBindingLimits(*in.env)
+                                                             : MG_Util::ShaderTranspiler::ResourceBindingLimits{},
+                             .resourceBindingViolation = &resourceBindingViolation};
 
         MGLOG_D("ProgramObject %u: Calling ShaderCompiler::LinkProgram", in.externalIndex);
         auto result = ShaderCompiler::LinkProgram(attrib);
@@ -671,9 +679,14 @@ namespace MobileGL::MG_State::GLState {
             return;
         }
 
-        // GL_GEOMETRY_INPUT_TYPE. A draw's primitive type has to be compatible with it
-        // (GL 4.6 core 11.3.1), so it is resolved for every link, not only a capturing one.
+        // The geometry stage's link properties. GL_GEOMETRY_INPUT_TYPE is load-bearing beyond the
+        // query surface - a draw's primitive type has to be compatible with it (GL 4.6 core
+        // 11.3.1) - so this block runs for every link, not only a capturing one. The other three
+        // are pure glGetProgramiv answers that previously had no source at all.
         artifacts.gsInputPrimitive = GL_NONE;
+        artifacts.gsOutputPrimitive = GL_NONE;
+        artifacts.gsMaxVertices = 0;
+        artifacts.gsInvocations = 0;
         if (const glslang::TIntermediate* gs = artifacts.program->getIntermediate(EShLangGeometry)) {
             switch (gs->getInputPrimitive()) {
             case glslang::ElgPoints: artifacts.gsInputPrimitive = GL_POINTS; break;
@@ -683,6 +696,77 @@ namespace MobileGL::MG_State::GLState {
             case glslang::ElgTrianglesAdjacency: artifacts.gsInputPrimitive = GL_TRIANGLES_ADJACENCY; break;
             default: break;
             }
+            switch (gs->getOutputPrimitive()) {
+            case glslang::ElgPoints: artifacts.gsOutputPrimitive = GL_POINTS; break;
+            case glslang::ElgLineStrip: artifacts.gsOutputPrimitive = GL_LINE_STRIP; break;
+            case glslang::ElgTriangleStrip: artifacts.gsOutputPrimitive = GL_TRIANGLE_STRIP; break;
+            default: break;
+            }
+            // glslang leaves both at TQualifier::layoutNotSet (-1) when the shader declared no
+            // such layout, and `invocations` defaults to one per GLSL 4.60 4.4.2.2 - so clamp
+            // rather than forward, or GL_GEOMETRY_SHADER_INVOCATIONS reports the sentinel.
+            artifacts.gsMaxVertices = std::max(gs->getVertices(), 0);
+            artifacts.gsInvocations = std::max(gs->getInvocations(), 1);
+        }
+
+        // The tessellation evaluation stage's link properties, GL 4.6 core table 23.35: the
+        // primitive generator's mode, spacing, winding and point mode. (The control stage's
+        // output patch size is captured below, together with the limit check that goes with it.)
+        artifacts.tessGenMode = GL_NONE;
+        artifacts.tessGenSpacing = GL_NONE;
+        artifacts.tessGenVertexOrder = GL_NONE;
+        artifacts.tessGenPointMode = false;
+        if (const glslang::TIntermediate* tes = artifacts.program->getIntermediate(EShLangTessEvaluation)) {
+            switch (tes->getInputPrimitive()) {
+            case glslang::ElgTriangles: artifacts.tessGenMode = GL_TRIANGLES; break;
+            case glslang::ElgQuads: artifacts.tessGenMode = GL_QUADS; break;
+            case glslang::ElgIsolines: artifacts.tessGenMode = GL_ISOLINES; break;
+            default: break;
+            }
+            // GLSL 4.60 4.4.2.3: equal_spacing and ccw are the defaults, which is what an unset
+            // qualifier means here.
+            switch (tes->getVertexSpacing()) {
+            case glslang::EvsFractionalEven: artifacts.tessGenSpacing = GL_FRACTIONAL_EVEN; break;
+            case glslang::EvsFractionalOdd: artifacts.tessGenSpacing = GL_FRACTIONAL_ODD; break;
+            default: artifacts.tessGenSpacing = GL_EQUAL; break;
+            }
+            switch (tes->getVertexOrder()) {
+            case glslang::EvoCw: artifacts.tessGenVertexOrder = GL_CW; break;
+            default: artifacts.tessGenVertexOrder = GL_CCW; break;
+            }
+            artifacts.tessGenPointMode = tes->getPointMode();
+        }
+
+        // GL_TESS_CONTROL_OUTPUT_VERTICES, i.e. the `layout(vertices = N) out` the control stage
+        // declared, and the limit that goes with it.
+        //
+        // GL 4.6 core 11.2.1.1: the LINK fails when N is greater than MAX_PATCH_VERTICES. Nothing
+        // enforced it - glslang's layout handling only rejects N <= 0 (ParseHelper.cpp "must be
+        // greater than 0") and carries maxPatchVertices in TBuiltInResource purely so
+        // gl_MaxPatchVertices can expand from it, exactly the gap ValidateImageUniformLimits
+        // documents for image uniforms. Checked at LINK rather than at compile on purpose: the CTS
+        // requires the offending shader to COMPILE ("Compilation passed as allowed") and only the
+        // link to fail, and turning it into a parse error would newly break an application that
+        // compiles such a shader and never links it.
+        //
+        // The limit is the one glGetIntegerv answers (GL_Getter.cpp reads the same
+        // DynamicBackendParameters field), so the advertised number and the enforced number cannot
+        // drift apart.
+        artifacts.tcsOutputVertices = 0;
+        if (const glslang::TIntermediate* tcs = artifacts.program->getIntermediate(EShLangTessControl)) {
+            artifacts.tcsOutputVertices = static_cast<Int>(tcs->getVertices());
+            if (artifacts.tcsOutputVertices > env.params.MaxPatchVertices) {
+                artifacts.linkStatus = false;
+                // Same invariant as the compute local-size gate above: a rejected link leaves no
+                // TProgram behind for a query surface to find.
+                artifacts.program.reset();
+                artifacts.infoLog = std::format(
+                    "Tessellation control shader declares an output patch of {} vertices, more than the {} "
+                    "GL_MAX_PATCH_VERTICES allows.",
+                    artifacts.tcsOutputVertices, env.params.MaxPatchVertices);
+                DeferLog(std::format("ProgramObject {}: Link failed - {}", in.externalIndex, artifacts.infoLog));
+                return;
+            }
         }
 
         // ---- everything below this line up to GenerateSpirv() is the GL query surface ----
@@ -1103,6 +1187,20 @@ namespace MobileGL::MG_State::GLState {
                         in.externalIndex, uniform.name.c_str());
                 continue;
             }
+            // The gl_NumSamples stand-in InjectNumSamplesBuiltinShim declared. It is a driver
+            // uniform, not the application's: gl_NumSamples is a BUILT-IN, so a conformant
+            // implementation reports nothing for it in GL_ACTIVE_UNIFORMS, glGetActiveUniform or
+            // glGetUniformLocation, and nothing may write it through glUniform* either. Filtering
+            // it here does both, and costs it no storage: BuildGlobalUboRouting takes its offset
+            // from the SPIR-V metadata by name, not from the GL location space.
+            if (isGlobalUboMember(uniform) &&
+                uniform.name == MG_Util::ShaderTranspiler::NUM_SAMPLES_UNIFORM_NAME) {
+                artifacts.usesReservedNumSamples = true;
+                MGLOG_D("ProgramObject %u: Reflection - reserved gl_NumSamples stand-in '%s' hidden from the GL "
+                        "uniform surface",
+                        in.externalIndex, uniform.name.c_str());
+                continue;
+            }
             if (isBufferVariable(uniform)) {
                 MGLOG_D("ProgramObject %u: Reflection - buffer variable '%s' filtered from the GL uniform "
                         "surface",
@@ -1573,9 +1671,46 @@ namespace MobileGL::MG_State::GLState {
             // (DirectGLES.cpp / UniformManager.cpp), all 14 elements also read the same
             // buffer. This is the rule the storage-block path in ProgramInterface.cpp
             // already applies, and whose comment there claims uniform blocks follow.
-            const Int declaredBinding = ubo.getBinding();
+            //
+            // "Declared" cannot be read back off the reflection, though. MobileGL asks glslang
+            // to auto-map bindings, so mapIO writes an invented one into every block's
+            // qualifier before reflection ever runs and ubo.getBinding() is never negative;
+            // worse, glslang packs uniform blocks into the SAME slot space as samplers and
+            // images (setEnvClient(EShClientVulkan) leaves spvVersion.openGl at 0, so
+            // TDefaultGlslIoResolver::resolveBinding keys every resource kind on set 0), so a
+            // block declared after an unbound image gets 1. GL 4.6 core 7.6.2 says an
+            // unqualified block reports ZERO. The set below is the shader's own answer,
+            // captured during mapIO while the qualifier still meant it - the same mechanism
+            // SeedDefaultStorageBlockBindings uses for storage blocks, and the aliasing at 0
+            // that results is GL's, not a bug: unqualified blocks collide there until the
+            // application rebinds them.
+            //
+            // Only this GL-visible binding POINT changes. The backends' descriptor lookups run
+            // off glslang's assignment through uniformBlockIndexByBinding, which is untouched.
+            const String blockTypeName = StripArrayElementSuffix(ubo.name);
+            const Int declaredBinding =
+                artifacts.uniformBlocksWithoutBinding.contains(blockTypeName) ? 0 : ubo.getBinding();
             artifacts.uniformBlockBinding[i] =
                 declaredBinding < 0 ? declaredBinding : declaredBinding + BlockArrayElement(ubo.name);
+            // The second way a binding reaches the state layer's indexed-binding array, and the
+            // one glUniformBlockBinding's new bound cannot see. glslang does not range-check a
+            // uniform block's layout(binding = N) against anything - TBuiltInResource has no
+            // maxUniformBufferBindings field at all, and ParseHelper bounds only samplers and
+            // atomic counters - so `layout(binding = 5000) uniform Blk {...}` compiled and linked
+            // clean and then had both backends subscript the array at 5000 on the first draw.
+            // Stated against the same ceiling glGetIntegerv(GL_MAX_UNIFORM_BUFFER_BINDINGS)
+            // advertises; an instance array whose LAST element passes it is a link error even
+            // though its base fits, same rule as the explicit-location check above.
+            if (artifacts.uniformBlockBinding[i] >=
+                static_cast<Int>(MG_State::GLState::BufferBindingPointCount)) {
+                artifacts.infoLog =
+                    std::format("Uniform block '{}' declares binding {}, which is not less than "
+                                "GL_MAX_UNIFORM_BUFFER_BINDINGS ({}).",
+                                ubo.name, artifacts.uniformBlockBinding[i],
+                                static_cast<Int>(MG_State::GLState::BufferBindingPointCount));
+                ProgramObject::ResetLinkArtifacts(artifacts);
+                return false;
+            }
             MGLOG_D("ProgramObject %u: Reflection - UBO[%d] name='%s' size=%u binding=%d", in.externalIndex, i,
                     ubo.name.c_str(), ubo.size, ubo.getBinding());
         }

MobileGL/MG_State/GLState/ProgramState/ProgramObject.cpp

@@ -369,6 +369,7 @@ namespace MobileGL::MG_State::GLState {
         // link, so a stale set would otherwise default a block the new sources do declare a
         // binding for.
         artifacts.storageBlocksWithoutBinding.clear();
+        artifacts.uniformBlocksWithoutBinding.clear();
         artifacts.attribs.clear();
         artifacts.attribTypes.clear();
         artifacts.activeUniformCount = 0;
@@ -491,6 +492,10 @@ namespace MobileGL::MG_State::GLState {
         // time, for anything cached during the pending window itself.)
         ++m_backendStateVersion;
         BumpLinkObservableVersions();
+        // The separable flag takes effect HERE, at the link, and nowhere else (GL 4.6 core 7.3).
+        // Latched before the early-outs below so a link that fails still counts as a link -
+        // what must not update it is a link that never happened at all.
+        m_linkedSeparable = m_separable;
         // A whole-struct reset, unlike ResetLinkArtifacts(): during the pending window this
         // is what every gated reader sees, so it has to be the complete "not linked" state -
         // including the fields ResetLinkArtifacts deliberately preserves for its own callers.
@@ -535,6 +540,33 @@ namespace MobileGL::MG_State::GLState {
         task->in.explicitFragDataIndex = m_explicitFragDataIndex;
         task->in.requestedXfbVaryings = m_requestedXfbVaryings;
         task->in.requestedXfbBufferMode = m_requestedXfbBufferMode;
+        // ARB_gl_spirv: a program built from SPIR-V declares its transform feedback through
+        // XfbBuffer/XfbStride/Offset DECORATIONS, and glTransformFeedbackVaryings has no effect on
+        // it at all. glSpecializeShader translated those decorations into the equivalent name
+        // request (ShaderCompiler::SpecializeAndDecompileSpirvModule), and this is where it enters
+        // the link - so everything downstream, the frontend packer and both backends, sees one
+        // declaration form instead of two.
+        //
+        // The capture stage is the LAST vertex-processing stage the program has, which is the same
+        // rule ProgramLinkTask::ResolveTransformFeedbackVaryings resolves the names against. The
+        // application's own request wins if it made one: that can only happen on a mixed program,
+        // which is not a shape ARB_gl_spirv defines, and honouring what the application explicitly
+        // asked for is the safer of the two readings.
+        if (task->in.requestedXfbVaryings.empty()) {
+            for (const ShaderStage captureStage:
+                 {ShaderStage::Geometry, ShaderStage::TessEval, ShaderStage::Vertex}) {
+                Bool stagePresent = false;
+                for (const auto& shader : m_shaders) {
+                    if (!shader || shader->GetShaderStage() != captureStage) continue;
+                    stagePresent = true;
+                    if (shader->GetSpirvXfbVaryings().empty()) continue;
+                    task->in.requestedXfbVaryings = shader->GetSpirvXfbVaryings();
+                    task->in.requestedXfbBufferMode = shader->GetSpirvXfbBufferMode();
+                    break;
+                }
+                if (stagePresent) break;
+            }
+        }
         task->in.maxFragmentOutputColorNumber = m_maxFragmentOutputColorNumber;
 
         Vector<SharedPtr<ShaderCompileTask>> deps;

MobileGL/MG_State/GLState/ProgramState/ProgramObject.h

@@ -787,6 +787,33 @@ namespace MobileGL::MG_State::GLState {
         void MarkUBOContentDirty() const {
             if (++m_uboContentVersion == ~0u) m_uboContentVersion = 0;
         }
+
+        // ---- the reserved gl_NumSamples stand-in (ShaderTranspiler::NUM_SAMPLES_UNIFORM_NAME) ----
+        //
+        // PHASE A: answerable without joining the SPIR-V job, which is what lets the draw path ask
+        // every program this question and pay nothing for the overwhelming majority that say no.
+        Bool UsesReservedNumSamples() const { return Artifacts().usesReservedNumSamples; }
+
+        // Publishes `samples` into the global-UBO shadow. Returns false when there is nowhere to
+        // put it - no shim in this program, no SPIR-V (a cancelled phase B), or the optimizer
+        // dropped the member because nothing read it after all - all of which are ordinary states,
+        // not errors. A value-identical write is dropped without bumping the content version, so a
+        // steady stream of draws into one framebuffer does not force a re-upload per draw.
+        Bool WriteReservedNumSamples(Int samples) {
+            if (!UsesReservedNumSamples()) return false;
+            SpirvArtifacts& spirv = Spirv();
+            const Uint offset = spirv.reservedNumSamplesOffset;
+            if (offset == kInvalidUniformOffset) return false;
+            if (static_cast<SizeT>(offset) + sizeof(Int) > spirv.globalUboScratch.size()) return false;
+
+            Uint8* const slot = spirv.globalUboScratch.data() + offset;
+            Int current = 0;
+            Memcpy(&current, slot, sizeof(Int));
+            if (current == samples) return true;
+            Memcpy(slot, &samples, sizeof(Int));
+            MarkUBOContentDirty();
+            return true;
+        }
         // ---- glUniform* inside the phase-A -> phase-B window ----
         //
         // True while the program is fully linked and fully queryable but its uniform shadow's
@@ -888,6 +915,14 @@ namespace MobileGL::MG_State::GLState {
         // subset of the stages of a program pipeline. Only takes effect on the next link,
         // which is why it is plain state here rather than something Link() consults.
         Bool GetSeparable() const { return m_separable; }
+        // What GL_PROGRAM_SEPARABLE actually reports, and what glUseProgramStages actually
+        // requires: the value the flag held at the program's LAST LINK, not the live flag.
+        // GL 4.6 core 7.3 - "the flag takes effect the next time the program is linked" - so a
+        // program that was told to be separable and then never linked is still NOT separable,
+        // which is precisely what es31cSeparateShaderObjsTests's PipelineApi and CreateShadProgApi
+        // assert. The live flag stays available as GetSeparable() for glGetProgramiv's sibling
+        // state and for the next link to latch.
+        Bool GetLinkedSeparable() const { return m_linkedSeparable; }
         void SetSeparable(Bool separable) {
             m_separable = separable;
             // ---- arming the uniform-write tracking latch ----
@@ -1024,8 +1059,10 @@ namespace MobileGL::MG_State::GLState {
         Uint32 GetBlockBindingVersion() const { return m_blockBindingVersion; }
 
         // Set by glUniformBlockBinding. The vector is seeded at link with each block's DECLARED
-        // binding (layout(binding=N), else -1), so an untouched program already reports what its
-        // shaders asked for.
+        // binding (layout(binding=N)), and with GL's default of 0 for a block that declared none
+        // - which the reflection cannot tell apart on its own, so the seeder consults
+        // uniformBlocksWithoutBinding. Either way an untouched program already reports what GL
+        // says it should.
         void SetUniformBlockBinding(Uint index, Uint binding) {
             if (index >= Artifacts().uniformBlockBinding.size() || Artifacts().uniformBlockBinding[index] == static_cast<Int>(binding)) {
                 return;
@@ -1287,8 +1324,21 @@ namespace MobileGL::MG_State::GLState {
             // binding from an invented one - and, unlike the per-shader lexer this replaced,
             // sees the declaration with its macros expanded.
             std::set<String> storageBlocksWithoutBinding;
+            // The same list for UNIFORM blocks, and it is needed for the same reason: glslang's
+            // auto-mapper assigns every uniform block a binding whether or not the shader asked
+            // for one, so uniformBlockBinding below cannot tell "declared 1" from "invented 1".
+            // GL 4.6 core 7.6.2 requires an unqualified block to report ZERO.
+            std::set<String> uniformBlocksWithoutBinding;
 
             Uint activeUniformCount = 0;
+            // This program's fragment stage read gl_NumSamples, so the source pipeline lowered it
+            // onto the reserved default-block uniform (ShaderTranspiler::NUM_SAMPLES_UNIFORM_NAME)
+            // and the draw path owes it the draw framebuffer's sample count before every draw.
+            //
+            // PHASE A on purpose, even though the byte offset it needs is phase-B output: the
+            // gate has to be answerable without joining the SPIR-V job, or every draw of every
+            // program would pay a join to discover it has nothing to write.
+            Bool usesReservedNumSamples = false;
             Uint maxUniformLocation = 0;
             Int uniformNameMaxLength = 0;
             Int attribInNameMaxLength = 0;
@@ -1310,6 +1360,27 @@ namespace MobileGL::MG_State::GLState {
             Vector<Uint32> gsStripTriangles;
             Bool gsStripCaptureFixup = false;
             GLenum gsInputPrimitive = GL_NONE;
+            // GL_TESS_CONTROL_OUTPUT_VERTICES: the `layout(vertices = N) out` of the linked
+            // tessellation control stage, or 0 when the program has none. Checked against
+            // GL_MAX_PATCH_VERTICES at link (GL 4.6 core 11.2.1.1).
+            Int tcsOutputVertices = 0;
+            // The rest of the geometry stage's link properties, and the tessellation evaluation
+            // stage's. Every one of these is a glGetProgramiv answer that had no source at all:
+            // the query surface listed the geometry pnames only to fall through to
+            // GL_INVALID_ENUM, and the GL_TESS_GEN_* pnames were not mentioned anywhere. They
+            // come from the linked intermediates for the same reason gsInputPrimitive and
+            // tcsOutputVertices do - glslang has already merged the compilation units' layout
+            // qualifiers and diagnosed contradictions, so the linked program is the thing that
+            // knows.
+            GLenum gsOutputPrimitive = GL_NONE;
+            Int gsMaxVertices = 0;
+            Int gsInvocations = 0;
+            // The tessellation evaluation stage's layout: GL_QUADS / GL_TRIANGLES / GL_ISOLINES,
+            // GL_EQUAL / GL_FRACTIONAL_EVEN / GL_FRACTIONAL_ODD, GL_CW / GL_CCW, and point mode.
+            GLenum tessGenMode = GL_NONE;
+            GLenum tessGenSpacing = GL_NONE;
+            GLenum tessGenVertexOrder = GL_NONE;
+            Bool tessGenPointMode = false;
             GLenum xfbBufferMode = GL_INTERLEAVED_ATTRIBS;
             Int xfbVaryingNameMaxLength = 0;
             Bool xfbNeedsScatteredCapture = false;
@@ -1337,6 +1408,11 @@ namespace MobileGL::MG_State::GLState {
             // kInvalidUniformOffset. Sized maxUniformLocation + 1 by the routing pass.
             Vector<Uint> uniformOffsets;
             Vector<Uint8> globalUboScratch;
+            // Byte offset of the reserved gl_NumSamples stand-in inside globalUboScratch, or
+            // kInvalidUniformOffset. Taken by NAME from the SPIR-V metadata rather than through
+            // uniformOffsets, because the member has no GL location at all: the link task keeps
+            // it out of the GL-visible uniform index space so no application can see or write it.
+            Uint reservedNumSamplesOffset = kInvalidUniformOffset;
             // False for a program whose SPIR-V was never produced (phase B cancelled at
             // teardown or by a relink) or whose optimizer run failed. GL has no way to
             // retract a LINK_STATUS it already reported true, so such a program stays
@@ -1460,6 +1536,14 @@ namespace MobileGL::MG_State::GLState {
             m_requestedXfbVaryings = Move(names);
             m_requestedXfbBufferMode = bufferMode;
         }
+        // NO ACCESSOR FOR THE PENDING REQUEST, deliberately. A program pipeline's draw composite
+        // needs the capture list of the stage program it flattens, and the obvious source - what
+        // glTransformFeedbackVaryings last recorded - is the wrong one: that request does not take
+        // effect until the stage program's next link, and it bumps no version, so reading it makes
+        // the composite's capture list depend on when the composite cache happened to be
+        // invalidated. GetTransformFeedbackInterfaceNames() below is the source that is correct
+        // AND cache-safe, because linked state only moves at a link and the composite signature
+        // already keys on the link version. See GLContext::GetProgramForDraw.
         GLenum GetTransformFeedbackBufferMode() const { return Artifacts().xfbBufferMode; }
         SizeT GetTransformFeedbackVaryingCount() const { return Artifacts().xfbVaryings.size(); }
         const XfbVarying* GetTransformFeedbackVarying(SizeT index) const {
@@ -1494,6 +1578,22 @@ namespace MobileGL::MG_State::GLState {
         // GL_LINES_ADJACENCY, GL_TRIANGLES or GL_TRIANGLES_ADJACENCY), or GL_NONE when the
         // program has no geometry stage. Draws must present a compatible primitive type.
         GLenum GetGeometryInputType() const { return Artifacts().gsInputPrimitive; }
+        // GL_GEOMETRY_OUTPUT_TYPE (GL_POINTS, GL_LINE_STRIP or GL_TRIANGLE_STRIP),
+        // GL_GEOMETRY_VERTICES_OUT and GL_GEOMETRY_SHADER_INVOCATIONS of the linked geometry
+        // stage. Meaningless without one - glGetProgramiv raises INVALID_OPERATION there.
+        GLenum GetGeometryOutputType() const { return Artifacts().gsOutputPrimitive; }
+        Int GetGeometryVerticesOut() const { return Artifacts().gsMaxVertices; }
+        Int GetGeometryShaderInvocations() const { return Artifacts().gsInvocations; }
+        // GL_TESS_CONTROL_OUTPUT_VERTICES of the linked tessellation control stage, or 0 when
+        // the program has no such stage. Never greater than GL_MAX_PATCH_VERTICES: a program
+        // that declared more does not link at all (GL 4.6 core 11.2.1.1).
+        Int GetTessControlOutputVertices() const { return Artifacts().tcsOutputVertices; }
+        // GL_TESS_GEN_MODE / _SPACING / _VERTEX_ORDER / _POINT_MODE of the linked tessellation
+        // evaluation stage.
+        GLenum GetTessGenMode() const { return Artifacts().tessGenMode; }
+        GLenum GetTessGenSpacing() const { return Artifacts().tessGenSpacing; }
+        GLenum GetTessGenVertexOrder() const { return Artifacts().tessGenVertexOrder; }
+        Bool GetTessGenPointMode() const { return Artifacts().tessGenPointMode; }
 
         Uint GetExternalIndex() const { return m_externalIndex; }
         // Globally-unique, never-reused id for this program object's lifetime. Unlike the GL
@@ -1619,6 +1719,11 @@ namespace MobileGL::MG_State::GLState {
         Bool m_deleteStatus = false;
         Bool m_binaryRetrievableHint = false;
         Bool m_separable = false;
+        // m_separable as of the last link; see GetLinkedSeparable. Latched by Link() rather than
+        // carried in LinkArtifacts because it is a GL-thread-owned decision made at enqueue time,
+        // not a result the worker computes - and because a FAILED link still latches it, exactly
+        // as a successful one does.
+        Bool m_linkedSeparable = false;
         // Monotone "this program may ever be a pipeline stage" latch; see SetSeparable for why
         // it is a latch and not just m_separable. Outside LinkArtifacts on purpose: a relink
         // clears the write SET, but a program that was separable is still separable after it.

MobileGL/MG_State/GLState/ProgramState/ProgramSpirvTask.cpp

@@ -278,6 +278,7 @@ namespace MobileGL::MG_State::GLState {
 
         artifacts.uniformOffsets.clear();
         artifacts.globalUboScratch.clear();
+        artifacts.reservedNumSamplesOffset = ProgramObject::kInvalidUniformOffset;
         // kInvalidUniformOffset marks locations that end up without global-UBO backing
         // (e.g. the optimizer eliminated every use of the uniform); the fallback pass
         // below gives those locations tail storage so glUniform* always has a target.
@@ -311,6 +312,18 @@ namespace MobileGL::MG_State::GLState {
                     artifacts.globalUboScratch.resize(size);
                 }
                 for (const auto& [name, offset] : meta.plainUniformOffsetsInUBO) {
+                    // The gl_NumSamples stand-in is routed by NAME and nothing else. It has no GL
+                    // location to look up - DoReflection hides it from the GL uniform index space
+                    // precisely so no application can address it - so the lookup below would find
+                    // nothing and log it as unbacked. Only the fragment stage declares it, and
+                    // every stage's copy sits at the same offset in the one shared global UBO.
+                    if (name == NUM_SAMPLES_UNIFORM_NAME) {
+                        artifacts.reservedNumSamplesOffset = offset;
+                        MGLOG_D("ProgramObject %u: BuildGlobalUboRouting - reserved gl_NumSamples stand-in '%s' "
+                                "backed at UBO offset %u",
+                                externalIndex, name.c_str(), offset);
+                        continue;
+                    }
                     // SPIRV-Reflect leaf names never carry a "[0]" suffix; frontend
                     // reflection keys arrays as "arr[0]" (GL naming), so retry with the
                     // suffix before declaring the uniform unbacked.

MobileGL/MG_State/GLState/ProgramState/ShaderCompileTask.cpp

@@ -140,17 +140,21 @@ namespace {
         return std::nullopt;
     }
 
-    // What glGetIntegerv(GL_MAX_SHADER_STORAGE_BUFFER_BINDINGS) answers, recomputed rather than
-    // queried: the compile runs on a worker with no context, and the pname is not a plain backend
-    // parameter - the getter caps the backend's count by the state layer's fixed binding-point
-    // array (GL_Getter's GetIndexedBufferQueryPointCount). A shader must be judged against the
-    // number the application was told, not against either half of it.
+    // What glGetIntegerv(GL_MAX_SHADER_STORAGE_BUFFER_BINDINGS) answers. Derived by the shared
+    // ResolveResourceBindingLimits so the compile-time scan below and the link-time general check
+    // (TMglGlslIoResolver::CheckDeclaredBindingRange) can never disagree about the number.
+    //
+    // Why BOTH still exist. GLSL makes an over-range binding a COMPILE-time error, and this scan
+    // is the only place MobileGL can raise one - glslang's own ceilings are switched off by the
+    // relaxed Vulkan parse and cannot be turned back on without changing the parse everything
+    // else depends on. The link-time check covers the four kinds a lexical scan of unexpanded
+    // source cannot see at all (samplers, images, uniform blocks, atomic counters, whose binding
+    // only survives inside a synthesized block NAME) and re-covers storage blocks as a backstop.
+    // The conformance predicate is compile AND link, so either site satisfies it; the split is
+    // about WHICH error GL reports, not about whether the shader is rejected.
     static MobileGL::Int MaxShaderStorageBufferBindings(
         const MobileGL::MG_Util::ShaderTranspiler::CompileEnv& env) {
-        const MobileGL::Int frontendPoints =
-            static_cast<MobileGL::Int>(MobileGL::MG_State::GLState::BufferBindingPointCount);
-        if (!env.HasBackend()) return frontendPoints;
-        return std::min<MobileGL::Int>(frontendPoints, std::max<MobileGL::Int>(env.params.MaxShaderStorageBufferBindings, 0));
+        return MobileGL::MG_State::GLState::ResolveResourceBindingLimits(env).MaxShaderStorageBufferBindings;
     }
 
     // The half of a compile that depends on nothing but the source text, the stage and the

MobileGL/MG_State/GLState/ProgramState/ShaderCompileTask.h

@@ -10,9 +10,54 @@
 #include <Includes.h>
 #include <MG_Util/Async/JobNode.h>
 #include <MG_Util/ShaderTranspiler/CompileEnv.h>
+#include <MG_Util/ShaderTranspiler/Types.h>
+#include <MG_State/GLState/BufferState/BufferState.h>
 #include <MG_State/GLState/ProgramState/ShaderPreprocessCache.h>
 
 namespace MobileGL::MG_State::GLState {
+    // THE one derivation of the binding ceilings a shader-declared layout(binding = N) is judged
+    // against. Two readers have to agree on them - the compile-time storage-block scan below and
+    // the link-time general check in TMglGlslIoResolver - and the numbers are recomputed here
+    // rather than queried because both readers run on a worker with no context.
+    //
+    // Each is exactly what glGetIntegerv answers for the matching pname, and none of them is a
+    // plain backend parameter: the buffer families are additionally capped by the state layer's
+    // indexed-binding array (GL_Getter's GetIndexedBufferQueryPointCount does the same), because
+    // a shader must be judged against the number the APPLICATION was told, not against either
+    // half of it. Lives in MG_State rather than in MG_Util/ShaderTranspiler/Types.h purely
+    // because BufferBindingPointCount is state-layer knowledge that the transpiler layer must
+    // not reach up for.
+    inline MG_Util::ShaderTranspiler::ResourceBindingLimits ResolveResourceBindingLimits(
+        const MG_Util::ShaderTranspiler::CompileEnv& env) {
+        namespace ST = MG_Util::ShaderTranspiler;
+        ST::ResourceBindingLimits limits;
+        const Int bindingPoints = static_cast<Int>(BufferBindingPointCount);
+        // The atomic-counter ceiling is a frontend constant, so it holds even with no backend -
+        // and it is the number BuildTBuiltInResource compiles a layout(binding = N) atomic_uint
+        // against, which is what makes it enforceable at all.
+        limits.MaxAtomicCounterBufferBindings = std::min<Int>(bindingPoints, ST::MAX_ATOMIC_COUNTER_BUFFER_BINDINGS);
+        // So is the uniform-buffer one: GL_MAX_UNIFORM_BUFFER_BINDINGS is clamped to the indexed
+        // binding array in the getter and its floor (the GL 4.5 core minimum of 84) is that same
+        // array's width, so the backend's own number never moves it.
+        limits.MaxUniformBufferBindings = bindingPoints;
+        // The storage-buffer ceiling has the same shape as GetIndexedBufferQueryPointCount's: the
+        // backend's count capped by the array, and the array alone when there is no backend. That
+        // "no backend" arm is not a detail - it is what the GPU-free test binary runs under, and
+        // it has to keep matching what glGetIntegerv answers there.
+        limits.MaxShaderStorageBufferBindings =
+            env.HasBackend()
+                ? std::min<Int>(bindingPoints, std::max<Int>(env.params.MaxShaderStorageBufferBindings, 0))
+                : bindingPoints;
+        if (!env.HasBackend()) {
+            // The two genuinely per-DEVICE ceilings have nothing to be measured against here, and
+            // zero means "do not enforce this kind" rather than "reject everything".
+            return limits;
+        }
+        limits.MaxSamplerBindings = std::max<Int>(env.params.MaxCombinedTextureImageUnits, 0);
+        limits.MaxImageBindings = std::max<Int>(env.params.MaxImageUnits, 0);
+        return limits;
+    }
+
     // glslang has no "detach this thread" API in the vendored revision, but TShader::parse
     // leaves the calling thread's TLS pool allocator pointing at the shader's own pool and
     // never restores it. Left there, the next allocation this thread makes - in an unrelated

MobileGL/MG_State/GLState/ProgramState/ShaderObject.cpp

@@ -15,7 +15,77 @@
 #include <MG_Util/ShaderTranspiler/Types.h>
 
 namespace MobileGL::MG_State::GLState {
+    void ShaderObject::SetSpirvBinary(Vector<Uint32>&& binary) {
+        // A module replaces whatever this object stood for, so the compiled state of the old
+        // source goes with it - including a compile still in flight.
+        ReleaseCompileNode();
+        m_spirvBinary = Move(binary);
+        m_hasSpirvBinary = true;
+        m_specialized = false;
+        m_specializationFailed = false;
+        m_specializationInfoLog.clear();
+        m_spirvXfbVaryings.clear();
+        m_spirvXfbBufferMode = GL_INTERLEAVED_ATTRIBS;
+        m_source = MakeShared<const String>(String{});
+        InvalidateCompiledState();
+    }
+
+    const String& ShaderObject::GetApplicationShaderSource() const {
+        static const String kNoSource;
+        // Both the unspecialized and the specialized windows answer empty: in the first m_source
+        // already is empty, in the second it holds generated GLSL that the application never wrote.
+        return m_hasSpirvBinary ? kNoSource : *m_source;
+    }
+
+    void ShaderObject::SpecializeFromSpirv(String&& glsl, Vector<String>&& xfbVaryings, GLenum xfbBufferMode) {
+        ReleaseCompileNode();
+        // The latch goes up HERE and nowhere else - this is the one path that actually specialized
+        // the shader.
+        m_specialized = true;
+        m_specializationFailed = false;
+        m_specializationInfoLog.clear();
+        m_spirvXfbVaryings = Move(xfbVaryings);
+        m_spirvXfbBufferMode = xfbBufferMode;
+        // The GLSL the module specializes to enters the ORDINARY pipeline from here: preprocess,
+        // glslang parse, reflection, transpile, both backends. Nothing downstream needs to know
+        // the source was not written by the application - which is the whole reason this hop
+        // exists, and the reason a SPIR-V program's GL-visible surface (uniform locations, block
+        // indices, transform-feedback layout) is populated at all.
+        m_source = MakeShared<const String>(Move(glsl));
+        InvalidateCompiledState();
+        Compile();
+    }
+
+    void ShaderObject::RecordSpecializationFailure(String&& infoLog) {
+        ReleaseCompileNode();
+        m_source = MakeShared<const String>(String{});
+        InvalidateCompiledState();
+        m_specializationFailed = true;
+        m_specializationInfoLog = Move(infoLog);
+    }
+
     void ShaderObject::SetShaderSource(const String& source) {
+        // glShaderSource on a SPIR-V shader takes the object back to being a GLSL one, and
+        // GL_SPIR_V_BINARY must then read FALSE (ARB_gl_spirv; gl4cGlSpirvTests'
+        // spirv_modules_state_queries_test checks exactly this transition). The stored module goes
+        // with the flag - re-specializing it would be re-specializing a shader the application has
+        // already replaced. The memo below is skipped on purpose: the source may well be
+        // byte-identical to the empty string this object has been holding, and keeping the
+        // "compiled state" of that would keep the module's verdict too.
+        if (m_hasSpirvBinary || m_specializationFailed) {
+            m_hasSpirvBinary = false;
+            m_spirvBinary.clear();
+            m_spirvBinary.shrink_to_fit();
+            m_specialized = false;
+            m_specializationFailed = false;
+            m_specializationInfoLog.clear();
+            m_spirvXfbVaryings.clear();
+            m_spirvXfbBufferMode = GL_INTERLEAVED_ATTRIBS;
+            ReleaseCompileNode();
+            m_source = MakeShared<const String>(source);
+            InvalidateCompiledState();
+            return;
+        }
         // P0b layer 1. glShaderSource always REPLACES the source, but replacing it with a
         // byte-identical one cannot change what a compile would produce: the whole
         // pipeline (preprocess -> lexical checks -> glslang parse) is a pure function of
@@ -36,6 +106,10 @@ namespace MobileGL::MG_State::GLState {
     }
 
     void ShaderObject::SetShaderSource(String&& source) {
+        if (m_hasSpirvBinary || m_specializationFailed) {
+            SetShaderSource(static_cast<const String&>(source));
+            return;
+        }
         if (SourceMatchesCompiledState(source)) return;
         ReleaseCompileNode();
         m_source = MakeShared<const String>(Move(source));

MobileGL/MG_State/GLState/ProgramState/ShaderObject.h

@@ -65,6 +65,51 @@ namespace MobileGL {
             void SetShaderSource(const String& source);
             void SetShaderSource(String&& source);
             void Compile();
+
+            // ---- GL_ARB_gl_spirv ----
+            // glShaderBinary(GL_SHADER_BINARY_FORMAT_SPIR_V): the object stops standing for a
+            // GLSL source and starts standing for an application-supplied SPIR-V module. The
+            // module is held verbatim until glSpecializeShader names an entry point for it -
+            // ARB_gl_spirv makes the pair a two-step operation, and glCompileShader in between is
+            // INVALID_OPERATION rather than a compile of anything.
+            //
+            // Both directions clear the other: glShaderSource on a SPIR-V shader takes it back to
+            // being a GLSL shader with GL_SPIR_V_BINARY reading FALSE, which the conformance suite
+            // checks explicitly.
+            void SetSpirvBinary(Vector<Uint32>&& binary);
+            Bool HasSpirvBinary() const { return m_hasSpirvBinary; }
+            // ARB_gl_spirv: "Once specialized, a shader may not be re-specialized without first
+            // re-associating the original SPIR-V module with it, through ShaderBinary." A second
+            // glSpecializeShader is GL_INVALID_OPERATION, and this latch is what answers that.
+            //
+            // Set ONLY on the success path. A specialization that FAILED did not specialize the
+            // shader, and the conformance suite relies on that distinction: it deliberately fails
+            // specialization (a bad entry point, then an unknown constant id) on one shader object
+            // and then requires the next, well-formed call on that same object to be accepted.
+            Bool HasBeenSpecialized() const { return m_specialized; }
+            const Vector<Uint32>& GetSpirvBinary() const { return m_spirvBinary; }
+            // glSpecializeShader's half: hand the object the GLSL its module specializes to and
+            // let the ordinary pipeline compile it.
+            void SpecializeFromSpirv(String&& glsl, Vector<String>&& xfbVaryings, GLenum xfbBufferMode);
+            // The capture the object's SPIR-V module DECLARED, as the equivalent
+            // glTransformFeedbackVaryings request. Empty for a GLSL shader and for a SPIR-V module
+            // that declares no transform feedback. ProgramObject::Link picks this up from the
+            // program's last vertex-processing stage, because ARB_gl_spirv makes decorations the
+            // only declaration form for a SPIR-V program and glTransformFeedbackVaryings has no
+            // effect on one.
+            const Vector<String>& GetSpirvXfbVaryings() const { return m_spirvXfbVaryings; }
+            GLenum GetSpirvXfbBufferMode() const { return m_spirvXfbBufferMode; }
+            // What glGetShaderSource / GL_SHADER_SOURCE_LENGTH must answer. A shader created from
+            // glShaderBinary never had glShaderSource called on it, so GL 4.6 core 7.1 makes its
+            // source the empty string - even after glSpecializeShader, when m_source holds the
+            // SPIRV-Cross GLSL the module was translated into. That text is MobileGL's, not the
+            // application's, and handing it back invites an application to cache and re-submit it.
+            const String& GetApplicationShaderSource() const;
+            // The other half: specialization itself failed (a bad entry point, a constant id the
+            // module does not declare, a module spirv-val rejects). There is nothing to compile,
+            // so the verdict is recorded directly - COMPILE_STATUS false with this log - and both
+            // queries answer from it without touching the compile pipeline.
+            void RecordSpecializationFailure(String&& infoLog);
             // Gives up this object's claim on its compile node, cancelling the node only if
             // this object was its LAST claimant. Called at the points where the object's
             // compiled state stops being observable through THIS name: a real source change,
@@ -99,14 +144,16 @@ namespace MobileGL {
             const SharedPtr<const String>& GetShaderSourcePtr() const { return m_source; }
 
             const SharedPtr<glslang::TShader>& GetCompiledShader() const { return Compiled().shader; }
-            const String& GetInfoLog() const { return Compiled().infoLog; }
+            const String& GetInfoLog() const {
+                return m_specializationFailed ? m_specializationInfoLog : Compiled().infoLog;
+            }
             // Explicit layout(location = N) qualifiers on this shader's default-block
             // uniforms, as glslang recorded them at the point its Vulkan-relaxed remap
             // discarded them (see CollectExplicitUniformLocations).
             const UnorderedMap<String, Int>& GetExplicitUniformLocations() const {
                 return Compiled().explicitUniformLocations;
             }
-            Bool GetCompileStatus() const { return Compiled().compileStatus; }
+            Bool GetCompileStatus() const { return m_specializationFailed ? false : Compiled().compileStatus; }
             Bool GetDeleteStatus() const { return m_deleteStatus; }
 
             // Blocks until a pending compile has published its artifacts. Public for the
@@ -248,6 +295,25 @@ namespace MobileGL {
             // query optimistically for the current node. Cleared wherever the node
             // changes hands (AdoptCompileNode) or goes away (DropCompileNode).
             mutable Bool m_optimisticAnswerLatched = false;
+            // The application-supplied SPIR-V module and the flag GL_SPIR_V_BINARY reports. The
+            // module is kept after specialization too: glSpecializeShader may legally run again on
+            // the same object with different constants, and the second call has to re-specialize
+            // the ORIGINAL words rather than the ones the first call folded.
+            Vector<Uint32> m_spirvBinary;
+            Bool m_hasSpirvBinary = false;
+            // "This shader has been specialized"; see HasBeenSpecialized. Cleared by anything that
+            // re-associates a module (SetSpirvBinary) or turns the object back into a GLSL shader
+            // (either SetShaderSource overload) - which is exactly the re-association ARB_gl_spirv
+            // names as the way to make a second specialization legal again.
+            Bool m_specialized = false;
+            Vector<String> m_spirvXfbVaryings;
+            GLenum m_spirvXfbBufferMode = GL_INTERLEAVED_ATTRIBS;
+            // A specialization that failed before any compile could start. Kept beside the
+            // compile artifacts rather than inside them because there is no compile job to hang
+            // it on - see RecordSpecializationFailure. Cleared by anything that gives the object
+            // a new meaning (a new source, a new module, a fresh specialization).
+            Bool m_specializationFailed = false;
+            String m_specializationInfoLog;
         };
     } // namespace MG_State::GLState
 } // namespace MobileGL

MobileGL/MG_State/GLState/RenderState/RenderState.cpp

@@ -217,12 +217,43 @@ namespace MobileGL {
                 return m_parameters.PatchVertices;
             }
 
-            void RenderState::SetPolygonOffset(Float factor, Float units) {
-                if (m_parameters.PolygonOffsetFactor == factor && m_parameters.PolygonOffsetUnits == units) return;
+            // BumpVersions(), not just ++m_version, for the same reason SetPatchVertices does it:
+            // these levels are compiled INTO the synthesized pass-through tessellation control
+            // stage on both backends, so changing one makes an already-built program stale.
+            //
+            // The redundant-write guard compares BIT PATTERNS, not floats: glPatchParameterfv
+            // accepts NaN, and a float compare would let a re-set of the identical NaN tuple fall
+            // through and bump the pipeline-state version - invalidating DirectVulkan's pipeline
+            // memo and DirectGLES's render-state span - on every single call.
+            void RenderState::SetPatchDefaultOuterLevel(const FloatVec4& levels) {
+                if (BitwiseEqual(m_parameters.PatchDefaultOuterLevel, levels)) return;
 
-                m_parameters.PolygonOffsetFactor = factor;
-                m_parameters.PolygonOffsetUnits = units;
-                ++m_version;
+                m_parameters.PatchDefaultOuterLevel = levels;
+                BumpVersions();
+            }
+
+            const FloatVec4& RenderState::GetPatchDefaultOuterLevel() const {
+                return m_parameters.PatchDefaultOuterLevel;
+            }
+
+            void RenderState::SetPatchDefaultInnerLevel(const FloatVec2& levels) {
+                if (BitwiseEqual(m_parameters.PatchDefaultInnerLevel, levels)) return;
+
+                m_parameters.PatchDefaultInnerLevel = levels;
+                BumpVersions();
+            }
+
+            const FloatVec2& RenderState::GetPatchDefaultInnerLevel() const {
+                return m_parameters.PatchDefaultInnerLevel;
+            }
+
+            void RenderState::SetPolygonOffset(Float factor, Float units) {
+                // GL 4.6 core 14.6.5 defines PolygonOffset(factor, units) as EQUIVALENT to
+                // PolygonOffsetClamp(factor, units, 0) - the equivalence is total, so the clamp is
+                // written too, not merely left alone. Leaving it meant a glPolygonOffsetClamp(1, 1,
+                // 0.5) followed by a plain glPolygonOffset(3, 4) still reported a clamp of 0.5, and
+                // the early-out below could even skip the version bump while doing it.
+                SetPolygonOffsetClamped(factor, units, 0.0f);
             }
 
             Float RenderState::GetPolygonOffsetFactor() const {
@@ -233,6 +264,37 @@ namespace MobileGL {
                 return m_parameters.PolygonOffsetUnits;
             }
 
+            void RenderState::SetPolygonOffsetClamped(Float factor, Float units, Float clamp) {
+                if (m_parameters.PolygonOffsetFactor == factor && m_parameters.PolygonOffsetUnits == units &&
+                    m_parameters.PolygonOffsetClamp == clamp)
+                    return;
+
+                m_parameters.PolygonOffsetFactor = factor;
+                m_parameters.PolygonOffsetUnits = units;
+                m_parameters.PolygonOffsetClamp = clamp;
+                ++m_version;
+            }
+
+            Float RenderState::GetPolygonOffsetClamp() const {
+                return m_parameters.PolygonOffsetClamp;
+            }
+
+            void RenderState::SetClipControl(GLenum origin, GLenum depth) {
+                if (m_parameters.ClipOrigin == origin && m_parameters.ClipDepthMode == depth) return;
+
+                m_parameters.ClipOrigin = origin;
+                m_parameters.ClipDepthMode = depth;
+                ++m_version;
+            }
+
+            GLenum RenderState::GetClipOrigin() const {
+                return m_parameters.ClipOrigin;
+            }
+
+            GLenum RenderState::GetClipDepthMode() const {
+                return m_parameters.ClipDepthMode;
+            }
+
             // -------------------- Capabilities --------------------
             namespace {
                 // CapabilityInput lists ClipDistance0..7 contiguously (RenderState.h); the caller
@@ -270,6 +332,7 @@ namespace MobileGL {
                     SET_CAPABILITY(SampleAlphaToOne, enabled);
                     SET_CAPABILITY(SampleCoverage, enabled);
                     SET_CAPABILITY(SampleMask, enabled);
+                    SET_CAPABILITY(SampleShading, enabled);
                     SET_CAPABILITY(StencilTest, enabled);
                     SET_CAPABILITY(ProgramPointSize, enabled);
                 case CapabilityInput::Blend: {
@@ -344,6 +407,7 @@ namespace MobileGL {
                     RETURN_CAPABILITY(SampleAlphaToOne);
                     RETURN_CAPABILITY(SampleCoverage);
                     RETURN_CAPABILITY(SampleMask);
+                    RETURN_CAPABILITY(SampleShading);
                     RETURN_CAPABILITY(StencilTest);
                     RETURN_CAPABILITY(ProgramPointSize);
                 case CapabilityInput::Blend:
@@ -737,6 +801,20 @@ namespace MobileGL {
                 return m_parameters.SampleMaskValue;
             }
 
+            void RenderState::SetMinSampleShadingValue(Float value) {
+                if (m_parameters.MinSampleShadingValue == value) return;
+
+                m_parameters.MinSampleShadingValue = value;
+                // BumpVersions, not just ++m_version: DirectVulkan bakes the fraction into
+                // VkPipelineMultisampleStateCreateInfo::minSampleShading, so a cached pipeline
+                // built with the old value must not be reused.
+                BumpVersions();
+            }
+
+            Float RenderState::GetMinSampleShadingValue() const {
+                return m_parameters.MinSampleShadingValue;
+            }
+
             // -------------------- Pixel Store --------------------
             void RenderState::SetPixelStoreParam(PixelStoreParam param, Int value) {
 #define SET_PIXEL_STORE_PARAM(paramNameHead, paramNameTail, val)                                                       \

MobileGL/MG_State/GLState/RenderState/RenderState.h

@@ -240,8 +240,24 @@ namespace MobileGL {
         Float PointSize = 1.0f;
         // GL_PATCH_VERTICES: how many vertices one tessellation patch consumes.
         Uint PatchVertices = 3;
+        // GL_PATCH_DEFAULT_OUTER_LEVEL / GL_PATCH_DEFAULT_INNER_LEVEL (glPatchParameterfv). The
+        // tessellation levels used when a program has an evaluation stage and NO control stage -
+        // GL's fixed-function pass-through (4.6 core 11.2.2). Both backends have to synthesize
+        // that stage, and they bake these numbers into it, so a change here makes an already-built
+        // one stale exactly as PATCH_VERTICES does. Default 1.0, per table 23.44.
+        FloatVec4 PatchDefaultOuterLevel = FloatVec4(1.0f, 1.0f, 1.0f, 1.0f);
+        FloatVec2 PatchDefaultInnerLevel = FloatVec2(1.0f, 1.0f);
         Float PolygonOffsetFactor = 0.0f;
         Float PolygonOffsetUnits = 0.0f;
+        // GL_POLYGON_OFFSET_CLAMP (GL 4.6 core 14.6.5 / GL_EXT_polygon_offset_clamp): the maximum
+        // magnitude of the offset glPolygonOffsetClamp's third argument allows. Zero - the default
+        // - means "no clamp", which is exactly the behaviour glPolygonOffset leaves behind.
+        Float PolygonOffsetClamp = 0.0f;
+
+        // glClipControl (GL 4.5 core 13.5). Defaults per table 23.7 are the pre-4.5 fixed
+        // behaviour: origin at the lower left, depth mapped from -1..1.
+        GLenum ClipOrigin = GL_LOWER_LEFT;
+        GLenum ClipDepthMode = GL_NEGATIVE_ONE_TO_ONE;
 
         // Blending
         Array<PerBufferBlendState, MG_State::GLState::FramebufferObject::MAX_DRAW_BUFFERS> BlendStates;
@@ -271,6 +287,10 @@ namespace MobileGL {
         Float SampleCoverageValue = 1.0f;
         Bool SampleCoverageInvert = false;
         Uint32 SampleMaskValue = 0xffffffffu;
+        // glMinSampleShading (ARB_sample_shading / GL 4.0 core 14.3.1). The fraction of samples
+        // that get their own independent shading when GL_SAMPLE_SHADING is enabled; the initial
+        // value is 0, and the value is clamped to [0, 1] on the way in.
+        Float MinSampleShadingValue = 0.0f;
         Array<StencilFaceState, 2> StencilStates{};
 
         // Cull Face
@@ -319,6 +339,7 @@ namespace MobileGL {
         Bool SampleAlphaToOneEnabled = false;
         Bool SampleCoverageEnabled = false;
         Bool SampleMaskEnabled = false;
+        Bool SampleShadingEnabled = false;
         Bool StencilTestEnabled = false;
         Bool ProgramPointSizeEnabled = false;
         // glEnable(GL_SCISSOR_TEST) enables the test for EVERY viewport, glEnablei for one
@@ -374,9 +395,21 @@ namespace MobileGL {
                 Float GetPointSize() const;
                 void SetPatchVertices(Uint vertices);
                 Uint GetPatchVertices() const;
+                void SetPatchDefaultOuterLevel(const FloatVec4& levels);
+                const FloatVec4& GetPatchDefaultOuterLevel() const;
+                void SetPatchDefaultInnerLevel(const FloatVec2& levels);
+                const FloatVec2& GetPatchDefaultInnerLevel() const;
                 void SetPolygonOffset(Float factor, Float units);
+                // glPolygonOffsetClamp. Writes the same factor/units as glPolygonOffset plus the
+                // clamp, because that is what the entry point does - glPolygonOffset is the
+                // clamp = 0 case of it (GL 4.6 core 14.6.5).
+                void SetPolygonOffsetClamped(Float factor, Float units, Float clamp);
                 Float GetPolygonOffsetFactor() const;
                 Float GetPolygonOffsetUnits() const;
+                Float GetPolygonOffsetClamp() const;
+                void SetClipControl(GLenum origin, GLenum depth);
+                GLenum GetClipOrigin() const;
+                GLenum GetClipDepthMode() const;
                 // Hints. target must be one of the 4 GL 3.3 core hint targets (validated by the caller).
                 void SetHint(GLenum target, GLenum mode);
                 GLenum GetHint(GLenum target) const;
@@ -454,6 +487,9 @@ namespace MobileGL {
                 Bool GetSampleCoverageInvert() const;
                 void SetSampleMaskValue(Uint32 mask);
                 Uint32 GetSampleMaskValue() const;
+                // glMinSampleShading. `value` is stored as given; the entry point clamps.
+                void SetMinSampleShadingValue(Float value);
+                Float GetMinSampleShadingValue() const;
 
                 // Pixel Store
                 void SetPixelStoreParam(PixelStoreParam param, Int value);

MobileGL/MG_State/GLState/SamplerState/SamplerObject.cpp

@@ -155,9 +155,21 @@ namespace MobileGL {
             // an answer whichever form was written. Integer <-> float uses the plain value, matching
             // what glTexParameterIiv/Iuiv mean: those forms are for integer texture formats, whose
             // border components are the raw integers rather than a normalized fraction.
+            //
+            // Which of the three the application actually WROTE is recorded separately in
+            // borderColorForm, because the derived values erase it: a backend has to know whether to
+            // forward the colour through glSamplerParameterfv or glSamplerParameterIiv (and which
+            // VkBorderColor family to ask Vulkan for), and the numbers alone cannot say. That is also
+            // why every setter's early-out tests the form as well as the value - a float (0,0,0,1)
+            // followed by an integer (0,0,0,1) is a real state change even though nothing numeric
+            // moved, and swallowing it would leave the backend syncing the wrong entry point forever.
             void SamplerObject::SetBorderColor(const FloatVec4& color) {
-                if (color == m_samplerParameters.borderColor) return;
+                if (color == m_samplerParameters.borderColor &&
+                    m_samplerParameters.borderColorForm == BorderColorForm::Float) {
+                    return;
+                }
 
+                m_samplerParameters.borderColorForm = BorderColorForm::Float;
                 m_samplerParameters.borderColor = color;
                 m_samplerParameters.borderColorI =
                     IntVec4(static_cast<Int32>(color.x()), static_cast<Int32>(color.y()),
@@ -169,8 +181,12 @@ namespace MobileGL {
             }
 
             void SamplerObject::SetBorderColorI(const IntVec4& color) {
-                if (color == m_samplerParameters.borderColorI) return;
+                if (color == m_samplerParameters.borderColorI &&
+                    m_samplerParameters.borderColorForm == BorderColorForm::Int) {
+                    return;
+                }
 
+                m_samplerParameters.borderColorForm = BorderColorForm::Int;
                 m_samplerParameters.borderColorI = color;
                 m_samplerParameters.borderColorUI =
                     UintVec4(static_cast<Uint32>(color.x()), static_cast<Uint32>(color.y()),
@@ -182,8 +198,12 @@ namespace MobileGL {
             }
 
             void SamplerObject::SetBorderColorUI(const UintVec4& color) {
-                if (color == m_samplerParameters.borderColorUI) return;
+                if (color == m_samplerParameters.borderColorUI &&
+                    m_samplerParameters.borderColorForm == BorderColorForm::Uint) {
+                    return;
+                }
 
+                m_samplerParameters.borderColorForm = BorderColorForm::Uint;
                 m_samplerParameters.borderColorUI = color;
                 m_samplerParameters.borderColorI =
                     IntVec4(static_cast<Int32>(color.x()), static_cast<Int32>(color.y()),
@@ -206,6 +226,10 @@ namespace MobileGL {
                 return m_samplerParameters.borderColorUI;
             }
 
+            BorderColorForm SamplerObject::GetBorderColorForm() const {
+                return m_samplerParameters.borderColorForm;
+            }
+
             SamplerCompareMode SamplerObject::GetCompareMode() const {
                 return m_samplerParameters.compareMode;
             }

MobileGL/MG_State/GLState/SamplerState/SamplerObject.h

@@ -56,6 +56,19 @@ namespace MobileGL {
         Unknown = -1
     };
 
+    // Which of the three GL_TEXTURE_BORDER_COLOR entry-point families last wrote the border colour,
+    // and therefore which of the three stored representations is AUTHORITATIVE. GL 4.6 core 8.10:
+    // TexParameterIiv/Iuiv store an integer border colour "unmodified, with an internal data type of
+    // integer", TexParameterfv stores a floating-point one, and the derived forms are only a
+    // convenience for a getter of the other spelling. A backend cannot pick the right driver entry
+    // point (glSamplerParameterIiv vs fv) or the right VkBorderColor family without this: numerically
+    // the three representations are always populated, so the value alone says nothing about the form.
+    enum class BorderColorForm : Uint8 {
+        Float,
+        Int,
+        Uint
+    };
+
     struct SamplerParameters {
         SamplerWrapMode wrapS = SamplerWrapMode::Repeat;
         SamplerWrapMode wrapT = SamplerWrapMode::Repeat;
@@ -79,6 +92,7 @@ namespace MobileGL {
         FloatVec4 borderColor = {0.0f, 0.0f, 0.0f, 0.0f};
         IntVec4 borderColorI = {0, 0, 0, 0};
         UintVec4 borderColorUI = {0, 0, 0, 0};
+        BorderColorForm borderColorForm = BorderColorForm::Float;
     };
 
     namespace MG_State {
@@ -117,6 +131,7 @@ namespace MobileGL {
                 const FloatVec4& GetBorderColor() const;
                 const IntVec4& GetBorderColorI() const;
                 const UintVec4& GetBorderColorUI() const;
+                BorderColorForm GetBorderColorForm() const;
                 Uint GetExternalIndex() const;
                 Uint16 GetVersion() const;
                 // Globally-unique, never-reused id for this sampler object's lifetime. Lets a

MobileGL/MG_State/GLState/TextureState/TextureObject.cpp

@@ -113,8 +113,14 @@ namespace MobileGL {
                 return m_sampler->GetBorderColor();
             }
 
+            // The redundancy filters test the FORM as well as the value: the derived representations
+            // make a float (0,0,0,1) and an integer (0,0,0,1) numerically identical, but they are
+            // different GL state and the DirectGLES sync memoises on m_textureParamsVersion.
             void TextureObjectBase::SetBorderColor(const FloatVec4& color) {
-                if (color == m_sampler->GetBorderColor()) return;
+                if (color == m_sampler->GetBorderColor() &&
+                    m_sampler->GetBorderColorForm() == BorderColorForm::Float) {
+                    return;
+                }
 
                 m_sampler->SetBorderColor(color);
                 ++m_textureParamsVersion;
@@ -125,7 +131,10 @@ namespace MobileGL {
             }
 
             void TextureObjectBase::SetBorderColorI(const IntVec4& color) {
-                if (color == m_sampler->GetBorderColorI()) return;
+                if (color == m_sampler->GetBorderColorI() &&
+                    m_sampler->GetBorderColorForm() == BorderColorForm::Int) {
+                    return;
+                }
 
                 m_sampler->SetBorderColorI(color);
                 ++m_textureParamsVersion;
@@ -136,12 +145,19 @@ namespace MobileGL {
             }
 
             void TextureObjectBase::SetBorderColorUI(const UintVec4& color) {
-                if (color == m_sampler->GetBorderColorUI()) return;
+                if (color == m_sampler->GetBorderColorUI() &&
+                    m_sampler->GetBorderColorForm() == BorderColorForm::Uint) {
+                    return;
+                }
 
                 m_sampler->SetBorderColorUI(color);
                 ++m_textureParamsVersion;
             }
 
+            BorderColorForm TextureObjectBase::GetBorderColorForm() const {
+                return m_sampler->GetBorderColorForm();
+            }
+
             TextureSwizzleParam TextureObjectBase::GetSwizzleParam(TextureSwizzleParam param) const {
                 switch (param) {
                 case TextureSwizzleParam::Red:
@@ -291,6 +307,13 @@ namespace MobileGL {
                 return m_lifetimeId;
             }
 
+            const SharedPtr<ITextureObject>& TextureObjectBase::GetViewStorageOwner() const {
+                // A plain texture owns its own storage. Only TextureObjectView overrides this,
+                // which is what IsTextureView() keys on everywhere else.
+                static const SharedPtr<ITextureObject> noStorageOwner = nullptr;
+                return noStorageOwner;
+            }
+
             Uint TextureObjectWithOneMipmap::GetMipmapLevelCount() const {
                 return m_textureStorage.GetLevelCount();
             }

MobileGL/MG_State/GLState/TextureState/TextureObject.h

@@ -40,6 +40,8 @@ namespace MobileGL::MG_State::GLState {
         virtual void SetBorderColorI(const IntVec4& color) = 0;
         virtual const UintVec4& GetBorderColorUI() const = 0;
         virtual void SetBorderColorUI(const UintVec4& color) = 0;
+        // Which of the three setters above last ran; see SamplerParameters::borderColorForm.
+        virtual BorderColorForm GetBorderColorForm() const = 0;
         virtual TextureSwizzleParam GetSwizzleParam(TextureSwizzleParam param) const = 0;
         virtual void SetSwizzleParam(TextureSwizzleParam param, TextureSwizzleParam value) = 0;
         virtual void SetSwizzleParamRGBA(const Vec4<TextureSwizzleParam>& values) = 0;
@@ -49,6 +51,12 @@ namespace MobileGL::MG_State::GLState {
         virtual void SetMaxLevel(Uint maxLevel) = 0;
         virtual Bool IsImmutable() const = 0;
         virtual Uint GetImmutableLevels() const = 0;
+        // How many levels THIS object can address, i.e. the bound a level argument has to
+        // stay under. The same number as GetImmutableLevels() for an ordinary immutable
+        // texture, but NOT for a view: GL 4.6 core 8.18 defines TEXTURE_IMMUTABLE_LEVELS on a
+        // view as the ORIGINAL texture's value, which says nothing about what the view itself
+        // can reach, and bounding by it lets a level the view does not have through.
+        virtual Uint GetAddressableLevelCount() const = 0;
         virtual void SetImmutableLevels(Uint levels) = 0;
         virtual Uint16 GetTextureParamsVersion() const = 0;
         // Monotonic counter bumped on every CPU-side pixel mutation (see MarkStorageDirty).
@@ -78,6 +86,29 @@ namespace MobileGL::MG_State::GLState {
         virtual GLenum GetDepthStencilTextureMode() const = 0;
         virtual void SetDepthStencilTextureMode(GLenum mode) = 0;
 
+        // ---- Texture views (ARB_texture_view / GL 4.6 core 8.18) ----
+        // The texture object whose immutable storage this one's texels actually live in, or
+        // nullptr when this texture owns its storage. It is itself NEVER a view: glTextureView
+        // composes a view-of-a-view onto the ROOT at creation, which is exactly what the spec's
+        // additive "<minlevel> plus the value of TEXTURE_VIEW_MIN_LEVEL from the original
+        // texture" rule describes, so one hop always reaches the storage.
+        //
+        // Holding it as a SharedPtr is what gives GL's name-deletion semantics for free: after
+        // glDeleteTextures(origtexture) the name is gone and TextureState has dropped its entry,
+        // but the object - and therefore the storage and every backend resource keyed on it -
+        // stays alive as long as some view still references it (GL 4.6 core 5.1.2).
+        virtual const SharedPtr<ITextureObject>& GetViewStorageOwner() const = 0;
+        Bool IsTextureView() const { return GetViewStorageOwner() != nullptr; }
+        // GL 4.6 core table 23.17, expressed in the storage owner's level/layer coordinates
+        // (see above - composition makes the two the same number). All four are 0 on a mutable
+        // texture; TexStorage* seeds them with (0, levels, 0, layers) because the spec makes an
+        // immutable texture a full-extent view of itself, and glTextureView composes onto those.
+        virtual Uint GetViewMinLevel() const = 0;
+        virtual Uint GetViewNumLevels() const = 0;
+        virtual Uint GetViewMinLayer() const = 0;
+        virtual Uint GetViewNumLayers() const = 0;
+        virtual void SetViewLevelLayerRange(Uint minLevel, Uint numLevels, Uint minLayer, Uint numLayers) = 0;
+
     protected:
         virtual Uint GetIndexOfTextureUploadTarget(TextureUploadTarget target) const = 0;
     };
@@ -100,6 +131,7 @@ namespace MobileGL::MG_State::GLState {
         void SetBorderColorI(const IntVec4& color) override;
         const UintVec4& GetBorderColorUI() const override;
         void SetBorderColorUI(const UintVec4& color) override;
+        BorderColorForm GetBorderColorForm() const override;
         TextureSwizzleParam GetSwizzleParam(TextureSwizzleParam param) const override;
         const Vec4<TextureSwizzleParam>& GetAllSwizzleParams() const override;
         void SetSwizzleParam(TextureSwizzleParam param, TextureSwizzleParam value) override;
@@ -109,6 +141,10 @@ namespace MobileGL::MG_State::GLState {
         void SetMaxLevel(Uint maxLevel) override;
         Bool IsImmutable() const override;
         Uint GetImmutableLevels() const override;
+        // m_immutableLevels is already the VIEW-relative count for a view (its constructor
+        // stores <numlevels> there so the level-range clamp works in view coordinates), so
+        // this one accessor is correct for both and needs no override.
+        Uint GetAddressableLevelCount() const override { return m_immutableLevels; }
         void SetImmutableLevels(Uint levels) override;
         Uint16 GetTextureParamsVersion() const override;
         Uint64 GetContentVersion() const override;
@@ -123,6 +159,18 @@ namespace MobileGL::MG_State::GLState {
         Bool HasFixedSampleLocations() const override;
         void SetFixedSampleLocations(Bool fixedSampleLocations) override;
         Uint64 GetLifetimeId() const override;
+        // A plain texture owns its storage; TextureObjectView overrides this.
+        const SharedPtr<ITextureObject>& GetViewStorageOwner() const override;
+        Uint GetViewMinLevel() const override { return m_viewMinLevel; }
+        Uint GetViewNumLevels() const override { return m_viewNumLevels; }
+        Uint GetViewMinLayer() const override { return m_viewMinLayer; }
+        Uint GetViewNumLayers() const override { return m_viewNumLayers; }
+        void SetViewLevelLayerRange(Uint minLevel, Uint numLevels, Uint minLayer, Uint numLayers) override {
+            m_viewMinLevel = minLevel;
+            m_viewNumLevels = numLevels;
+            m_viewMinLayer = minLayer;
+            m_viewNumLayers = numLayers;
+        }
         GLenum GetDepthStencilTextureMode() const override { return m_depthStencilTextureMode; }
         // Bumps the params version like every other backend-visible texture parameter: the mode
         // decides which ASPECT of a packed depth/stencil image a sampler reads, which DirectGLES
@@ -165,6 +213,12 @@ namespace MobileGL::MG_State::GLState {
         // matches before its first sync. Bumped only on dirty=true in MarkStorageDirty.
         Uint64 m_contentVersion = 1;
         GLenum m_depthStencilTextureMode = GL_DEPTH_COMPONENT;
+        // GL 4.6 core table 23.17: all four are 0 until immutable storage exists, which is what
+        // makes glGetTexParameteriv(GL_TEXTURE_VIEW_NUM_LEVELS) answer 0 on a mutable texture.
+        Uint m_viewMinLevel = 0;
+        Uint m_viewNumLevels = 0;
+        Uint m_viewMinLayer = 0;
+        Uint m_viewNumLayers = 0;
         Int m_samples = 0;
         Bool m_fixedSampleLocations = true;
     };

MobileGL/MG_State/GLState/TextureState/TextureObjectView.cpp

@@ -0,0 +1,368 @@
+// MobileGL - MobileGL/MG_State/GLState/TextureState/TextureObjectView.cpp
+// Copyright (c) 2025-2026 MobileGL-Dev
+// Licensed under the GNU Lesser General Public License v3.0:
+//   https://www.gnu.org/licenses/gpl-3.0.txt
+//   https://www.gnu.org/licenses/lgpl-3.0.txt
+// SPDX-License-Identifier: LGPL-3.0-only
+// End of Source File Header
+
+#include "TextureObjectView.h"
+
+#include <algorithm>
+#include <cstring>
+
+namespace MobileGL::MG_State::GLState {
+    namespace {
+        // Where a target keeps its LAYER count. GL puts a 1D array's layers in the state-side
+        // height (that is what glTexImage2D(GL_TEXTURE_1D_ARRAY, width, layers) means, and what
+        // TextureObject.cpp's completeness walk assumes); every other layered target keeps them
+        // in z. GL_TEXTURE_3D is deliberately None: its depth is a spatial axis, not layers, and
+        // ARB_texture_view forbids anything but a full-depth 3D->3D view of it.
+        enum class LayerAxis { None, Y, Z };
+
+        LayerAxis LayerAxisOf(TextureTarget target) {
+            switch (target) {
+            case TextureTarget::Texture1DArray:
+                return LayerAxis::Y;
+            case TextureTarget::Texture2DArray:
+            case TextureTarget::TextureCubeMapArray:
+            case TextureTarget::Texture2DMultisampleArray:
+                return LayerAxis::Z;
+            default:
+                return LayerAxis::None;
+            }
+        }
+
+        Vector<TextureUploadTarget> UploadTargetsForViewTarget(TextureTarget target) {
+            switch (target) {
+            case TextureTarget::Texture1D:
+                return {TextureUploadTarget::Texture1D};
+            case TextureTarget::Texture2D:
+                return {TextureUploadTarget::Texture2D};
+            case TextureTarget::Texture3D:
+                return {TextureUploadTarget::Texture3D};
+            case TextureTarget::TextureRectangle:
+                return {TextureUploadTarget::TextureRectangle};
+            case TextureTarget::Texture1DArray:
+                return {TextureUploadTarget::Texture1DArray};
+            case TextureTarget::Texture2DArray:
+                return {TextureUploadTarget::Texture2DArray};
+            case TextureTarget::TextureCubeMapArray:
+                return {TextureUploadTarget::CubeMapArray};
+            case TextureTarget::Texture2DMultisample:
+                return {TextureUploadTarget::Texture2DMultisample};
+            case TextureTarget::Texture2DMultisampleArray:
+                return {TextureUploadTarget::Texture2DMultisampleArray};
+            case TextureTarget::TextureCubeMap:
+                return {TextureUploadTarget::CubeMapPositiveX, TextureUploadTarget::CubeMapNegativeX,
+                        TextureUploadTarget::CubeMapPositiveY, TextureUploadTarget::CubeMapNegativeY,
+                        TextureUploadTarget::CubeMapPositiveZ, TextureUploadTarget::CubeMapNegativeZ};
+            default:
+                MOBILEGL_ASSERT(false, "TextureObjectView: target %d cannot be a texture view", (int)target);
+                return {TextureUploadTarget::Texture2D};
+            }
+        }
+    } // namespace
+
+    TextureObjectView::TextureObjectView(Uint externalIndex, TextureTarget target,
+                                         SharedPtr<ITextureObject> storageOwner, Uint minLevel, Uint numLevels,
+                                         Uint minLayer, Uint numLayers)
+        : TextureObjectMipmap(target, externalIndex), m_storageOwner(Move(storageOwner)),
+          m_uploadTargets(UploadTargetsForViewTarget(target)) {
+        MOBILEGL_ASSERT(m_storageOwner != nullptr, "TextureObjectView: storage owner is null");
+        MOBILEGL_ASSERT(!m_storageOwner->IsTextureView(),
+                        "TextureObjectView: storage owner must be a root texture, not another view");
+        m_ownerMipmap = AsMipmapTexture(m_storageOwner.get());
+        SetViewLevelLayerRange(minLevel, numLevels, minLayer, numLayers);
+        // Held rather than forwarded so the base class's level-range clamp works against the
+        // view's OWN level count - TEXTURE_BASE_LEVEL / TEXTURE_MAX_LEVEL on a view are relative
+        // to the view. GetImmutableLevels() forwards to the owner for the actual GL query, which
+        // GL 4.6 core 8.18 defines as the ORIGINAL texture's value.
+        SetImmutableLevels(numLevels);
+    }
+
+    Uint TextureObjectView::GetImmutableLevels() const {
+        return m_storageOwner->GetImmutableLevels();
+    }
+
+    Uint64 TextureObjectView::GetContentVersion() const {
+        return m_storageOwner->GetContentVersion();
+    }
+
+    Int TextureObjectView::GetSamples() const {
+        return m_storageOwner->GetSamples();
+    }
+
+    Bool TextureObjectView::HasFixedSampleLocations() const {
+        return m_storageOwner->HasFixedSampleLocations();
+    }
+
+    TextureUploadTarget TextureObjectView::ToOwnerUploadTarget(TextureUploadTarget viewTarget) const {
+        const auto& ownerTargets = m_storageOwner->GetUploadTargets();
+        MOBILEGL_ASSERT(!ownerTargets.empty(), "TextureObjectView: storage owner has no upload target");
+        if (ownerTargets.size() == 1) {
+            // The owner keeps every layer in one blob, so there is nothing to choose.
+            return ownerTargets[0];
+        }
+        // The owner is a cube map: six independent blobs, one per face, and the view's layer
+        // index selects among them. A cube-map view of a cube map maps face to face; any other
+        // view target addresses layers, which for a cube-map owner ARE its faces.
+        const Uint faceCount = static_cast<Uint>(ownerTargets.size());
+        Uint face = m_viewMinLayer;
+        if (GetTarget() == TextureTarget::TextureCubeMap) {
+            for (Uint i = 0; i < m_uploadTargets.size(); ++i) {
+                if (m_uploadTargets[i] == viewTarget) {
+                    face = m_viewMinLayer + i;
+                    break;
+                }
+            }
+        }
+        return ownerTargets[std::min(face, faceCount - 1)];
+    }
+
+    IntVec3 TextureObjectView::ToViewLevelSize(const IntVec3& ownerLevelSize) const {
+        IntVec3 size = ownerLevelSize;
+        // Collapse whichever axis the OWNER stored its layers in down to a single slice, then
+        // impose this view's own layer count on whichever axis THIS target stores layers in.
+        // Doing it in that order makes every legal target pair fall out: 2D_ARRAY->2D clears z,
+        // 2D->2D_ARRAY sets it, 2D_ARRAY->2D_ARRAY replaces it, and 3D->3D touches neither
+        // (LayerAxis::None on both sides), which is what keeps a 3D view's full depth intact.
+        switch (LayerAxisOf(m_storageOwner->GetTarget())) {
+        case LayerAxis::Y:
+            size.y() = 1;
+            break;
+        case LayerAxis::Z:
+            size.z() = 1;
+            break;
+        case LayerAxis::None:
+            break;
+        }
+        switch (LayerAxisOf(GetTarget())) {
+        case LayerAxis::Y:
+            size.y() = static_cast<Int>(m_viewNumLayers);
+            break;
+        case LayerAxis::Z:
+            size.z() = static_cast<Int>(m_viewNumLayers);
+            break;
+        case LayerAxis::None:
+            break;
+        }
+        return size;
+    }
+
+    SizeT TextureObjectView::LayerByteOffset(TextureUploadTarget viewTarget, Uint mipmapLevel) const {
+        if (m_viewMinLayer == 0 || m_ownerMipmap == nullptr) return 0;
+        const LayerAxis ownerAxis = LayerAxisOf(m_storageOwner->GetTarget());
+        if (ownerAxis == LayerAxis::None) {
+            // A cube-map owner keeps each face in its OWN blob, and ToOwnerUploadTarget already
+            // picked the right one; a 3D or plain 2D owner has no layers to skip.
+            return 0;
+        }
+        const TextureUploadTarget ownerTarget = ToOwnerUploadTarget(viewTarget);
+        const Uint ownerLevel = ToOwnerLevel(mipmapLevel);
+        const IntVec3 ownerSize = m_ownerMipmap->GetMipmapTexelSize(ownerTarget, ownerLevel);
+        const SizeT ownerBytes = m_ownerMipmap->GetMipmapByteSize(ownerTarget, ownerLevel);
+        const SizeT ownerTexels = static_cast<SizeT>(std::max(ownerSize.x(), 0)) *
+                                  static_cast<SizeT>(std::max(ownerSize.y(), 0)) *
+                                  static_cast<SizeT>(std::max(ownerSize.z(), 1));
+        if (ownerTexels == 0 || ownerBytes == 0) return 0;
+        const SizeT bytesPerTexel = ownerBytes / ownerTexels;
+        // One "layer" is a whole x*y slice for a 2D/cube array, and a single row of `width`
+        // texels for a 1D array (whose layer count lives in the state-side height).
+        const SizeT layerTexels = ownerAxis == LayerAxis::Y
+                                      ? static_cast<SizeT>(std::max(ownerSize.x(), 0))
+                                      : static_cast<SizeT>(std::max(ownerSize.x(), 0)) *
+                                            static_cast<SizeT>(std::max(ownerSize.y(), 0));
+        const SizeT offset = static_cast<SizeT>(m_viewMinLayer) * layerTexels * bytesPerTexel;
+        return offset < ownerBytes ? offset : 0;
+    }
+
+    IntVec3 TextureObjectView::ToOwnerRegionOffset(const IntVec3& viewOffset) const {
+        if (m_viewMinLayer == 0) return viewOffset;
+        IntVec3 offset = viewOffset;
+        // The dirty region is recorded in the OWNER's blob coordinates - that is the space its
+        // upload path walks - so the view's layer origin has to be added here even though
+        // MapMipmapData hands back an already-shifted POINTER. The two are not double-counting:
+        // one moves the bytes, the other tells the owner which of its layers moved.
+        switch (LayerAxisOf(m_storageOwner->GetTarget())) {
+        case LayerAxis::Y:
+            offset.y() += static_cast<Int>(m_viewMinLayer);
+            break;
+        case LayerAxis::Z:
+            offset.z() += static_cast<Int>(m_viewMinLayer);
+            break;
+        case LayerAxis::None:
+            break;
+        }
+        return offset;
+    }
+
+    Uint TextureObjectView::GetMipmapLevelCount() const {
+        if (m_ownerMipmap == nullptr) return 0;
+        const Uint ownerLevels = m_ownerMipmap->GetMipmapLevelCount();
+        if (m_viewMinLevel >= ownerLevels) return 0;
+        return std::min(m_viewNumLevels, ownerLevels - m_viewMinLevel);
+    }
+
+    const IntVec3 TextureObjectView::GetMipmapTexelSize(TextureUploadTarget target, Uint mipmapLevel) const {
+        if (m_ownerMipmap == nullptr) return {0, 0, 0};
+        return ToViewLevelSize(
+            m_ownerMipmap->GetMipmapTexelSize(ToOwnerUploadTarget(target), ToOwnerLevel(mipmapLevel)));
+    }
+
+    const SizeT TextureObjectView::GetMipmapByteSize(TextureUploadTarget target, Uint mipmapLevel) const {
+        if (m_ownerMipmap == nullptr) return 0;
+        const TextureUploadTarget ownerTarget = ToOwnerUploadTarget(target);
+        const Uint ownerLevel = ToOwnerLevel(mipmapLevel);
+        const IntVec3 ownerSize = m_ownerMipmap->GetMipmapTexelSize(ownerTarget, ownerLevel);
+        const SizeT ownerBytes = m_ownerMipmap->GetMipmapByteSize(ownerTarget, ownerLevel);
+        const SizeT ownerTexels = static_cast<SizeT>(std::max(ownerSize.x(), 0)) *
+                                  static_cast<SizeT>(std::max(ownerSize.y(), 0)) *
+                                  static_cast<SizeT>(std::max(ownerSize.z(), 1));
+        if (ownerTexels == 0 || ownerBytes == 0) return 0;
+        // Scaled rather than recomputed from a format table: the view's internalformat is
+        // required to be in the same view class as the owner's (GL 4.6 core table 8.21), i.e. to
+        // have the identical texel size, so bytes-per-texel is shared by construction and the
+        // only difference is how many texels the view addresses.
+        const IntVec3 viewSize = ToViewLevelSize(ownerSize);
+        const SizeT viewTexels = static_cast<SizeT>(std::max(viewSize.x(), 0)) *
+                                 static_cast<SizeT>(std::max(viewSize.y(), 0)) *
+                                 static_cast<SizeT>(std::max(viewSize.z(), 1));
+        const SizeT viewBytes = (ownerBytes / ownerTexels) * viewTexels;
+        // Clamped against what remains of the owner's blob past this view's layer origin. A view
+        // whose layer window the shadow cannot lay out contiguously - several faces of a cube-map
+        // owner, which are separate blobs - would otherwise advertise more bytes than
+        // MapMipmapData can hand back, and a caller sizing a copy off this would overrun.
+        const SizeT layerOffset = LayerByteOffset(target, mipmapLevel);
+        const SizeT available = layerOffset < ownerBytes ? ownerBytes - layerOffset : 0;
+        return std::min(viewBytes, available);
+    }
+
+    void TextureObjectView::AllocateStorage(TextureUploadTarget uploadTarget, Uint mipmapLevel, MipmapInput input) {
+        // Unreachable through the API: a view is immutable from birth (GL 4.6 core 8.18 sets its
+        // TEXTURE_IMMUTABLE_FORMAT), and every entry point that would allocate is gated on
+        // ValidateTextureMutable. Forwarded rather than asserted so an internal caller that
+        // re-specifies the storage still hits the one real allocation.
+        if (m_ownerMipmap == nullptr) return;
+        m_ownerMipmap->AllocateStorage(ToOwnerUploadTarget(uploadTarget), ToOwnerLevel(mipmapLevel), input);
+    }
+
+    void TextureObjectView::TruncateMipmapLevels(TextureUploadTarget uploadTarget, Uint levelCount) {
+        if (m_ownerMipmap == nullptr) return;
+        m_ownerMipmap->TruncateMipmapLevels(ToOwnerUploadTarget(uploadTarget), ToOwnerLevel(levelCount));
+    }
+
+    void TextureObjectView::UpdateMipmapSubData(TextureUploadTarget uploadTarget, Uint mipmapLevel, DataPtr input) {
+        if (m_ownerMipmap == nullptr) return;
+        const TextureUploadTarget ownerTarget = ToOwnerUploadTarget(uploadTarget);
+        const Uint ownerLevel = ToOwnerLevel(mipmapLevel);
+        const SizeT layerOffset = LayerByteOffset(uploadTarget, mipmapLevel);
+        if (layerOffset == 0) {
+            m_ownerMipmap->UpdateMipmapSubData(ownerTarget, ownerLevel, input);
+            return;
+        }
+        // The owner's whole-level write starts at ITS level origin, which for a layer-sliced view
+        // is the wrong place: writing there would silently overwrite the parent's layers 0..n
+        // instead of the window this view opened. Write through the shifted pointer instead, and
+        // mark exactly the layers that moved.
+        auto* destination = static_cast<Uint8*>(m_ownerMipmap->MapMipmapData(ownerTarget, ownerLevel));
+        if (destination == nullptr || input.data == nullptr || input.size == 0) return;
+        const SizeT capacity = GetMipmapByteSize(uploadTarget, mipmapLevel);
+        std::memcpy(destination + layerOffset, input.data, std::min(input.size, capacity));
+        const IntVec3 viewSize = GetMipmapTexelSize(uploadTarget, mipmapLevel);
+        MarkStorageDirtyRegion(uploadTarget, mipmapLevel, IntVec3{0, 0, 0},
+                               IntVec3{viewSize.x(), viewSize.y(), std::max(viewSize.z(), 1)});
+    }
+
+    void* TextureObjectView::MapMipmapData(TextureUploadTarget uploadTarget, Uint mipmapLevel) {
+        if (m_ownerMipmap == nullptr) return nullptr;
+        auto* data = static_cast<Uint8*>(
+            m_ownerMipmap->MapMipmapData(ToOwnerUploadTarget(uploadTarget), ToOwnerLevel(mipmapLevel)));
+        if (data == nullptr) return nullptr;
+        // Shifted to the view's first LAYER, so that a caller which maps this pointer and then
+        // offsets into it using the extents GetMipmapTexelSize reports - which is what every
+        // glTexSubImage*/glGetTexImage path does - lands on the layers this view addresses rather
+        // than on the parent's first ones.
+        return data + LayerByteOffset(uploadTarget, mipmapLevel);
+    }
+
+    void TextureObjectView::MarkStorageDirty(TextureUploadTarget uploadTarget, Uint mipmapLevel, Bool dirty) {
+        if (m_ownerMipmap == nullptr) return;
+        m_ownerMipmap->MarkStorageDirty(ToOwnerUploadTarget(uploadTarget), ToOwnerLevel(mipmapLevel), dirty);
+    }
+
+    Bool TextureObjectView::IsStorageDirty(TextureUploadTarget uploadTarget, Uint mipmapLevel) const {
+        if (m_ownerMipmap == nullptr) return false;
+        return m_ownerMipmap->IsStorageDirty(ToOwnerUploadTarget(uploadTarget), ToOwnerLevel(mipmapLevel));
+    }
+
+    void TextureObjectView::MarkStorageDirtyRegion(TextureUploadTarget uploadTarget, Uint mipmapLevel, IntVec3 offset,
+                                                   IntVec3 size) {
+        if (m_ownerMipmap == nullptr) return;
+        m_ownerMipmap->MarkStorageDirtyRegion(ToOwnerUploadTarget(uploadTarget), ToOwnerLevel(mipmapLevel),
+                                              ToOwnerRegionOffset(offset), size);
+    }
+
+    MipmapDirtyRegion TextureObjectView::GetStorageDirtyRegion(TextureUploadTarget uploadTarget,
+                                                               Uint mipmapLevel) const {
+        if (m_ownerMipmap == nullptr) return {};
+        return m_ownerMipmap->GetStorageDirtyRegion(ToOwnerUploadTarget(uploadTarget), ToOwnerLevel(mipmapLevel));
+    }
+
+    void TextureObjectView::SetMipmapCompressedImage(TextureUploadTarget uploadTarget, Uint mipmapLevel,
+                                                     GLenum internalFormat, const void* data, SizeT size) {
+        if (m_ownerMipmap == nullptr) return;
+        m_ownerMipmap->SetMipmapCompressedImage(ToOwnerUploadTarget(uploadTarget), ToOwnerLevel(mipmapLevel),
+                                                internalFormat, data, size);
+    }
+
+    GLenum TextureObjectView::GetMipmapCompressedFormat(TextureUploadTarget uploadTarget, Uint mipmapLevel) const {
+        if (m_ownerMipmap == nullptr) return GL_NONE;
+        return m_ownerMipmap->GetMipmapCompressedFormat(ToOwnerUploadTarget(uploadTarget), ToOwnerLevel(mipmapLevel));
+    }
+
+    SizeT TextureObjectView::GetMipmapCompressedByteSize(TextureUploadTarget uploadTarget, Uint mipmapLevel) const {
+        if (m_ownerMipmap == nullptr) return 0;
+        return m_ownerMipmap->GetMipmapCompressedByteSize(ToOwnerUploadTarget(uploadTarget), ToOwnerLevel(mipmapLevel));
+    }
+
+    const void* TextureObjectView::MapMipmapCompressedImage(TextureUploadTarget uploadTarget, Uint mipmapLevel) const {
+        if (m_ownerMipmap == nullptr) return nullptr;
+        return m_ownerMipmap->MapMipmapCompressedImage(ToOwnerUploadTarget(uploadTarget), ToOwnerLevel(mipmapLevel));
+    }
+
+    void TextureObjectView::SetMipmapRequestedCompressedFormat(TextureUploadTarget uploadTarget, Uint mipmapLevel,
+                                                               GLenum internalFormat) {
+        if (m_ownerMipmap == nullptr) return;
+        m_ownerMipmap->SetMipmapRequestedCompressedFormat(ToOwnerUploadTarget(uploadTarget), ToOwnerLevel(mipmapLevel),
+                                                          internalFormat);
+    }
+
+    GLenum TextureObjectView::GetMipmapRequestedCompressedFormat(TextureUploadTarget uploadTarget,
+                                                                 Uint mipmapLevel) const {
+        if (m_ownerMipmap == nullptr) return GL_NONE;
+        return m_ownerMipmap->GetMipmapRequestedCompressedFormat(ToOwnerUploadTarget(uploadTarget),
+                                                                 ToOwnerLevel(mipmapLevel));
+    }
+
+    IntVec3 TextureObjectView::GetBaseSize() const {
+        if (GetMipmapLevelCount() == 0) return {0, 0, 0};
+        return GetMipmapTexelSize(m_uploadTargets[0], 0);
+    }
+
+    Bool TextureObjectView::IsComplete() const {
+        if (!TextureObjectBase::IsComplete()) return false;
+        // The view's own level set is what sampling walks, and it can be shorter than the
+        // owner's. Everything below it - that the owner has real storage at all - is the owner's
+        // answer, because these texels are its texels.
+        if (GetMipmapLevelCount() == 0) return false;
+        return m_storageOwner->IsComplete();
+    }
+
+    Uint TextureObjectView::GetIndexOfTextureUploadTarget(TextureUploadTarget target) const {
+        for (Uint i = 0; i < static_cast<Uint>(m_uploadTargets.size()); ++i) {
+            if (m_uploadTargets[i] == target) return i;
+        }
+        return 0;
+    }
+} // namespace MobileGL::MG_State::GLState

MobileGL/MG_State/GLState/TextureState/TextureObjectView.h

@@ -0,0 +1,126 @@
+// MobileGL - MobileGL/MG_State/GLState/TextureState/TextureObjectView.h
+// Copyright (c) 2025-2026 MobileGL-Dev
+// Licensed under the GNU Lesser General Public License v3.0:
+//   https://www.gnu.org/licenses/gpl-3.0.txt
+//   https://www.gnu.org/licenses/lgpl-3.0.txt
+// SPDX-License-Identifier: LGPL-3.0-only
+// End of Source File Header
+
+#pragma once
+#include "TextureObject.h"
+
+namespace MobileGL::MG_State::GLState {
+    // A texture created by glTextureView (ARB_texture_view / GL 4.6 core 8.18): a texture object
+    // in every respect - own name, own target, own internal format, own sampler and own
+    // per-texture parameters - whose TEXELS are somebody else's. That last part is the whole
+    // point of the extension, and the reason this cannot be a plain TextureObject2D with a copy:
+    // the application samples the view and the original SIMULTANEOUSLY, reading different aspects
+    // or different formats out of one storage, and writes through either name must be visible
+    // through the other.
+    //
+    // So this class owns no MipmapStorage at all. Every storage question is answered by
+    // m_storageOwner, shifted by the view's level offset; every parameter question is answered
+    // by this object's own TextureObjectBase state. The owner is held by SharedPtr, which is
+    // exactly GL's name-deletion rule (5.1.2): glDeleteTextures on the original frees the NAME
+    // immediately, but the storage - and every backend resource keyed on the owner object -
+    // lives until the last view referencing it is gone too.
+    //
+    // m_storageOwner is guaranteed never to be a view itself. glTextureView composes a
+    // view-of-a-view onto the root at creation time, which is what the spec's additive
+    // "<minlevel> plus the value of TEXTURE_VIEW_MIN_LEVEL from the original texture" rule
+    // means; one hop therefore always reaches real storage and no recursion is possible.
+    //
+    // LEVEL offsets are applied by shifting the level index; LAYER offsets cannot be, because the
+    // TextureObjectMipmap interface addresses storage as (upload target, level) and a layer lives
+    // INSIDE a level's blob. They are applied two other ways instead, and the pair is what keeps
+    // a layer-sliced view from corrupting its parent:
+    //   * MapMipmapData returns a pointer already advanced to the view's first layer, so a caller
+    //     that maps it and then offsets using the extents GetMipmapTexelSize reports - which is
+    //     what every glTexSubImage*/glGetTexImage path does - writes the layers it meant to; and
+    //   * MarkStorageDirtyRegion moves the region's origin into the OWNER's layer space, which is
+    //     the space its upload path walks.
+    // Those two are not double-counting: one moves the bytes, the other names which of the
+    // owner's layers moved.
+    class TextureObjectView : public TextureObjectMipmap {
+    public:
+        TextureObjectView(Uint externalIndex, TextureTarget target, SharedPtr<ITextureObject> storageOwner,
+                          Uint minLevel, Uint numLevels, Uint minLayer, Uint numLayers);
+
+        const SharedPtr<ITextureObject>& GetViewStorageOwner() const override { return m_storageOwner; }
+        const Vector<TextureUploadTarget>& GetUploadTargets() const override { return m_uploadTargets; }
+
+        // A view is immutable from birth (GL 4.6 core 8.18 sets its TEXTURE_IMMUTABLE_FORMAT), and
+        // unconditionally so: the base class infers immutability from a non-zero level count, and
+        // a degenerate view - one the spec's min() composition narrowed to zero levels - would
+        // otherwise report GL_FALSE, walk straight past ValidateTextureMutable and let
+        // glTexImage2D respecify the PARENT's immutable storage through AllocateStorage.
+        Bool IsImmutable() const override { return true; }
+
+        // GL 4.6 core 8.18: "TEXTURE_IMMUTABLE_LEVELS is set to the value of
+        // TEXTURE_IMMUTABLE_LEVELS from the original texture" - NOT to <numlevels>. Kept as a
+        // forward rather than in m_immutableLevels so that the base class's level-range clamp
+        // keeps using the view's own level count, which is what TEXTURE_BASE_LEVEL /
+        // TEXTURE_MAX_LEVEL on a view are relative to.
+        Uint GetImmutableLevels() const override;
+
+        // Both follow the storage, not this object: a backend that memoised on the view's own
+        // counter would keep serving stale texels after the owner was written through its own
+        // name (KHR-GL43.texture_view.coherency is exactly this test).
+        Uint64 GetContentVersion() const override;
+        Int GetSamples() const override;
+        Bool HasFixedSampleLocations() const override;
+
+        Uint GetMipmapLevelCount() const override;
+        const IntVec3 GetMipmapTexelSize(TextureUploadTarget target, Uint mipmapLevel) const override;
+        const SizeT GetMipmapByteSize(TextureUploadTarget target, Uint mipmapLevel) const override;
+        void AllocateStorage(TextureUploadTarget uploadTarget, Uint mipmapLevel, MipmapInput input) override;
+        void TruncateMipmapLevels(TextureUploadTarget uploadTarget, Uint levelCount) override;
+        void UpdateMipmapSubData(TextureUploadTarget uploadTarget, Uint mipmapLevel, DataPtr input) override;
+        void* MapMipmapData(TextureUploadTarget uploadTarget, Uint mipmapLevel) override;
+        void MarkStorageDirty(TextureUploadTarget uploadTarget, Uint mipmapLevel, Bool dirty) override;
+        Bool IsStorageDirty(TextureUploadTarget uploadTarget, Uint mipmapLevel) const override;
+        void MarkStorageDirtyRegion(TextureUploadTarget uploadTarget, Uint mipmapLevel, IntVec3 offset,
+                                    IntVec3 size) override;
+        MipmapDirtyRegion GetStorageDirtyRegion(TextureUploadTarget uploadTarget, Uint mipmapLevel) const override;
+        void SetMipmapCompressedImage(TextureUploadTarget uploadTarget, Uint mipmapLevel, GLenum internalFormat,
+                                      const void* data, SizeT size) override;
+        GLenum GetMipmapCompressedFormat(TextureUploadTarget uploadTarget, Uint mipmapLevel) const override;
+        SizeT GetMipmapCompressedByteSize(TextureUploadTarget uploadTarget, Uint mipmapLevel) const override;
+        const void* MapMipmapCompressedImage(TextureUploadTarget uploadTarget, Uint mipmapLevel) const override;
+        void SetMipmapRequestedCompressedFormat(TextureUploadTarget uploadTarget, Uint mipmapLevel,
+                                                GLenum internalFormat) override;
+        GLenum GetMipmapRequestedCompressedFormat(TextureUploadTarget uploadTarget, Uint mipmapLevel) const override;
+
+        IntVec3 GetBaseSize() const override;
+        Bool IsComplete() const override;
+
+    protected:
+        Uint GetIndexOfTextureUploadTarget(TextureUploadTarget target) const override;
+
+    private:
+        // The owner-side upload target a given view-side one addresses. Only GL_TEXTURE_CUBE_MAP
+        // stores its six faces as six separate blobs (MipmapUploadTargetArray<6>); every other
+        // target - arrays and cube-map arrays included - keeps all its layers in one blob, so
+        // the mapping is "the owner's only target" unless one of the two sides is a cube map.
+        TextureUploadTarget ToOwnerUploadTarget(TextureUploadTarget viewTarget) const;
+        Uint ToOwnerLevel(Uint viewLevel) const { return m_viewMinLevel + viewLevel; }
+        // The owner's level extent rewritten into this view's shape: the owner's layer axis is
+        // collapsed to one slice and the view's own layer count is imposed on the view's layer
+        // axis. A GL 1D array carries its layer count in the state-side HEIGHT while every other
+        // layered target carries it in z, so the axis is target-dependent.
+        IntVec3 ToViewLevelSize(const IntVec3& ownerLevelSize) const;
+        // Where this view's first LAYER starts inside the owner's level blob. The layer axis a
+        // level's bytes are laid out along is the OWNER's, so this is a slice for a 2D/cube array
+        // and a single row for a 1D array; a cube-map owner returns 0 because its faces are
+        // separate blobs that ToOwnerUploadTarget already selects between.
+        SizeT LayerByteOffset(TextureUploadTarget viewTarget, Uint mipmapLevel) const;
+        // A dirty-region origin moved from the view's layer space into the owner's.
+        IntVec3 ToOwnerRegionOffset(const IntVec3& viewOffset) const;
+
+        SharedPtr<ITextureObject> m_storageOwner;
+        // Non-owning; m_storageOwner keeps it alive and is never a view, so this is set once in
+        // the constructor and is null only for the (rejected at creation) buffer-texture case.
+        TextureObjectMipmap* m_ownerMipmap = nullptr;
+        Vector<TextureUploadTarget> m_uploadTargets;
+    };
+} // namespace MobileGL::MG_State::GLState

MobileGL/MG_State/GLState/TextureState/TextureState.cpp

@@ -18,6 +18,7 @@
 #include "TextureObject2DCube.h"
 #include "TextureObjectBuffer.h"
 #include "TextureObjectStubs.h"
+#include "TextureObjectView.h"
 
 namespace MobileGL::MG_State::GLState {
     static std::atomic<Uint64> s_nextTextureStateContextId = 1;
@@ -104,6 +105,16 @@ namespace MobileGL::MG_State::GLState {
         return textureObject;
     }
 
+    const SharedPtr<ITextureObject>& TextureState::CreateTextureViewObject(
+        Uint index, TextureTarget target, const SharedPtr<ITextureObject>& storageOwner, Uint minLevel,
+        Uint numLevels, Uint minLayer, Uint numLayers) {
+        MOBILEGL_ASSERT(storageOwner != nullptr, "CreateTextureViewObject: storage owner is null");
+        auto& textureObject = m_textureObjects[index];
+        textureObject = MakeShared<TextureObjectView>(index, target, storageOwner, minLevel, numLevels, minLayer,
+                                                      numLayers);
+        return textureObject;
+    }
+
     void TextureState::MarkTextureObjectForDeletion(Uint index, Bool keepUnboundReservation) {
         if (m_indexGenerator.IsValid(index)) {
             auto it = m_textureObjects.find(index);

MobileGL/MG_State/GLState/TextureState/TextureState.h

@@ -48,6 +48,13 @@ namespace MobileGL::MG_State::GLState {
         TextureState();
         void GenerateNames(Uint number, Vector<Uint>& textures);
         const SharedPtr<ITextureObject>& CreateTextureObject(Uint index, TextureTarget target);
+        // glTextureView (GL 4.6 core 8.18). `storageOwner` must already be a texture with
+        // immutable storage and must NOT itself be a view - the caller composes a view-of-a-view
+        // onto the root first, and passes the composed (root-relative) level/layer range here.
+        const SharedPtr<ITextureObject>& CreateTextureViewObject(Uint index, TextureTarget target,
+                                                                 const SharedPtr<ITextureObject>& storageOwner,
+                                                                 Uint minLevel, Uint numLevels, Uint minLayer,
+                                                                 Uint numLayers);
         const SharedPtr<ITextureObject>& GetTextureObject(Uint index);
         // The context's default texture object (name 0) for `target`. GL 3.3 core 3.8: texture
         // zero names a real, per-target texture object shared by every texture unit; binding 0

MobileGL/MG_Test/Backend/DirectGLES/EsslShaderPassTest.cpp

@@ -15,6 +15,8 @@
 
 #include <MG_Backend/DirectGLES/Utils.h>
 
+#include <limits>
+
 using namespace MobileGL;
 using MobileGL::MG_Backend::DirectGLES::PrgramImpl::BakeImageFormatQualifiers;
 using MobileGL::MG_Backend::DirectGLES::PrgramImpl::BuildPassthroughTessControlEssl;
@@ -1296,8 +1298,13 @@ void main() { gl_ViewportIndex = 1; imageStore(uni_image, ivec2(0), uvec4(1u));
 // rather than pick a shape, because a redeclaration that disagrees with the stage it feeds is an
 // ES link error against a program that has nothing else wrong with it.
 
+namespace {
+    const FloatVec4 kDefaultOuter(1.0f, 1.0f, 1.0f, 1.0f);
+    const FloatVec2 kDefaultInner(1.0f, 1.0f);
+} // namespace
+
 TEST(PassthroughTessControlEsslTest, DeclaresThePatchSizeAndWritesEveryTessLevel) {
-    const String out = BuildPassthroughTessControlEssl(320, 4, "", "");
+    const String out = BuildPassthroughTessControlEssl(320, 4, "", "", kDefaultOuter, kDefaultInner);
     EXPECT_EQ(out.find("#version 320 es"), 0u) << out;
     EXPECT_TRUE(Contains(out, "layout(vertices = 4) out;")) << out;
     EXPECT_TRUE(Contains(out,
@@ -1314,10 +1321,63 @@ TEST(PassthroughTessControlEsslTest, DeclaresThePatchSizeAndWritesEveryTessLevel
     EXPECT_FALSE(Contains(out, "gl_PerVertex")) << out;
 }
 
+// glPatchParameterfv's state is compiled INTO this stage: ES has no PATCH_DEFAULT_*_LEVEL and no
+// entry point to forward it to, so a generator that ignored these arguments would tessellate every
+// control-stage-less program at level 1 whatever the application asked for.
+TEST(PassthroughTessControlEsslTest, BakesTheDefaultTessLevelsIn) {
+    const String out = BuildPassthroughTessControlEssl(320, 4, "", "", FloatVec4(2.0f, 3.0f, 4.0f, 5.0f),
+                                                       FloatVec2(6.5f, 7.25f));
+    EXPECT_TRUE(Contains(out, "gl_TessLevelOuter[0] = 2.0;")) << out;
+    EXPECT_TRUE(Contains(out, "gl_TessLevelOuter[1] = 3.0;")) << out;
+    EXPECT_TRUE(Contains(out, "gl_TessLevelOuter[2] = 4.0;")) << out;
+    EXPECT_TRUE(Contains(out, "gl_TessLevelOuter[3] = 5.0;")) << out;
+    EXPECT_TRUE(Contains(out, "gl_TessLevelInner[0] = 6.5;")) << out;
+    EXPECT_TRUE(Contains(out, "gl_TessLevelInner[1] = 7.25;")) << out;
+}
+
+// Every level literal carries a decimal point even when the value is integral: ESSL reads
+// `gl_TessLevelOuter[0] = 1;` as an int assigned to a float and refuses to compile the stage,
+// which would take the whole program down with it.
+TEST(PassthroughTessControlEsslTest, SpellsIntegralLevelsAsFloatLiterals) {
+    const String out = BuildPassthroughTessControlEssl(320, 4, "", "", FloatVec4(2.0f, 2.0f, 2.0f, 2.0f),
+                                                       FloatVec2(2.0f, 2.0f));
+    EXPECT_FALSE(Contains(out, "= 2;")) << out;
+}
+
+// glPatchParameterfv accepts any float, NaN and infinity included, and GL 4.6 core 11.2.2
+// discards a patch ONLY when a relevant outer level is <= 0 - everything else is clamped into
+// [1, MAX_TESS_GEN_LEVEL]. So the three non-finite inputs do not share one answer: NaN is
+// unspecified and 0.0 is the safe reading, -inf really does discard, and +inf must tessellate at
+// the maximum. Baking 0.0 for +inf inverted "as finely as possible" into "draw nothing".
+TEST(PassthroughTessControlEsslTest, NonFiniteLevelsFollowTheDiscardRule) {
+    const Float notANumber = std::numeric_limits<Float>::quiet_NaN();
+    const Float infinity = std::numeric_limits<Float>::infinity();
+    const String out = BuildPassthroughTessControlEssl(320, 4, "", "",
+                                                       FloatVec4(notANumber, -infinity, infinity, 1.0f),
+                                                       FloatVec2(notANumber, 1.0f));
+    EXPECT_TRUE(Contains(out, "gl_TessLevelOuter[0] = 0.0;")) << out;
+    EXPECT_TRUE(Contains(out, "gl_TessLevelOuter[1] = 0.0;")) << out;
+    EXPECT_FALSE(Contains(out, "gl_TessLevelOuter[2] = 0.0;"))
+        << "a positive infinity clamps to GL_MAX_TESS_GEN_LEVEL, not to a discarded patch" << out;
+    EXPECT_TRUE(Contains(out, "gl_TessLevelInner[0] = 0.0;")) << out;
+    EXPECT_FALSE(Contains(out, "nan")) << out;
+    EXPECT_FALSE(Contains(out, "inf")) << out;
+}
+
+// A level below the old six-decimal format's resolution is still a POSITIVE level, which GL clamps
+// to 1 and draws; rendering it as "0.000000" discarded the patch instead.
+TEST(PassthroughTessControlEsslTest, TinyPositiveLevelsDoNotFlushToZero) {
+    const String out = BuildPassthroughTessControlEssl(320, 4, "", "",
+                                                       FloatVec4(1e-7f, 1.0f, 1.0f, 1.0f),
+                                                       FloatVec2(1.0f, 1.0f));
+    EXPECT_FALSE(Contains(out, "gl_TessLevelOuter[0] = 0.0;")) << out;
+    EXPECT_FALSE(Contains(out, "gl_TessLevelOuter[0] = 0.000000;")) << out;
+}
+
 // ES 3.1 reaches tessellation only through the extension; the caller has already established
 // that the driver runs the evaluation stage at all, so the only question is the spelling.
 TEST(PassthroughTessControlEsslTest, RequestsTheExtensionBelowEs32) {
-    const String out = BuildPassthroughTessControlEssl(310, 3, "", "");
+    const String out = BuildPassthroughTessControlEssl(310, 3, "", "", kDefaultOuter, kDefaultInner);
     EXPECT_EQ(out.find("#version 310 es"), 0u) << out;
     EXPECT_TRUE(Contains(out, "#extension GL_EXT_tessellation_shader : require")) << out;
 }
@@ -1325,7 +1385,7 @@ TEST(PassthroughTessControlEsslTest, RequestsTheExtensionBelowEs32) {
 TEST(PassthroughTessControlEsslTest, MirrorsTheNeighboursPerVertexBlocks) {
     const String inMembers = " highp vec4 gl_Position; highp float gl_PointSize; ";
     const String outMembers = " highp vec4 gl_Position; ";
-    const String out = BuildPassthroughTessControlEssl(320, 4, inMembers, outMembers);
+    const String out = BuildPassthroughTessControlEssl(320, 4, inMembers, outMembers, kDefaultOuter, kDefaultInner);
     EXPECT_TRUE(Contains(out, "in gl_PerVertex {" + inMembers + "} gl_in[gl_MaxPatchVertices];")) << out;
     EXPECT_TRUE(Contains(out, "out gl_PerVertex {" + outMembers + "} gl_out[];")) << out;
 }

MobileGL/MG_Test/BackendLoader/BackendLoaderTest.cpp

@@ -1087,22 +1087,76 @@ TEST(TextureAnisotropyCapabilities, ExtensionIsAdvertisedOnlyWhenTheHostDriverSu
         return std::find(extensions.begin(), extensions.end(), wanted) != extensions.end();
     };
 
-    const auto without = MobileGL::MG_Backend::DirectGLES::BuildAdvertisedExtensions(false, false, false, false);
+    const auto without = MobileGL::MG_Backend::DirectGLES::BuildAdvertisedExtensions(false, false, false, false, false, false);
     EXPECT_FALSE(contains(without, MobileGL::E_GL_EXT_texture_filter_anisotropic));
     EXPECT_FALSE(contains(without, MobileGL::E_GL_ARB_texture_filter_anisotropic));
 
-    const auto with = MobileGL::MG_Backend::DirectGLES::BuildAdvertisedExtensions(false, true, false, false);
+    const auto with = MobileGL::MG_Backend::DirectGLES::BuildAdvertisedExtensions(false, true, false, false, false, false);
     EXPECT_TRUE(contains(with, MobileGL::E_GL_EXT_texture_filter_anisotropic));
     EXPECT_TRUE(contains(with, MobileGL::E_GL_ARB_texture_filter_anisotropic));
 
     // Same rule on the Vulkan backend, where the gate is the samplerAnisotropy device feature.
-    const auto vkWithout = MobileGL::MG_Backend::DirectVulkan::BuildAdvertisedExtensions(false, false, false, false);
+    const auto vkWithout = MobileGL::MG_Backend::DirectVulkan::BuildAdvertisedExtensions(false, false, false, false, false);
     EXPECT_FALSE(contains(vkWithout, MobileGL::E_GL_EXT_texture_filter_anisotropic));
-    const auto vkWith = MobileGL::MG_Backend::DirectVulkan::BuildAdvertisedExtensions(false, false, true, false);
+    const auto vkWith = MobileGL::MG_Backend::DirectVulkan::BuildAdvertisedExtensions(false, false, true, false, false);
     EXPECT_TRUE(contains(vkWith, MobileGL::E_GL_EXT_texture_filter_anisotropic));
     EXPECT_TRUE(contains(vkWith, MobileGL::E_GL_ARB_texture_filter_anisotropic));
 }
 
+// Cube map arrays are core at the version MobileGL claims, but there is nothing underneath on a
+// pre-ES-3.2 driver without EXT/OES_texture_cube_map_array, and no VK_IMAGE_VIEW_TYPE_CUBE_ARRAY
+// without the imageCubeArray feature. The string has to follow the capability on both backends -
+// and it has to BE there when the capability is, because KHR-GL4*.texture_gather.*-cube-array
+// gates on the string with no core-version fallback.
+TEST(CubeMapArrayAdvertisement, FollowsTheHostCapabilityOnBothBackends) {
+    const auto contains = [](const MobileGL::Vector<MobileGL::GLExtension>& extensions,
+                             MobileGL::GLExtension wanted) {
+        return std::find(extensions.begin(), extensions.end(), wanted) != extensions.end();
+    };
+
+    const auto esWithout =
+        MobileGL::MG_Backend::DirectGLES::BuildAdvertisedExtensions(false, false, false, false, false, false);
+    EXPECT_FALSE(contains(esWithout, MobileGL::E_GL_ARB_texture_cube_map_array));
+    const auto esWith =
+        MobileGL::MG_Backend::DirectGLES::BuildAdvertisedExtensions(false, false, false, false, false, true);
+    EXPECT_TRUE(contains(esWith, MobileGL::E_GL_ARB_texture_cube_map_array));
+
+    const auto vkWithout = MobileGL::MG_Backend::DirectVulkan::BuildAdvertisedExtensions(false, false, false, false,
+                                                                                         false);
+    EXPECT_FALSE(contains(vkWithout, MobileGL::E_GL_ARB_texture_cube_map_array));
+    const auto vkWith = MobileGL::MG_Backend::DirectVulkan::BuildAdvertisedExtensions(false, false, false, false, true);
+    EXPECT_TRUE(contains(vkWith, MobileGL::E_GL_ARB_texture_cube_map_array));
+}
+
+// The core-plumbing strings carry no capability gate: they name entry points that have been real
+// on both backends for as long as the backends have existed, and an application that gates its
+// entry-point resolution on the string (LWJGL does) would otherwise call through null. Pinned
+// together so a future edit cannot quietly drop one, and pinned on BOTH backends so the two
+// cannot disagree about what MobileGL is.
+TEST(CorePlumbingAdvertisement, IsUnconditionalAndIdenticalOnBothBackends) {
+    const auto contains = [](const MobileGL::Vector<MobileGL::GLExtension>& extensions,
+                             MobileGL::GLExtension wanted) {
+        return std::find(extensions.begin(), extensions.end(), wanted) != extensions.end();
+    };
+    const MobileGL::GLExtension expected[] = {
+        MobileGL::E_GL_ARB_sync,          MobileGL::E_GL_ARB_shader_atomic_counters,
+        MobileGL::E_GL_ARB_vertex_array_object, MobileGL::E_GL_ARB_sampler_objects,
+        MobileGL::E_GL_ARB_map_buffer_range,    MobileGL::E_GL_ARB_copy_buffer,
+        MobileGL::E_GL_ARB_copy_image,          MobileGL::E_GL_ARB_texture_swizzle,
+        MobileGL::E_GL_ARB_vertex_type_2_10_10_10_rev, MobileGL::E_GL_ARB_texture_rg,
+        MobileGL::E_GL_ARB_depth_buffer_float,  MobileGL::E_GL_ARB_texture_float,
+        MobileGL::E_GL_ARB_viewport_array};
+
+    // Every gate off: none of these may depend on one.
+    const auto es = MobileGL::MG_Backend::DirectGLES::BuildAdvertisedExtensions(false, false, false, false, false,
+                                                                                false);
+    const auto vk = MobileGL::MG_Backend::DirectVulkan::BuildAdvertisedExtensions(false, false, false, false, false);
+    for (const auto extension : expected) {
+        EXPECT_TRUE(contains(es, extension)) << "DirectGLES stopped advertising extension " << extension;
+        EXPECT_TRUE(contains(vk, extension)) << "DirectVulkan stopped advertising extension " << extension;
+    }
+}
+
 // Minecraft 26.3 checks ARB_draw_indirect before it considers the already-advertised
 // ARB_multi_draw_indirect, then separately requires ARB_base_instance before enabling its terrain
 // indirect path. Pin both strings and, just as importantly, the non-zero firstInstance gate.
@@ -1113,27 +1167,27 @@ TEST(IndirectDrawAdvertisement, MatchesEachBackendsUsableCommandSemantics) {
     };
 
     const auto esWithoutIndirect =
-        MobileGL::MG_Backend::DirectGLES::BuildAdvertisedExtensions(false, false, false, false);
+        MobileGL::MG_Backend::DirectGLES::BuildAdvertisedExtensions(false, false, false, false, false, false);
     EXPECT_FALSE(contains(esWithoutIndirect, MobileGL::E_GL_ARB_draw_indirect));
     EXPECT_FALSE(contains(esWithoutIndirect, MobileGL::E_GL_ARB_base_instance));
 
     const auto esWithoutBaseInstance =
-        MobileGL::MG_Backend::DirectGLES::BuildAdvertisedExtensions(false, false, true, false);
+        MobileGL::MG_Backend::DirectGLES::BuildAdvertisedExtensions(false, false, true, false, false, false);
     EXPECT_TRUE(contains(esWithoutBaseInstance, MobileGL::E_GL_ARB_draw_indirect));
     EXPECT_FALSE(contains(esWithoutBaseInstance, MobileGL::E_GL_ARB_base_instance));
 
     const auto esWithBoth =
-        MobileGL::MG_Backend::DirectGLES::BuildAdvertisedExtensions(false, false, true, true);
+        MobileGL::MG_Backend::DirectGLES::BuildAdvertisedExtensions(false, false, true, true, false, false);
     EXPECT_TRUE(contains(esWithBoth, MobileGL::E_GL_ARB_draw_indirect));
     EXPECT_TRUE(contains(esWithBoth, MobileGL::E_GL_ARB_base_instance));
 
     const auto vkWithoutBaseInstance =
-        MobileGL::MG_Backend::DirectVulkan::BuildAdvertisedExtensions(false, false, false, false);
+        MobileGL::MG_Backend::DirectVulkan::BuildAdvertisedExtensions(false, false, false, false, false);
     EXPECT_TRUE(contains(vkWithoutBaseInstance, MobileGL::E_GL_ARB_draw_indirect));
     EXPECT_FALSE(contains(vkWithoutBaseInstance, MobileGL::E_GL_ARB_base_instance));
 
     const auto vkWithBoth =
-        MobileGL::MG_Backend::DirectVulkan::BuildAdvertisedExtensions(false, false, false, true);
+        MobileGL::MG_Backend::DirectVulkan::BuildAdvertisedExtensions(false, false, false, true, false);
     EXPECT_TRUE(contains(vkWithBoth, MobileGL::E_GL_ARB_draw_indirect));
     EXPECT_TRUE(contains(vkWithBoth, MobileGL::E_GL_ARB_base_instance));
 }

MobileGL/MG_Test/Buffer/BufferTest.cpp

@@ -711,6 +711,62 @@ TEST_F(BufferTest, ClearNamedBufferSubDataRepeatsPattern) {
     EXPECT_EQ(actual, (Vector<Uint32>{0, pattern, pattern, pattern, 0}));
     EXPECT_EQ(MobileGL::MG_Impl::GLImpl::GetError(), GL_NO_ERROR);
 }
+
+// GL 4.6 core table 8.2 pairs GL_INT with the non-integer base formats as a signed-normalized
+// source, so a GL_R8 clear whose pattern arrives as (GL_RED, GL_INT) is legal. The pair used to be
+// rejected with INVALID_VALUE, which is the first call
+// KHR-GL45.direct_state_access.buffers_functional makes.
+TEST_F(BufferTest, ClearNamedBufferSubDataAcceptsSignedNormalizedIntPattern) {
+    GLuint buffer = 0;
+    MobileGL::MG_Impl::GLImpl::CreateBuffers(1, &buffer);
+
+    const Vector<Uint8> initial(24, 0x7F);
+    MobileGL::MG_Impl::GLImpl::NamedBufferStorage(
+        buffer, initial.size(), initial.data(),
+        GL_MAP_READ_BIT | GL_MAP_WRITE_BIT | GL_DYNAMIC_STORAGE_BIT | GL_MAP_PERSISTENT_BIT);
+    ASSERT_EQ(MobileGL::MG_Impl::GLImpl::GetError(), GL_NO_ERROR);
+
+    const GLint zero = 0;
+    MobileGL::MG_Impl::GLImpl::ClearNamedBufferSubData(buffer, GL_R8, 0, sizeof(GLint), GL_RED, GL_INT, &zero);
+    EXPECT_EQ(MobileGL::MG_Impl::GLImpl::GetError(), GL_NO_ERROR);
+
+    Vector<Uint8> actual(initial.size());
+    auto bufferObject = MobileGL::MG_State::pGLContext->GetBufferObject(buffer);
+    ASSERT_NE(bufferObject, nullptr);
+    Memcpy(actual.data(), bufferObject->AcquireMemory(false, true, false), actual.size());
+    Vector<Uint8> expected(initial);
+    for (SizeT i = 0; i < sizeof(GLint); ++i) expected[i] = 0;
+    EXPECT_EQ(actual, expected);
+
+    MobileGL::MG_Impl::GLImpl::DeleteBuffers(1, &buffer);
+    DrainPendingGlErrors();
+}
+
+// The same pair on the bound-target entry point: the DSA and the bound call share
+// ClearBufferRange_State, and a regression in either direction has to show up here too.
+TEST_F(BufferTest, ClearBufferSubDataAcceptsSignedNormalizedIntPattern) {
+    GLuint buffer = 0;
+    MobileGL::MG_Impl::GLImpl::GenBuffers(1, &buffer);
+    MobileGL::MG_Impl::GLImpl::BindBuffer(GL_ARRAY_BUFFER, buffer);
+
+    const Vector<Uint8> initial(8, 0x7F);
+    MobileGL::MG_Impl::GLImpl::BufferData(GL_ARRAY_BUFFER, initial.size(), initial.data(), GL_STATIC_DRAW);
+    // GL_INT is signed-normalized against 2^31-1, so the maximum maps to a saturated GL_R8 texel.
+    const GLint one = 2147483647;
+    MobileGL::MG_Impl::GLImpl::ClearBufferSubData(GL_ARRAY_BUFFER, GL_R8, 0, 4, GL_RED, GL_INT, &one);
+    EXPECT_EQ(MobileGL::MG_Impl::GLImpl::GetError(), GL_NO_ERROR);
+
+    Vector<Uint8> actual(initial.size());
+    auto bufferObject = MobileGL::MG_State::pGLContext->GetBufferObject(buffer);
+    ASSERT_NE(bufferObject, nullptr);
+    Memcpy(actual.data(), bufferObject->AcquireMemory(false, true, false), actual.size());
+    EXPECT_EQ(actual, (Vector<Uint8>{0xFF, 0xFF, 0xFF, 0xFF, 0x7F, 0x7F, 0x7F, 0x7F}));
+
+    MobileGL::MG_Impl::GLImpl::BindBuffer(GL_ARRAY_BUFFER, 0);
+    MobileGL::MG_Impl::GLImpl::DeleteBuffers(1, &buffer);
+    DrainPendingGlErrors();
+}
+
 TEST_F(BufferTest, ClearBufferSubDataInitializesIrisStaticSsboRange) {
     GLuint buffer = 0;
     MobileGL::MG_Impl::GLImpl::GenBuffers(1, &buffer);

MobileGL/MG_Test/Framebuffer/FramebufferTest.cpp

@@ -1252,3 +1252,194 @@ TEST_F(FramebufferTest, ApplicationAlphaMaskOffIsStillHonouredOnANativeDrawBuffe
     EXPECT_EQ(g_driverIndexedColorMasks[2].a, GL_TRUE) << "a native buffer keeps its alpha writes";
     EXPECT_EQ(MG_Impl::GLImpl::GetError(), GL_NO_ERROR);
 }
+
+// --- glFramebufferTexture error conditions (GL 4.6 core 9.2.8) ---------------------------------
+//
+// Four of them were missing from the bound-target path while its DSA sibling
+// (glNamedFramebufferTexture) implemented all four, which is what KHR-GL4x.geometry_shader.
+// layered_fbo.fb_texture_* fails on. Two of them - the attachment-range check and the
+// default-framebuffer rejection - newly REFUSE calls that used to succeed, so they are pinned
+// here rather than left to the conformance suite.
+
+TEST_F(FramebufferTest, FramebufferTextureRejectsTheDefaultFramebuffer) {
+    GLuint texture = 0;
+    MG_Impl::GLImpl::CreateTextures(GL_TEXTURE_2D, 1, &texture);
+    MG_Impl::GLImpl::TextureStorage2D(texture, 1, GL_RGBA8, 64, 32);
+    ASSERT_EQ(MG_Impl::GLImpl::GetError(), GL_NO_ERROR);
+
+    // MobileGL models framebuffer 0 as a real FramebufferObject, so the null test that used to
+    // stand in for this could never fire and the attach silently "succeeded".
+    MG_Impl::GLImpl::BindFramebuffer(GL_DRAW_FRAMEBUFFER, 0);
+    MG_Impl::GLImpl::FramebufferTexture(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, texture, 0);
+    EXPECT_EQ(MG_Impl::GLImpl::GetError(), GL_INVALID_OPERATION);
+    DrainPendingGlErrors();
+}
+
+TEST_F(FramebufferTest, FramebufferTextureRejectsAColourAttachmentPastTheLimit) {
+    GLuint framebuffer = 0;
+    GLuint texture = 0;
+    MG_Impl::GLImpl::CreateFramebuffers(1, &framebuffer);
+    MG_Impl::GLImpl::CreateTextures(GL_TEXTURE_2D, 1, &texture);
+    MG_Impl::GLImpl::TextureStorage2D(texture, 1, GL_RGBA8, 64, 32);
+    MG_Impl::GLImpl::BindFramebuffer(GL_DRAW_FRAMEBUFFER, framebuffer);
+    ASSERT_EQ(MG_Impl::GLImpl::GetError(), GL_NO_ERROR);
+
+    // The same limit ValidateColorAttachmentInRange reads, so the test cannot disagree with the
+    // implementation about where the boundary is.
+    const GLint limit = MG_Backend::pActiveBackendObject
+                            ? static_cast<GLint>(
+                                  MG_Backend::pActiveBackendObject->GetDynamicParameters().MaxColorAttachments)
+                            : static_cast<GLint>(MG_State::GLState::FramebufferObject::MAX_DRAW_BUFFERS);
+    ASSERT_GT(limit, 0);
+    ASSERT_LT(limit, 32) << "the test needs a colour attachment enum past the limit to exist";
+
+    MG_Impl::GLImpl::FramebufferTexture(GL_DRAW_FRAMEBUFFER,
+                                        static_cast<GLenum>(GL_COLOR_ATTACHMENT0 + limit), texture, 0);
+    EXPECT_EQ(MG_Impl::GLImpl::GetError(), GL_INVALID_OPERATION);
+    DrainPendingGlErrors();
+
+    // The last legal one still attaches, so the boundary is off-by-none.
+    MG_Impl::GLImpl::FramebufferTexture(GL_DRAW_FRAMEBUFFER,
+                                        static_cast<GLenum>(GL_COLOR_ATTACHMENT0 + limit - 1), texture, 0);
+    EXPECT_EQ(MG_Impl::GLImpl::GetError(), GL_NO_ERROR);
+}
+
+TEST_F(FramebufferTest, FramebufferTextureReportsInvalidValueForANameThatWasNeverGenerated) {
+    GLuint framebuffer = 0;
+    MG_Impl::GLImpl::CreateFramebuffers(1, &framebuffer);
+    MG_Impl::GLImpl::BindFramebuffer(GL_DRAW_FRAMEBUFFER, framebuffer);
+    ASSERT_EQ(MG_Impl::GLImpl::GetError(), GL_NO_ERROR);
+
+    // INVALID_VALUE, not INVALID_OPERATION: the entry point used to resolve the texture object
+    // first and report the miss with the wrong code, pre-empting ValidateTextureName.
+    const GLuint neverGenerated = std::numeric_limits<GLuint>::max();
+    ASSERT_FALSE(MG_State::pGLContext->ValidateTextureName(neverGenerated));
+    MG_Impl::GLImpl::FramebufferTexture(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, neverGenerated, 0);
+    EXPECT_EQ(MG_Impl::GLImpl::GetError(), GL_INVALID_VALUE);
+    DrainPendingGlErrors();
+}
+
+TEST_F(FramebufferTest, FramebufferTextureRejectsALevelTheTextureDoesNotHave) {
+    GLuint framebuffer = 0;
+    GLuint texture = 0;
+    MG_Impl::GLImpl::CreateFramebuffers(1, &framebuffer);
+    MG_Impl::GLImpl::CreateTextures(GL_TEXTURE_2D, 1, &texture);
+    // Two levels of immutable storage: level 1 is legal, level 2 is not.
+    MG_Impl::GLImpl::TextureStorage2D(texture, 2, GL_RGBA8, 64, 32);
+    MG_Impl::GLImpl::BindFramebuffer(GL_DRAW_FRAMEBUFFER, framebuffer);
+    ASSERT_EQ(MG_Impl::GLImpl::GetError(), GL_NO_ERROR);
+
+    MG_Impl::GLImpl::FramebufferTexture(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, texture, 1);
+    EXPECT_EQ(MG_Impl::GLImpl::GetError(), GL_NO_ERROR) << "the last level the texture has is legal";
+
+    MG_Impl::GLImpl::FramebufferTexture(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, texture, 2);
+    EXPECT_EQ(MG_Impl::GLImpl::GetError(), GL_INVALID_VALUE);
+    DrainPendingGlErrors();
+
+    MG_Impl::GLImpl::FramebufferTexture(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, texture, -1);
+    EXPECT_EQ(MG_Impl::GLImpl::GetError(), GL_INVALID_VALUE);
+    DrainPendingGlErrors();
+}
+
+// The four conditions above are stated once in GL 4.6 core 9.2.8 for the WHOLE family, and
+// glFramebufferTexture2D / 3D / TextureLayer reach the attachment through their own code rather
+// than through the shared helper - so each of them has to be asked separately or one entry point
+// answers differently from its aliases. glFramebufferTexture2D is the most-used of the five, and
+// the default-framebuffer case is the damaging one: the attach used to succeed and replace
+// framebuffer 0's colour attachment, which nothing ever puts back.
+
+TEST_F(FramebufferTest, FramebufferTexture2DRejectsTheDefaultFramebufferAndBadAttachments) {
+    GLuint texture = 0;
+    MG_Impl::GLImpl::CreateTextures(GL_TEXTURE_2D, 1, &texture);
+    MG_Impl::GLImpl::TextureStorage2D(texture, 2, GL_RGBA8, 64, 32);
+    ASSERT_EQ(MG_Impl::GLImpl::GetError(), GL_NO_ERROR);
+
+    const auto defaultFramebuffer = MG_State::pGLContext->GetFramebufferObject(0);
+    ASSERT_NE(defaultFramebuffer, nullptr);
+    const auto& colorBefore = defaultFramebuffer->GetAttachment(FramebufferAttachmentType::Color0);
+    const Bool hadTextureBefore = colorBefore.IsTexture();
+
+    MG_Impl::GLImpl::BindFramebuffer(GL_DRAW_FRAMEBUFFER, 0);
+    MG_Impl::GLImpl::FramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, texture, 0);
+    EXPECT_EQ(MG_Impl::GLImpl::GetError(), GL_INVALID_OPERATION);
+    DrainPendingGlErrors();
+    // ...and, more to the point, the default framebuffer still describes the surface.
+    const auto& colorAfter = defaultFramebuffer->GetAttachment(FramebufferAttachmentType::Color0);
+    EXPECT_EQ(colorAfter.IsTexture(), hadTextureBefore);
+    if (colorAfter.IsTexture() && hadTextureBefore) {
+        EXPECT_NE(colorAfter.GetTexture()->GetExternalIndex(), texture)
+            << "the refused attach must not have replaced framebuffer 0's colour attachment";
+    }
+
+    GLuint framebuffer = 0;
+    MG_Impl::GLImpl::CreateFramebuffers(1, &framebuffer);
+    MG_Impl::GLImpl::BindFramebuffer(GL_DRAW_FRAMEBUFFER, framebuffer);
+    ASSERT_EQ(MG_Impl::GLImpl::GetError(), GL_NO_ERROR);
+
+    const GLint limit = MG_Backend::pActiveBackendObject
+                            ? static_cast<GLint>(
+                                  MG_Backend::pActiveBackendObject->GetDynamicParameters().MaxColorAttachments)
+                            : static_cast<GLint>(MG_State::GLState::FramebufferObject::MAX_DRAW_BUFFERS);
+    MG_Impl::GLImpl::FramebufferTexture2D(GL_DRAW_FRAMEBUFFER,
+                                          static_cast<GLenum>(GL_COLOR_ATTACHMENT0 + limit), GL_TEXTURE_2D,
+                                          texture, 0);
+    EXPECT_EQ(MG_Impl::GLImpl::GetError(), GL_INVALID_OPERATION);
+    DrainPendingGlErrors();
+
+    MG_Impl::GLImpl::FramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, texture, 2);
+    EXPECT_EQ(MG_Impl::GLImpl::GetError(), GL_INVALID_VALUE) << "the texture has two levels, not three";
+    DrainPendingGlErrors();
+
+    MG_Impl::GLImpl::FramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, texture, -1);
+    EXPECT_EQ(MG_Impl::GLImpl::GetError(), GL_INVALID_VALUE);
+    DrainPendingGlErrors();
+
+    // The legal call still works, so the boundary is off-by-none.
+    MG_Impl::GLImpl::FramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, texture, 1);
+    EXPECT_EQ(MG_Impl::GLImpl::GetError(), GL_NO_ERROR);
+}
+
+TEST_F(FramebufferTest, FramebufferTextureLayerRejectsTheDefaultFramebufferAndBadLevels) {
+    GLuint texture = 0;
+    MG_Impl::GLImpl::CreateTextures(GL_TEXTURE_2D_ARRAY, 1, &texture);
+    MG_Impl::GLImpl::TextureStorage3D(texture, 2, GL_RGBA8, 16, 16, 4);
+    ASSERT_EQ(MG_Impl::GLImpl::GetError(), GL_NO_ERROR);
+
+    // The attach path used to bypass every one of these while the DETACH path (texture == 0) went
+    // through the fixed helper, so one entry point answered two different ways.
+    MG_Impl::GLImpl::BindFramebuffer(GL_DRAW_FRAMEBUFFER, 0);
+    MG_Impl::GLImpl::FramebufferTextureLayer(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, texture, 0, 0);
+    EXPECT_EQ(MG_Impl::GLImpl::GetError(), GL_INVALID_OPERATION);
+    DrainPendingGlErrors();
+
+    GLuint framebuffer = 0;
+    MG_Impl::GLImpl::CreateFramebuffers(1, &framebuffer);
+    MG_Impl::GLImpl::BindFramebuffer(GL_DRAW_FRAMEBUFFER, framebuffer);
+    ASSERT_EQ(MG_Impl::GLImpl::GetError(), GL_NO_ERROR);
+
+    MG_Impl::GLImpl::FramebufferTextureLayer(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, texture, 2, 0);
+    EXPECT_EQ(MG_Impl::GLImpl::GetError(), GL_INVALID_VALUE);
+    DrainPendingGlErrors();
+
+    MG_Impl::GLImpl::FramebufferTextureLayer(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, texture, 1, 0);
+    EXPECT_EQ(MG_Impl::GLImpl::GetError(), GL_NO_ERROR);
+}
+
+// The DSA sibling is the entry point the bound-target family was aligned WITH, so an out-of-range
+// immutable level has to be rejected there too - otherwise the alignment created a fresh
+// asymmetry in the opposite direction.
+TEST_F(FramebufferTest, NamedFramebufferTextureRejectsALevelTheTextureDoesNotHave) {
+    GLuint framebuffer = 0;
+    GLuint texture = 0;
+    MG_Impl::GLImpl::CreateFramebuffers(1, &framebuffer);
+    MG_Impl::GLImpl::CreateTextures(GL_TEXTURE_2D, 1, &texture);
+    MG_Impl::GLImpl::TextureStorage2D(texture, 2, GL_RGBA8, 64, 32);
+    ASSERT_EQ(MG_Impl::GLImpl::GetError(), GL_NO_ERROR);
+
+    MG_Impl::GLImpl::NamedFramebufferTexture(framebuffer, GL_COLOR_ATTACHMENT0, texture, 1);
+    EXPECT_EQ(MG_Impl::GLImpl::GetError(), GL_NO_ERROR);
+
+    MG_Impl::GLImpl::NamedFramebufferTexture(framebuffer, GL_COLOR_ATTACHMENT0, texture, 2);
+    EXPECT_EQ(MG_Impl::GLImpl::GetError(), GL_INVALID_VALUE);
+    DrainPendingGlErrors();
+}

MobileGL/MG_Test/Pipeline/PassthroughTessControlTest.cpp

@@ -101,9 +101,14 @@ namespace {
         return builtIns;
     }
 
-    Vector<Uint32> CompileGeneratedSource(Uint32 patchVertices) {
+    const FloatVec4 kDefaultOuter(1.0f, 1.0f, 1.0f, 1.0f);
+    const FloatVec2 kDefaultInner(1.0f, 1.0f);
+
+    Vector<Uint32> CompileGeneratedSource(Uint32 patchVertices,
+                                          Uint32 perVertexMembers = ProgramFactory::kDefaultPerVertexMembers) {
         using namespace MG_Util::ShaderTranspiler;
-        const String source = ProgramFactory::BuildPassthroughTessControlSource(patchVertices);
+        const String source = ProgramFactory::BuildPassthroughTessControlSource(patchVertices, kDefaultOuter,
+                                                                                kDefaultInner, perVertexMembers);
 
         ShaderAttrib shaderAttrib{.shaderType = GL_TESS_CONTROL_SHADER, .sourceStr = source};
         auto shaderResult = ShaderCompiler::CompileShader(shaderAttrib);
@@ -162,11 +167,85 @@ TEST_F(PassthroughTessControlTest, ForwardsPositionAndWritesBothLevelArrays) {
 // user-defined varying, ReflectPassthroughTessControlNeed's "built-ins only" refusal stops being
 // the right gate and both have to move together.
 TEST_F(PassthroughTessControlTest, InterfaceIsBuiltInsOnly) {
-    const String source = ProgramFactory::BuildPassthroughTessControlSource(4);
+    const String source = ProgramFactory::BuildPassthroughTessControlSource(
+        4, kDefaultOuter, kDefaultInner, ProgramFactory::kDefaultPerVertexMembers);
     EXPECT_EQ(source.find("layout(location"), String::npos) << source;
     EXPECT_NE(source.find("layout(vertices = 4) out;"), String::npos) << source;
 }
 
+// glPatchParameterfv's levels are compiled into this stage - Vulkan has no dynamic state for them -
+// so two different level sets must produce two different sources AND two different cache keys.
+// Without the second half a pipeline memoised at one set of levels would be handed back after the
+// application changed them, and the tessellation would silently stay at the old levels.
+TEST_F(PassthroughTessControlTest, BakesTheDefaultTessLevelsInAndKeysOnThem) {
+    constexpr Uint32 kMembers = ProgramFactory::kDefaultPerVertexMembers;
+    const FloatVec4 outer(2.0f, 3.0f, 4.0f, 5.0f);
+    const FloatVec2 inner(6.5f, 7.25f);
+    const String source = ProgramFactory::BuildPassthroughTessControlSource(4, outer, inner,
+                                                                             ProgramFactory::kDefaultPerVertexMembers);
+    EXPECT_NE(source.find("gl_TessLevelOuter[0] = 2.0;"), String::npos) << source;
+    EXPECT_NE(source.find("gl_TessLevelOuter[3] = 5.0;"), String::npos) << source;
+    EXPECT_NE(source.find("gl_TessLevelInner[0] = 6.5;"), String::npos) << source;
+    EXPECT_NE(source.find("gl_TessLevelInner[1] = 7.25;"), String::npos) << source;
+
+    const Uint64 defaultKey =
+        ProgramFactory::ComputePassthroughTessControlKey(4, kDefaultOuter, kDefaultInner, kMembers);
+    EXPECT_NE(ProgramFactory::ComputePassthroughTessControlKey(4, outer, inner, kMembers), defaultKey);
+    EXPECT_NE(ProgramFactory::ComputePassthroughTessControlKey(4, kDefaultOuter, inner, kMembers), defaultKey);
+    EXPECT_NE(ProgramFactory::ComputePassthroughTessControlKey(3, kDefaultOuter, kDefaultInner, kMembers), defaultKey);
+    EXPECT_EQ(ProgramFactory::ComputePassthroughTessControlKey(4, kDefaultOuter, kDefaultInner, kMembers), defaultKey);
+    // ...and the gl_PerVertex member set is in the same key, for the same reason: two programs at
+    // different GLSL versions need differently-shaped modules, and a pipeline memoised against one
+    // shape must not be handed back for the other.
+    constexpr Uint32 kMembersWithCull =
+        ProgramFactory::kDefaultPerVertexMembers |
+        static_cast<Uint32>(ProgramFactory::PerVertexMemberBit::CullDistance);
+    EXPECT_NE(ProgramFactory::ComputePassthroughTessControlKey(4, kDefaultOuter, kDefaultInner, kMembersWithCull),
+              defaultKey);
+}
+
+// The generated stage still has to COMPILE with non-default levels: an integral level spelled
+// without a decimal point is an int literal, and `gl_TessLevelOuter[0] = 2;` does not compile.
+TEST_F(PassthroughTessControlTest, CompilesWithNonDefaultLevels) {
+    using namespace MG_Util::ShaderTranspiler;
+    const String source =
+        ProgramFactory::BuildPassthroughTessControlSource(4, FloatVec4(2.0f, 2.0f, 2.0f, 2.0f),
+                                                          FloatVec2(2.0f, 2.0f),
+                                                          ProgramFactory::kDefaultPerVertexMembers);
+    ShaderAttrib shaderAttrib{.shaderType = GL_TESS_CONTROL_SHADER, .sourceStr = source};
+    auto shaderResult = ShaderCompiler::CompileShader(shaderAttrib);
+    EXPECT_TRUE(shaderResult) << (shaderResult ? String{} : shaderResult.error().log) << source;
+}
+
+// The generator honours the mask it is given, in both directions. This is what covers the
+// pre-cutoff three-member form now that no authorable tessellation evaluation stage produces it
+// (ARB_tessellation_shader is GL 4.0 and gl_CullDistance joins the block at 400), and it is also
+// the fallback ReflectPassthroughTessControlNeed uses when it cannot read a module's block.
+TEST_F(PassthroughTessControlTest, RedeclaresExactlyTheRequestedMembers) {
+    using Bit = ProgramFactory::PerVertexMemberBit;
+    constexpr Uint32 kWithCull = ProgramFactory::kDefaultPerVertexMembers | static_cast<Uint32>(Bit::CullDistance);
+    constexpr Uint32 kBuiltInCullDistance = 4;
+    constexpr Uint32 kBuiltInClipDistance = 3;
+
+    const Vector<Uint32> withoutCull = CompileGeneratedSource(4, ProgramFactory::kDefaultPerVertexMembers);
+    ASSERT_FALSE(withoutCull.empty());
+    const std::set<Uint32> withoutCullBuiltIns = DeclaredBuiltIns(withoutCull);
+    EXPECT_TRUE(withoutCullBuiltIns.contains(kBuiltInClipDistance));
+    EXPECT_FALSE(withoutCullBuiltIns.contains(kBuiltInCullDistance))
+        << "the three-member mask must not emit gl_CullDistance";
+    for (const auto& [structId, shape] : BuiltInBlockShapes(withoutCull)) {
+        EXPECT_EQ(StructMemberCount(withoutCull, structId), 3u) << "structId=" << structId;
+    }
+
+    const Vector<Uint32> withCull = CompileGeneratedSource(4, kWithCull);
+    ASSERT_FALSE(withCull.empty());
+    EXPECT_TRUE(DeclaredBuiltIns(withCull).contains(kBuiltInCullDistance))
+        << "the four-member mask must emit gl_CullDistance";
+    for (const auto& [structId, shape] : BuiltInBlockShapes(withCull)) {
+        EXPECT_EQ(StructMemberCount(withCull, structId), 4u) << "structId=" << structId;
+    }
+}
+
 // THE load-bearing test. Vulkan matches built-in interface blocks by their whole shape, and this
 // stage is compiled ON ITS OWN - it never goes through the glslang link that gives a real program
 // its gl_PerVertex. So the shape it declares has to equal the shape a linked vertex+evaluation
@@ -179,13 +258,33 @@ TEST_F(PassthroughTessControlTest, InterfaceIsBuiltInsOnly) {
 TEST_F(PassthroughTessControlTest, MatchesTheFrontendPerVertexBlock) {
     using namespace MG_Util::ShaderTranspiler;
 
-    // Deliberately the shape of KHR-GL43.shader_storage_buffer_object.advanced-write-tessellation:
-    // a vertex stage feeding an evaluation stage with no control stage in between.
-    static const char* kVs = R"(#version 430 core
+    // MORE THAN ONE VERSION, because the shape is a function of the neighbour's GLSL version and
+    // a single-version case cannot see that. glslang gates gl_PerVertex's gl_CullDistance member
+    // on a version cutoff, and this case used to link #version 430 ONLY - which is exactly why a
+    // generator hardcoded to the pre-cutoff three-member form looked correct while every program
+    // above it, including every ESSL program (the source processor rewrites those to
+    // "#version 460 core"), was silently mismatched.
+    //
+    // The expected member COUNT is deliberately not spelled per version any more. It moved once
+    // already (the fork's GL_ARB_cull_distance work lowered the cutoff from 450 to 400, so 430
+    // went from three members to four), and pinning it here only produced a test that failed for
+    // being right. What must hold - and is what this case now asserts - is that the generator
+    // reproduces whatever glslang produced, at every version, plus the floor that a per-vertex
+    // block always has at least gl_Position. A tessellation evaluation stage cannot be authored
+    // below #version 400 at all (ARB_tessellation_shader is GL 4.0), so 400 is the bottom of the
+    // reachable range; the pre-cutoff three-member form is covered through the explicit-mask case
+    // below instead.
+    for (const char* version : {"#version 400 core", "#version 430 core", "#version 460 core"}) {
+        SCOPED_TRACE(version);
+
+        // Deliberately the shape of
+        // KHR-GL43.shader_storage_buffer_object.advanced-write-tessellation: a vertex stage
+        // feeding an evaluation stage with no control stage in between.
+        const String vs = String(version) + R"(
 layout(location = 0) in vec4 g_in_position;
 void main() { gl_Position = g_in_position; }
 )";
-    static const char* kTes = R"(#version 430 core
+        const String tes = String(version) + R"(
 layout(quads) in;
 void main() {
   vec4 p0 = mix(gl_in[0].gl_Position, gl_in[1].gl_Position, gl_TessCoord.x);
@@ -193,55 +292,62 @@ void main() {
   gl_Position = mix(p0, p1, gl_TessCoord.y);
 }
 )";
-    static const char* kFs = R"(#version 430 core
+        const String fs = String(version) + R"(
 layout(location = 0) out vec4 g_fs_out;
 void main() { g_fs_out = vec4(0, 1, 0, 1); }
 )";
 
-    const Vector<GLenum> types{GL_VERTEX_SHADER, GL_TESS_EVALUATION_SHADER, GL_FRAGMENT_SHADER};
-    const Vector<const char*> sources{kVs, kTes, kFs};
-    Vector<SharedPtr<glslang::TShader>> shaders;
-    for (SizeT i = 0; i < types.size(); ++i) {
-        ShaderAttrib attrib{.shaderType = types[i], .sourceStr = sources[i]};
-        auto compiled = ShaderCompiler::CompileShader(attrib);
-        ASSERT_TRUE(compiled) << compiled.error().log;
-        shaders.push_back(compiled.value());
-    }
-    ProgramAttrib programAttrib{.shaders = shaders};
-    auto linked = ShaderCompiler::LinkProgram(programAttrib);
-    ASSERT_TRUE(linked) << linked.error().log;
-    ProgramBinaryAttrib binaryAttrib{.shaderTypes = types, .program = *linked.value()};
-    auto binary = ShaderCompiler::GetSpirvBinaryFromProgram(binaryAttrib);
-    ASSERT_TRUE(binary);
-    ASSERT_EQ(binary->size(), types.size());
+        const Vector<GLenum> types{GL_VERTEX_SHADER, GL_TESS_EVALUATION_SHADER, GL_FRAGMENT_SHADER};
+        const Vector<const String*> sources{&vs, &tes, &fs};
+        Vector<SharedPtr<glslang::TShader>> shaders;
+        for (SizeT i = 0; i < types.size(); ++i) {
+            ShaderAttrib attrib{.shaderType = types[i], .sourceStr = *sources[i]};
+            auto compiled = ShaderCompiler::CompileShader(attrib);
+            ASSERT_TRUE(compiled) << (compiled ? String{} : compiled.error().log);
+            shaders.push_back(compiled.value());
+        }
+        ProgramAttrib programAttrib{.shaders = shaders};
+        auto linked = ShaderCompiler::LinkProgram(programAttrib);
+        ASSERT_TRUE(linked) << (linked ? String{} : linked.error().log);
+        ProgramBinaryAttrib binaryAttrib{.shaderTypes = types, .program = *linked.value()};
+        auto binary = ShaderCompiler::GetSpirvBinaryFromProgram(binaryAttrib);
+        ASSERT_TRUE(binary);
+        ASSERT_EQ(binary->size(), types.size());
 
-    // The evaluation stage's gl_in is the block the pass-through has to feed. It is the only
-    // built-in block that stage declares as an input, so the module holds exactly one such shape
-    // besides its own gl_PerVertex output - and both are the same shape, which is the point.
-    const auto tesShapes = BuiltInBlockShapes((*binary)[1]);
-    ASSERT_FALSE(tesShapes.empty());
-    const Vector<Uint32> frontendShape = tesShapes.begin()->second;
-    const Uint32 frontendMembers = StructMemberCount((*binary)[1], tesShapes.begin()->first);
-    for (const auto& [structId, shape] : tesShapes) {
-        EXPECT_EQ(shape, frontendShape) << "the evaluation stage's own built-in blocks disagree";
-        EXPECT_EQ(StructMemberCount((*binary)[1], structId), frontendMembers);
-    }
+        // The evaluation stage's gl_in is the block the pass-through has to feed. It is the only
+        // built-in block that stage declares as an input, so the module holds exactly one such
+        // shape besides its own gl_PerVertex output - and both are the same shape, which is the
+        // point.
+        const auto tesShapes = BuiltInBlockShapes((*binary)[1]);
+        ASSERT_FALSE(tesShapes.empty());
+        const Vector<Uint32> frontendShape = tesShapes.begin()->second;
+        const Uint32 frontendMembers = StructMemberCount((*binary)[1], tesShapes.begin()->first);
+        EXPECT_GE(frontendMembers, 1u) << "a gl_PerVertex block always carries at least gl_Position";
+        for (const auto& [structId, shape] : tesShapes) {
+            EXPECT_EQ(shape, frontendShape) << "the evaluation stage's own built-in blocks disagree";
+            EXPECT_EQ(StructMemberCount((*binary)[1], structId), frontendMembers);
+        }
 
-    const Vector<Uint32> passthrough = CompileGeneratedSource(4);
-    ASSERT_FALSE(passthrough.empty());
-    const auto passthroughShapes = BuiltInBlockShapes(passthrough);
-    ASSERT_FALSE(passthroughShapes.empty());
+        // ...and the generator is driven the way PRODUCTION drives it: the mask comes from the
+        // evaluation stage's own module, not from a constant the test picked.
+        const Uint32 reflectedMembers = ProgramFactory::ReflectPerVertexInputMembers((*binary)[1]);
+        EXPECT_NE(reflectedMembers, 0u) << "the input per-vertex block walk found nothing to match against";
+        const Vector<Uint32> passthrough = CompileGeneratedSource(4, reflectedMembers);
+        ASSERT_FALSE(passthrough.empty());
+        const auto passthroughShapes = BuiltInBlockShapes(passthrough);
+        ASSERT_FALSE(passthroughShapes.empty());
 
-    Uint32 perVertexBlocksChecked = 0;
-    for (const auto& [structId, shape] : passthroughShapes) {
-        // gl_TessLevelOuter/Inner are decorated on plain variables, not on a block, so every
-        // struct that reaches here is a gl_PerVertex - gl_in's and gl_out's.
-        EXPECT_EQ(shape, frontendShape)
-            << "the pass-through control stage's gl_PerVertex no longer matches the one the "
-               "frontend gives a linked vertex+evaluation program";
-        EXPECT_EQ(StructMemberCount(passthrough, structId), frontendMembers)
-            << "the pass-through control stage's gl_PerVertex has a different member count";
-        ++perVertexBlocksChecked;
+        Uint32 perVertexBlocksChecked = 0;
+        for (const auto& [structId, shape] : passthroughShapes) {
+            // gl_TessLevelOuter/Inner are decorated on plain variables, not on a block, so every
+            // struct that reaches here is a gl_PerVertex - gl_in's and gl_out's.
+            EXPECT_EQ(shape, frontendShape)
+                << "the pass-through control stage's gl_PerVertex no longer matches the one the "
+                   "frontend gives a linked vertex+evaluation program";
+            EXPECT_EQ(StructMemberCount(passthrough, structId), frontendMembers)
+                << "the pass-through control stage's gl_PerVertex has a different member count";
+            ++perVertexBlocksChecked;
+        }
+        EXPECT_EQ(perVertexBlocksChecked, 2u) << "expected both gl_in and gl_out to be gl_PerVertex blocks";
     }
-    EXPECT_EQ(perVertexBlocksChecked, 2u) << "expected both gl_in and gl_out to be gl_PerVertex blocks";
 }