Files
MobileGL/docs/Disaggregated/notes/p5b/p5b-results/c0b-v1.md
T
swung0x48 aba329cc53 [Docs] (Disaggregated): P6's plan and contract draft; the working notes move from a local directory into the repository
Two things land together: the plan and contract draft P6 needs before it starts, and the
working notes that until now existed on exactly one machine.

1. P6's plan and contract draft (neither started; not one line of code)

     docs/Disaggregated/P6-SPAWN-PLAN.md      the package plan
     docs/Disaggregated/P6-CONTRACT-DRAFT.md  the contract draft (moves to
                                              MG_Remote/CONTRACT-P6.md when c6 lands)

   The three findings that mattered, all written into the documents:

   a. P6 is SMALLER than its roadmap cell suggests. The transport primitives are already in
      the tree and already tested: SocketDoorbell including hangup detection and the two
      lost-wakeup fences (Doorbell.cpp:211-360, four cases in FdPassingTest), SCM_RIGHTS fd
      passing, ShmSegment::Adopt, and the SurfaceOp / WindowKind schema already defined in
      protocol.fbs. inproc uses a real second shm mapping rather than an alias, and
      SessionRings.h:22-28 says that was done so today is not the first time it runs. P6 is
      assembly plus a control plane plus a process, not "write a socket transport".

   b. The evidence for "P6 is only a transport swap" has EXPIRED. It was bought by P5c, whose
      audit found 59 direct memory accesses in a tree that already looked clean. P5e changed
      the tree again. So the first package, a6, is a read-only audit that writes no code, and
      every contract row that depends on it is marked PENDING a6 rather than decided in
      advance.

   c. P6 does not get a game running on the phone. ANativeWindow* is meaningless in the child,
      and protocol.fbs:186-196's own comment says Android transfers the window out of band -
      that band is P12. P6 lands pbuffer / surfaceless / offscreen, which is what HeadlessGL,
      the split-equivalent lane and trace replay need, and which is exactly what P6's own exit
      gate measures. A real window reaching the server is a named refusal. This gets its own
      section because "P6 is done" would otherwise read as "the game runs now".

   The contract adds rule G: a role may not name a process-local handle. It extends P5c's
   rule E (a role may not name the other role's memory) from memory to identity - EGLDisplay /
   EGLSurface / EGLContext / ANativeWindow* / fd number / pid / mapped address either cross as
   a token this contract defines or they do not cross, and a descriptor crosses only through
   SCM_RIGHTS, where the kernel renumbers it.

   What P5e left for P6 gets its own section. The latch semantics are already designed
   (ARCHITECTURE.md:504) and are copied unchanged. What is new is that after run-ahead the
   client no longer waits while the peer process can DIE: the client holds records that were
   published and will never be applied, and Android's freezer suspends the child without
   closing anything. So slow and dead must be distinguishable WITHOUT a timeout threshold.
   That is P6's only new design; everything else is carriage. It shares a predicate with E1's
   outstanding re-specification (ID-122), and the two should be designed together.

   Two gates carried over from P5e's lessons: G2 requires integration-spawn's test-name set to
   match integration-split's name for name, because a differing set means the lane is quietly
   running less; and the arm-proof gate requires every entry to record the child pid and
   transport=spawn in its OWN private log - ID-124 established that this kind of false green
   really happens, and a spawn lane that actually ran monolith would pass every other gate.
   S2's shape must be settled before it is written, and the document says why: E1 is the
   cautionary case.

2. The working notes move into the repository (docs/Disaggregated/notes/, 415 .md files)

   ROADMAP / CURRENT_STAGE_PROGRESS / MEASUREMENTS cite ruling numbers, audit line numbers and
   the provenance of gate figures from ~/w7/notes/, a directory that exists on one machine and
   is under no version control. A record cited hundreds of times that lives only locally is not
   a record, so it moves in whole, with its directory structure intact.

   Only .md moves. Scripts, run logs, census and gate result JSON, patches, APKs and nm dumps
   stay where they are: they are evidence DATA rather than documents - large, regenerated per
   run, and mostly carrying absolute paths and device serials that mean something only on the
   machine that produced them. The mapping and this split are written in
   docs/Disaggregated/notes/README.md.

   Two exceptions are recorded there too: P6's BRIEF has been promoted to a first-class
   document and keeps no copy under notes/, so the two cannot drift apart; and recovered/ holds
   material salvaged from failed workflow runs, which may duplicate the like-named phase
   directories and is kept because a few of those were the only complete version at the time.

   CI's citation lint takes docs/Disaggregated/*.md only (non-recursive), and notes/ is
   deliberately outside it: those notes' file:line citations were written against their own
   code heads, and resolving them against today's HEAD would fail wholesale - which is not an
   error, it is their correct state as a historical record. To check one, use the revision it
   declares (--rev).

3. Inconsistencies fixed in passing

   - README.md's status line still said P5e was in progress while the stage table already said
     closed.
   - CURRENT_STAGE_PROGRESS's next-step item 1 still listed the exit gate as pending, though
     (a) and (b) were run in the previous commit; rewritten to closed plus the one item that
     actually remains (E1).
   - ROADMAP's P6 row still had the pre-phase plan in three of its cells; it now carries the
     package order, the scope boundary and the two new gates.

Checked with check_doc_citations.py: 125 citations across docs/Disaggregated/*.md, 0 problems.
2026-09-18 23:31:39 -04:00

21 KiB
Raw Blame History

c0b — the P5b contract package, v1

Branch p5b/c0b, forked from feat/disaggregated @ c9878d69, merged feat/disaggregated @ f5614995 (v1 round 3) before finishing. Worktree ~/w7/p5b-c0b. Nothing pushed; nothing merged into the integration branch.

4dd4027e [Feat] (MG_Pipe, P5b): append the five P5b verb rows at opcodes 72..76, arm draw_vbo
         with the kDrawIsIndirect second tail, give resource_copy_region its two GL names and
         the clear discriminants one spelling, and stamp every P5b verb boundary
a1641130 [Feat] (MG_Remote, P5b): decode every P5b row to a WireVerbSink method, stub each
         ServerVerbSink body with a named UnmigratedVerb Fatal, and partition the emit table
         class-C list per migration package
02972f70 [Test] (MG_Test, P5b): pin the 76-row catalogue, the 18 stamp rows and the 77 verify
         payloads, and round-trip each P5b row once through the codec
538c3c19 [Docs] (MG_Remote): CONTRACT-P5B.md - the wire records, sinks, refusals and rulings
         for the 25 measured class-C slots
f5676103 [Merge] (P5b, c0b): merge feat/disaggregated at f5614995 (v1 round 3, ID-62/ID-67)
         into p5b/c0b

Range 4dd4027e..f5676103 (four single-line commits plus the merge). The working tree is clean; the worktree's submodule symlinks are left OFF (git-safe).

Notes written beside it: ~/w7/notes/p5b/PACKAGE-PREAMBLE.md, ~/w7/notes/p5b/BRIEF-P5B.md. The contract: MobileGL/MG_Remote/CONTRACT-P5B.md (beside CONTRACT-P5.md, which stays authoritative for everything it covers).


1. What landed

# item state
1 CONTRACT-P5B.md: every one of the 25 slots - wire record, class, sink signature, reply slot (none), thread rule, refusal shape, monolith (untouched) - grouped d1 / i1 / t2 / f1; table 0 additions; table 2 and 3 under P5b; knobs (none); 12 rulings with overturn conditions; the 71-slot partition; ownership with two pre-granted edits done
2 PipeCalls.def: five rows appended at opcodes 72..76 (BindShaderImage, PatchParameter, BindStreamOutput, SetStorageBlockBinding kHasBlob, CopyFramebufferToTexture), documented count 71 → 76; no opcode moved done
3 MGPipeTypes.h: the five PODs (40 / 8 / 16 / 40 / 48 B); kDrawIsIndirect; MGPCopyRegion 64 → 72 (two GL names); MGPMipPlan::Target ruled the GL enum; the clear discriminants kMGPipeClearKind* / kMGPipeClearValueClass* (one spelling; the two hand-minted copies in EmitTables.h and PipeApplier.h now alias them) done
4 PipeFields.def: five field lists + MGPCopyRegion's two names; verify payloads 72 → 77. FieldOwnership.def: six stamp rows (ResourceCopyRegion → CopyImageSubData and the five new ops), 12 → 18. Generated tables regenerated; gen_pipe.py --check / --self-test (9/9), gen_pipe_field_ownership.py --check / --self-test (15/15), gen_pipe_dirty_surface.py --check all green done
5 Codec: MGPW_FOR_EACH_CALL + name/size tables for the five rows; BlobSlotsFor for the name blob; draw_vbo's layout takes kDrawIsIndirect (exclusive with the span, NumDraws == 0, both Fatal{ProtocolCorruption} on both sides); WireRecordLayout::SecondTailIsHostSpans so the encoder's span honesty pass cannot read an indirect block as spans; arms for the six pre-existing rows that returned false (LaunchGrid, MemoryBarrier, the four stream-output rows), for ResourceCopyRegion and GenerateMipmap, and for the five new rows (the name blob is bounded at 4096 bytes, copied and re-terminated, rule C) done
6 WireVerbSink: OnDrawVbo gains const MGPDrawIndirect*; 13 new methods. ServerVerbSink: 13 overrides, every body Fatal{UnmigratedVerb, "<GL slot>"} by the slot's own name ("(server sink)" in the message text); OnDrawVbo declines the indirect block, the span, NumDraws > 1 and the instanced arms by name instead of by log line done
7 EmitTables.cpp: class C partitioned into MGR_UNMIGRATED_{D1,I1,T2,F1,TAIL}_SLOTS with per-package kEmittedSlots* and ownership static_asserts 19 / 7 / 7 / 11 / 20 = 64; EmitTables.h includes MGPipeTypes.h for the aliases. No slot flipped: UnmigratedSlotCount() == 64, ImplementedVerbCount() == 5 as RemoteClientTest pins done
8 Tests: PipeCatalogueTest (kCtxVerb 18, opcodes 72..76 by value, kOpCount 77, the P5b flags/sizes, the clear constants, verify payloads 77); FieldOwnershipTest (18 stamp rows, six by name); PipeWireCodecTest (13 new cases: 11 round trips - one per row, the four stream-output rows in one - and two forged-record Fatals for the draw flag exclusion and the indirect-with-ranges refusal) done
9 Notes: PACKAGE-PREAMBLE.md (numbers on this head, the ID-66 process rules, the setsid trap), BRIEF-P5B.md (goal, four packages with slots / rows / models, per-package gate, file ownership, the loop) done
10 Tools: ~/w7/p5b-c0b-{setup,submodlinks,gate,accept}.sh, ~/w7/p5b-c0b-census.{sh,py} (the P6 lane runner over any tree: p5b-c0b-census.sh <tree> <outdir>) done

2. Acceptance numbers

All on ~/w7/p5b-c0b @ f5676103 (the merged head), GLIBC_TUNABLES=glibc.malloc.tcache_count=0. Log: ~/w7/p5b-c0b-gate-run3.log; the census: ~/w7/p5b-c0b-census-logs/results.json (per entry: status, rc, first Fatal{…}, command, working directory, its own .log/.out).

2.1 Builds — four for four

build-linux (pull), build-push, build-verify, build-split: configure rc=0, build rc=0, all four, CTestTestfile.cmake confirmed in each.

2.2 G1 — did not move

nm --defined-only build-linux/libMobileGL.so | grep -ic MG_Remote  ->  0
nm --defined-only build-split/libMobileGL.so | grep -ic MG_Remote  ->  627
symbol-report vs ~/w7/p5-before-libMobileGL.so (.READY 4595163c), --threshold 0:
  .text 10806611 -> 10806611 (+0, +0.000%)
  27814 -> 27814 defined symbols: 0 added, 0 removed, 0 resized, 0 renamed

Held because everything c0b added is either a header-only type, a .def row whose generated tables the pull build never odr-uses (the baseline .so defines no gMGPipe* symbol), or an MG_Remote file compiled only under MOBILEGL_BUILD_DISAGGREGATED.

2.3 Generators and purity

gen_pipe.py --check                       : generated files are up to date
gen_pipe.py --self-test                   : 9 negative-control trips, positive control OK
gen_pipe_field_ownership.py --check       : generated file is up to date
gen_pipe_field_ownership.py --self-test   : 15 negative-control trips, positive control OK
gen_pipe_dirty_surface.py --check         : green
check_include_closure.py                  : 4 probes, 0 skipped, 0 problem(s)
check_doc_citations.py CONTRACT-P5B.md    : 36 citations, 0 problem(s)

2.4 Test lanes

lane result
ctest -L unit build-linux / push / verify 1817 / 1817 each (the head's baseline)
ctest -L unit build-split 2084 / 2084 = the head's 2071 + c0b's 13 codec cases; no known red (c1g landed before c0b forked)
ctest -L integration-split build-split, MOBILEGL_TRANSPORT=inproc 22 / 22 (2 skipped by design, as in ID-66's quick gate)
ctest -L integration-gpu build-push, MOBILEGL_TRANSPORT=monolith 1128 / 1128
G2 pull-vs-push ctest -N names, every label 0 diff lines
G14 build-linux names vs ~/w7/p5-before-ctest-names.txt 0 removed, 43 added (c0b's 13 + v1 round 3's 30 ServerLoopTest cases)
G5 p3a_untouched_regions.sh ff2994d9 HEAD rc=0, byte-identical
G5 p4a_untouched_regions.sh ff2994d9 HEAD rc=0, byte-identical

2.5 The inproc census on this head

~/w7/p5b-c0b-census.sh ~/w7/p5b-c0b ~/w7/p5b-c0b-census-logs (the P6 runner over this tree; 1,149 entries, eight workers, private logs):

status P6 census (e61d0012) c0b head (f5676103)
passed 426 432
skipped 185 185
subprocess-aborted 511 505
ordinary failed 27 27

The verb census is unchanged: 505 UnmigratedVerb aborts, the same 505, the same per-slot table (BindImageTexture 138, BeginTransformFeedback 95, DrawElements 56, DispatchCompute 49, PatchParameteri 43, CopyImageSubData 30, MemoryBarrier 18, MultiDrawElementsBaseVertex 18, ClearBufferiv 12, ClearBufferuiv 8, ClearBufferfv 6, DrawArraysInstancedBaseInstance 4, …). The six entries that moved are exactly the six the P6 census listed as Fatal{StageSnapshotTooNarrow, "respecify_whole_store"} (two LargeArenaAdoptionScenario pairs, MapPersistentRoundtrips.LargeArenaAdoptionScenario.AnAdoptionCostsExactlyOneMap, ResourceSubsystemControlScenario.ClearingTheP3aBits), now PASSED - v1 round 3's M-3 fix (1646049a, "aborted six LargeArenaAdoption / ResourceSubsystemControl joint entries by name"), which landed on feat/disaggregated while c0b ran and is in the merge. Entry names identical between the two runs; no other entry changed status. 432 / 185 / 505 / 27 is therefore the baseline the four packages diff against, and every abort in it is a verb abort.


3. The 25 rows and their carriers, in one table

R = an existing row; N = appended by c0b. Every payload is an inline POD in SEG_CMD; the two variable parts are named. No row owns a reply slot.

# GL slot (census) pkg row (op) R/N fixed payload var part / carrier sink stamp
1 DrawElements (56 + every MC trace) d1 draw_vbo (59) R MGPDrawInfo 56 B MGPDrawRange[NumDraws] in SEG_CMD; client-side indices: MGHostSpan tail naming a SEG_STAGE run OnDrawVbo DrawArrays
2 DrawElementsInstancedBaseVertex (trace) d1 59 R InstanceCount, IndexBias as 1 OnDrawVbo DrawArrays
3 MultiDrawElementsBaseVertex (18) d1 59 R NumDraws = drawcount MGPDrawRange[drawcount] OnDrawVbo DrawArrays
4 DrawArraysInstancedBaseInstance (4) d1 59 R IndexSize 0, InstanceCount, StartInstance one range OnDrawVbo DrawArrays
5 DrawElementsBaseVertex (4) d1 59 R IndexBias one range OnDrawVbo DrawArrays
6 MultiDrawArrays (2) d1 59 R IndexSize 0, NumDraws ranges OnDrawVbo DrawArrays
7 MultiDrawElements (2) d1 59 R NumDraws ranges OnDrawVbo DrawArrays
8 DrawArraysInstanced (2) d1 59 R InstanceCount one range OnDrawVbo DrawArrays
9 MultiDrawElementsIndirect (2) d1 59 R (tail N) Flags = kDrawIsIndirect, NumDraws 0 MGPDrawIndirect (40 B) as the second tail, in SEG_CMD OnDrawVbo(…, indirect) DrawArrays
10 MultiDrawArraysIndirectCount (2) d1 59 R (tail N) as 9 + ParameterBuffer, ParameterOffset, DrawCount = maxdrawcount MGPDrawIndirect OnDrawVbo DrawArrays
11 BindImageTexture (138) i1 bind_shader_image (72) N MGPImageBind 40 B (GL args verbatim + handle) — OnBindShaderImage BindImageTexture
12 DispatchCompute (49) i1 launch_grid (60) R MGPGridInfo 48 B — OnLaunchGrid DispatchCompute
13 CopyImageSubData (30) i1 resource_copy_region (53) R (72 B) MGPCopyRegion + SrcGlName/DstGlName — OnResourceCopyRegion CopyImageSubData (new stamp row)
14 MemoryBarrier (18) i1 memory_barrier (61) R MGPMemoryBarrier 8 B — OnMemoryBarrier MemoryBarrier
15 ShaderStorageBlockBinding (4) i1 set_storage_block_binding (75) N, kHasBlob MGPStorageBlockBinding 40 B the block name, SEG_STAGE, Size = strlen + 1 OnSetStorageBlockBinding(…, name) ShaderStorageBlockBinding
16 BeginTransformFeedback (95) t2 begin_stream_output (62) R MGPStreamOutputBegin 8 B — OnBeginStreamOutput BeginTransformFeedback
17 PatchParameteri (43) t2 patch_parameter (73) N MGPPatchParameter 8 B — OnPatchParameter PatchParameteri
18 BindTransformFeedback (2) t2 bind_stream_output (74) N MGPStreamOutputBind 16 B — OnBindStreamOutput BindTransformFeedback
19 ClearBufferiv (12) f1 clear (57) R MGPClear 48 B, Kind/ValueClass per the one spelling — OnClear (live) Clear
20 ClearBufferuiv (8) f1 57 R as 19 — OnClear (live) Clear
21 ClearBufferfv (6) f1 57 R as 19 — OnClear (live) Clear
22 ClearBufferfi (1) f1 57 R as 19 — OnClear (live) Clear
23 ClearNamedFramebufferfv (2) f1 57 R Fbo = the named handle (Named record precedes) — OnClear; unbound → refused by name Clear
24 CopyTexImage2D (3) f1 copy_framebuffer_to_texture (76) N MGPCopyFromFramebuffer 48 B — OnCopyFramebufferToTexture CopyTexImage2D
25 GenerateMipmap (2) f1 generate_mipmap (54) R MGPMipPlan 16 B, Target = GL enum — OnGenerateMipmap GenerateMipmap

Companions that share a row and whose slots the same package owns (unmeasured, no gate): DrawElementsInstanced* ×3, DrawRangeElements* ×2, *Indirect ×4 (d1); DispatchComputeIndirect, MemoryBarrierByRegion (i1); End/Pause/ResumeTransformFeedback (t2; DeleteTransformFeedback has no row); ClearNamedFramebufferfi/iv/uiv, CopyTexSubImage2D (f1).


4. Rulings I made, each with what would overturn it

The full list with evidence is CONTRACT-P5B.md §6; the ones a package will trip over:

  1. No P5b slot gains an MGPipeApply*; all 25 reach WireVerbSink. The applier's 37 entry points are the object/state families; a verb is a backend CALL, and the census's correction (the P5 five already go to the sink) is the pattern. Consequence: R-17's generated routing is untouched, PipeCatalogueTest still reads 33 installed + 4 escapes, and a migration is two files per slot. Overturned by a verb that has to be replayable from the monolith side through a table (P13's recorder).

  2. Rule D: a verb crosses AS THE CALL. GL enums and GL names verbatim beside the handle; the sink reproduces the backend call; the backend keeps reading its verb class's BARRIER-PULLED fields. This is why nothing on the monolith path changes and why rsp grows - the phase's measurement, not a regression. Overturned by nothing in P5b; retiring the pulls is P3b/P4b/P7/P8's per CONTRACT-P5 table 2.

  3. Four of the five appended rows exist because the backend does work AT THE CALL (Espryt: SyncImageTextureBinding at DirectGLES.cpp:9154, glPatchParameteri at :8760, the XFB object rebind at :1390, the driver-program rebinding at :9200) that no validate-time set record reproduces; the fifth (copy_framebuffer_to_texture) has a source no resource row can name. Overturned by the backend moving that work into its draw prep - G5-pinned regions, so not P5b's.

  4. resource_copy_region = glCopyImageSubData only, and the framebuffer-sourced copies get row 76 - which also settles the stamp row FieldOwnership.def had left to "the phase that emits it".

  5. kCapNeedsHostIndexBytes stays 0 while the user-index span is ARMED. The cap is the server asking for VBO-backed index bytes (P8's mirror); the span is the client handing over bytes only it has (the P8 resolve-on-client rule applied to the one case that needs it). CONTRACT-P5 table 0 conflated them because P5 produced neither. Overturned by a client index array larger than a SEG_STAGE run at 32 MiB (R-10's maxrec= would say).

  6. Renderbuffer copy endpoints, unbound DSA clears, and multi-draws with client-side indices are refused BY NAME (ID-57's shape) - no barrier-pulled forward can serve them and none was measured. Overturned by a measured workload hitting one.

  7. The copy-image-shadow-mirror site is skipped under split (behind #if MOBILEGL_BUILD_DISAGGREGATED, i1's edit); the two mipmap emulation sites stay Fatal (one may be re-derived from the descriptor by f1; the CPU fallback needs P9's OnTextureWriteback). Overturned by a measured GetTexImage of a copy destination before P9 - which is the wave-3 GetTexImage Fatal anyway.

  8. GetProgramForDispatch FATAL → BARRIER_PULLED "P7 (Magma), P8 (Espryt)" (i1's one-row edit; granted). kCapBackendOwnsXfbCapture (1<<9) for the EndTransformFeedback probe at GL_Drawing.cpp:1290 (t2's edit; granted). The only two c0b-file edits a package makes.

  9. The emit-table class-C list is partitioned per package with five ownership static_asserts, so four parallel landings edit four disjoint regions and a package that touches another's list breaks the other's line. Arithmetic; nothing overturns it.

  10. MGPCopyRegion 64 → 72. No producer, no consumer, no wire history; the names are the pull keys the P5b sink needs beside the handles P7 will use. Overturned by P7's handle-keyed endpoints, when the two fields become padding (and stay in the layout).

  11. Client-side vertex arrays are not a P5b question: served on the server through the barrier-pulled GetBoundVertexArray (CONTRACT-P5 table 2 "P8"), correct under inproc by the barrier and the shared address space, retired by P8's HostResolve.cpp. Overturned by P6's spawn - which ID-68 puts after P5b.

  12. Companions ride for free but are not gated. Nine draw variants, two compute/barrier variants, three XFB span controls and four clear/copy variants share a measured row; each package owns their slots (the ownership counts include them) and may flip them in the same change, but the census gate names only the 25.


5. What I believe the brief or the census got wrong or incomplete

  1. The brief's IVerbSink is MG_Remote::Wire::WireVerbSink (declared in the codec header so the codec does not depend on the session), and its server implementation is Server::ServerVerbSink. Nothing else in the tree is named IVerbSink.
  2. The census's "reuse MGPipeApplySetShaderImages / MGPipeApplySetPatchState" for BindImageTexture / PatchParameteri does not work: both appliers write the working block at VALIDATE time and the backend slots do their work AT THE CALL (§4.3). Those two slots needed rows of their own, which is where two of the five appended rows come from. The census's "add/finish MGPipeApplyLaunchGrid / MGPipeApplyDrawVbo / MGPipeApplyClear" is likewise not the shape: the census's own correction (verbs reach the sink) is.
  3. ShaderStorageBlockBinding was documented as "folded into the reflection archive" (PipeCalls.def footer, BackendObject.h). The archive carries the bindings as LINKED; glShaderStorageBlockBinding moves one AFTER link and both backends apply it to an already-built program. It needed a row (75); the footer is corrected in place.
  4. The 25 are 25 only if DrawElementsInstancedBaseVertex counts (trace-only, 0 lane entries). The brief's d1 list has 10 with it; the census's integration table has 24 non-zero rows. Both are right; the contract lists all 25 by name.
  5. FieldOwnership.def's stamp map covered the pre-existing rows but not ResourceCopyRegion, by design ("the phase that emits it adds its row"). Left to a package, an omitted stamp row is the ONE silent failure the table has (the record applies under the previous verb's serial). All six rows are landed here, not left to the packages.
  6. feat/disaggregated moved twice while c0b ran (c1g at c9878d69 before I forked; v1 round 3 at f5614995 after). Merged; one four-line overlap in PipeApplier.h, no conflict.

6. Debts, for the packages and the phases after

  • rsp per Minecraft frame is unmeasured until d1 renders one; each package reports it for its scenarios (PACKAGE-PREAMBLE §"五件事" 4).
  • Renderbuffer glCopyImageSubData endpoints need a renderbuffer forward or P7's handle keying (§4.6). Unbound DSA clears need Espryt's handle-keyed SyncAndBindFramebufferObject sibling (P7 / P3b-P4b, or f1 behind the option). Multi-draw with client-side indices needs P8's HostResolve.cpp.
  • DeleteTransformFeedback has no row: the driver object leaks on the server until P9's XFB namespace work; a bind of name 0 is what the backend does on delete of the bound one.
  • MGPGridInfo::Block* is 0 in P5b; P7 Magma fills it from the reflection archive if it needs it.
  • The Fatal NAME a pull prints under a P5b draw is DrawArrays for all twenty draw verbs (one stamp row, one kDraw mask). The wire's UnmigratedVerb names are exact; the poison's @DrawArrays is not, and d1's reports should say so once rather than be surprised.
  • P5's .text baseline moved between c0 (10806323) and this head (10806611) with the landed P5 packages; p5-before-libMobileGL.so (.READY 4595163c) is the comparand and the report's G1 is against it.