Files
MobileGL/docs/Disaggregated/notes/p5b/p5b-results/r1-codex-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

7.2 KiB
Raw Blame History

P5b r1 — server ownership, readback byte order, retirement back-pressure

Completed 2026-09-16. Branch p5b/r1-codex, isolated tree /home/swung/w7/p5b-r1-codex, commit db8b3479520862cd47c546e96506b7332158abcf. Base is r2 d50183cb, so merge r2 first or merge this branch including its ancestry. This implements ID-73 findings 1–4; it is not another adversarial review or a phase gate.

The inherited Windows tree /mnt/c/Users/geekerwan/p5b-r1 was at 37fc4fdb with three uncommitted files: PersistentMapTracker.cpp, .h, and BufferObject.cpp. Their complete file contents were copied into /home/swung/w7/p5b-r1-codex-backup/ before continuing in this new tree. The old tree and any old Claude process were left alone.

Changes and evidence

Finding Implementation Focused evidence
1: coherent map bypass Under split, retained backend SyncPersistentMappedRange calls return on the apply thread. PushAllMembers, PushBlocksFor, and the final PushMappedSpanBlock producer refuse server-role entry with Fatal{RoleViolation,...}, before reading map bytes or modifying frontend content serials. Server map-state teardown keeps membership valid but does not emit client live-write state. Real GPU store readback test now holds a coherently mapped frontend object with bytes A and a server shadow with bytes B; Ensure uploads B. A live client session also asserts the producer's named apply-thread refusal. Separately, removing only client pre-verb push calls makes both coherent-pointer pixel cases fail while backend sync sites remain intact; restoring the calls makes both pass.
2: framebuffer death on client GLES' shared frontend-death callback forwards synchronously through the existing apply-thread control mailbox before any twin destruction or binding-cache changes. This covers every kind in that callback; framebuffer still has no invented wire opcode. Native glDeleteFramebuffers interception observes one deletion on the apply thread and verifies that the native draw binding becomes zero. The previously failing DirectGLES.P4aFinalFixScenario.ARenderbufferAndItsFramebufferDeletedAfterAClearLeaveTheNextDrawIntact now passes.
3: PACK_SWAP_BYTES After validating the tight reply, swap each requested component/packed word on the client. Apply to both direct and bounce paths, before scatter; byte types remain unchanged. FLOAT_32_UNSIGNED_INT_24_8_REV swaps two separate 32-bit words. Live emit-table peer covers ushort, float, packed uint, mixed depth/stencil, and byte data with swap off/on and tight/scattered layouts; sentinel gaps are preserved. Actual GL test reads a non-symmetric ushort from RGBA16F with tight and skipped/strided pack destinations: both monolith and inproc pass. RGBA8 was deliberately avoided because expanded ushort components repeat the same byte and cannot expose a missing swap.
4: applied before retired Preserve r2's staging allocator wait and truthful wait counter unchanged. ClientSession::EmitAndWait now treats command-ring capacity exhaustion as back-pressure after the encoder's individual-record size check. It waits for enough retired bytes including the wrap pad, then retries the same record without changing its sequence. Shutdown/timeout gets a named RetirementWaitFailed instead of a silent drop or indefinite wait. Two live-session cases hold the server between applied publication and retirement using r2's scheduling-only hook. The staging case fills 3/4 of 1 MiB twice and observes one wait. The command case positions a legal 393 KiB draw record so its successor needs a wrap pad, observes the producer parked, releases retirement, and verifies the next ordinal is exactly prior+1.

The death callback uses the existing inproc control mailbox, not a new wire lifetime record. P6's separate address space still needs the already-planned lifetime/control transport work. Staging's r2 five-second shutdown-aware bound and once-per-blocked-allocation counter are intact. The command wait uses the session's existing kBarrierTimeoutMs (30 seconds); individually oversized records still fail at PipeWireEncoder::EncodeRecord.

Executed acceptance

All paths below are under /home/swung/w7/.

  • Split build completed, CTestTestfile.cmake verified by the build driver. Logs: p5b-r1-codex-config-split.log, p5b-r1-codex-build-split.log, p5b-r1-codex-build-fixes.log. Build concurrency was at most 8.
  • Final targeted run: 10/10 passed, zero skips, MOBILEGL_TRANSPORT=inproc, MOBILEGL_ITEST_REQUIRE_GPU=1, p5b-r1-codex-final-targeted.log. Six unit entries plus four integration entries: the six ownership/wait/swap controls, actual ushort swap pixels, FBO/RBO deletion pixels, and two coherent-pointer draw cases.
  • Actual GL ushort swap: 1/1 monolith and 1/1 inproc, zero skips, p5b-r1-codex-swap-{monolith,inproc}.log.
  • Client-only push suppression: two own pixel failures (CTest exit 8), plus the existing map-arm metadata case's one intentional skip. The skip is not counted as a red. p5b-r1-codex-coherent-red.log; driver p5b-r1-codex-coherent-control.sh. It restores the source through an EXIT trap and rebuilds the restored integration binary. The restored two pixel cases are included in the final 10/10 run.
  • git diff --check passed. Final worktree clean, submodule links off. Commit is a single line without trailers, author/committer Swung0x48 <swung0x48@outlook.com>.

The primary negative control removes only PushPersistentMapsBeforeVerb() from the client pre-verb hooks. It does not set the block-size-zero knob, alter the backend's retained sync sites, or remove the new server-role guard. It therefore proves client omission is no longer repaired by server producer re-entry.

No new all-lane census, four-flavor unit run, G1 binary comparison, full integration lane, device operation, push, or phase review was run here: the integrator explicitly owns the combined quick gate. All production changes outside MG_Remote are inside #if MOBILEGL_BUILD_DISAGGREGATED, so pull code is structurally unaffected; this is not reported as a new measured G1 result. No .def, generated catalogue, contract, or opcode changed.

Integration notes

  1. Keep r2's StageAllocate implementation, its session doorbell hookup, telemetry semantics, and before-retire hook. This commit changes command wait policy in ClientSession only; it does not replace r2's codec implementation.
  2. Five new unit entries; the existing staged-shadow unit is strengthened to use a persistent coherent mapping. One new ordinary integration scenario, registered for both backends; integration-split remains the same 22 selected entries. The new scenario is PixelStoreSweepScenario.ReadPixelsSwapsUnsignedShortComponentsAndPreservesPackGaps.
  3. The known 27 inproc ordinary failures should no longer be relabeled wholesale as unchanged: the GLES FBO/RBO deletion case is now green on this isolated tree. The merged census must determine the new aggregate; no count was extrapolated from this single case.

Tools retained: p5b-r1-codex-submodlinks.sh, p5b-r1-codex-build.sh, p5b-r1-codex-coherent-control.sh; external backup directory above. No integration worktree, other package source/build directory, adb device, or remote ref was modified.