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.
7.5 KiB
P5b Codex APK / Redmi runner handoff
2026-09-16. Dedicated scripts; original campaign scripts and evidence remain intact.
Build state
Final source head: 82683d4a83b5c8f4f7c375a7a92ebee9a1854646, including both closing-review fixes. The first real build started at 2026-09-16T14:12:42-04:00 in phase p5bcodex1 and was interrupted by an entire WSL restart at approximately 14:14 (retained exec session exited 1; uptime reset to 0 minutes; all native/Gradle processes disappeared). Its output directory, build log and native cache are preserved. Pull reached mergeTraceReleaseJniLibFolders; no signed APK/proof was produced.
State: waiting for the root task's memory/concurrency coordination. The next authorized build will use p5bcodex2, the same exact source head and dedicated APK tree, and a fresh output directory. The commands below name that next phase; they must not be run until its three APK proofs exist.
Build tree: /home/swung/w7/p5b-apk-codex; output: /home/swung/w7/notes/p5b/apk/p5bcodex2.
The wrapper notes/tools/p5b_codex_build_apks.sh requires --head with the full SHA and a clean dedicated checkout. It shares the old wsl_build_p5_apks.lock, refuses an existing output directory, retains each arm's build log and proofs, and builds pull/push/split sequentially. P5B_APK_JOBS=4 is the default. A generated Gradle init script also sets CMake compile/link job pools, because AGP invokes Ninja directly. No tracked source build file is modified.
In WSL Arch, after the root task has moved the clean APK tree to the final source commit:
HEAD_SHA=82683d4a83b5c8f4f7c375a7a92ebee9a1854646
test "$(git -C /home/swung/w7/p5b-apk-codex rev-parse HEAD)" = "$HEAD_SHA"
P5B_APK_JOBS=4 bash /home/swung/w7/notes/tools/p5b_codex_build_apks.sh --head "$HEAD_SHA" --phase p5bcodex2 /home/swung/w7/p5b-apk-codex
Prerequisites: local hydrated ~/w7/pipe dependencies, JDK 21 at /usr/lib/jvm/java-21-openjdk, SDK at ~/android-sdk, Gradle 8.10.2 and its offline dependencies. The native build must be optimized RelWithDebInfo, arm64 only. Each proof checks source SHA, embedded Git stamp, package/signing/ABI, push/remote symbol identities, and byte-for-byte equality between the packaged library and stripped build library. The three outputs are trace-pull.apk, trace-push.apk, and trace-split.apk; splitctl reuses the split APK with monolith runtime.
Windows Git Bash setup and prerequisites
Run in Windows Git Bash, not WSL, since the existing Android runner uses Windows Python/adb. The root task owns all device operations and synchronization of the clean Windows source tree.
TREE=/c/Users/geekerwan/AndroidStudioProjects/FoldCraftLauncher/MobileGL-disagg
TOOLS=//wsl.localhost/Arch/home/swung/w7/notes/tools
APKS=//wsl.localhost/Arch/home/swung/w7/notes/p5b/apk/p5bcodex2
FIXTURES=//wsl.localhost/Arch/home/swung/w7/pipe/tools/trace_replay/fixtures
OUT_ROOT=/c/Users/geekerwan/AndroidStudioProjects/FoldCraftLauncher/MobileGL/.trace-work/p5b-redmi
PHASE=p5bcodex2
cd "$TREE"
command -v bash git python adb
python --version
HEAD_SHA=$(python -c 'import json,sys; print(json.load(open(sys.argv[1]))["source_head"])' "$APKS/trace-pull.proof.json")
test "$(git rev-parse HEAD)" = "$HEAD_SHA"
git diff --quiet --ignore-submodules=all HEAD --
for arm in pull push split; do
test -s "$APKS/trace-$arm.apk" && test -s "$APKS/trace-$arm.proof.json" || exit 1
done
python "$TOOLS/p5b_codex_runner_selftest.py"
All four fixture archives and their manifest goldens must already be hydrated. The adapter validates their existence/content and refuses Git LFS pointer files; it never fetches LFS. Case names:
improved-transparency-minecraft-26.3minecraft-1.21.4-in-worldminecraft-1.21.4-fabric-sodium-in-worldminecraft-1.21.4-fabric-iris-bsl-in-world
Dry runs perform no device operations, locking, builds, or file writes:
bash "$TOOLS/p5b_codex_redmi.sh" --head "$HEAD_SHA" --phase "$PHASE" --mode correctness --tree "$TREE" --out-root "$OUT_ROOT" --dry-run
bash "$TOOLS/p5b_codex_redmi.sh" --head "$HEAD_SHA" --phase "$PHASE" --mode benchmark --tree "$TREE" --out-root "$OUT_ROOT" --dry-run
Expect DRY_GROUPS=8 and DRY_GROUPS=32. The target is fixed at Redmi serial 2f7cbe2e, which must support adb and su after reboot. The existing pin script must be readable at //wsl.localhost/Arch/home/swung/w7/notes/p2/devices/pin_device.sh. The session uses the existing campaign lock at /c/Users/geekerwan/AppData/Local/Temp/claude/C--Users-geekerwan-AndroidStudioProjects-FoldCraftLauncher-MobileGL/8ab6b3cb-c783-4826-9fd7-39ff3498af5b/scratchpad/w7/locks/2f7cbe2e; it waits at most 900 seconds and never breaks another owner's lock. No other campaign should operate that device concurrently.
Device commands (root task only)
bash "$TOOLS/p5b_codex_redmi.sh" --head "$HEAD_SHA" --phase "$PHASE" --mode correctness --tree "$TREE" --out-root "$OUT_ROOT"
Correctness must finish with 8/8 groups and session-exit-code.txt equal to 0. Each case/backend requires an actual fresh PNG, the manifest target call and SSIM threshold (including improved-transparency 0.995), successful replay, and private-log proof of inproc, apply thread, and positive wire records. A skip is failure. Then:
python "$TOOLS/p5b_codex_matrix.py" require-correctness "$OUT_ROOT/$PHASE/2f7cbe2e/correctness" --head "$HEAD_SHA" --phase "$PHASE"
bash "$TOOLS/p5b_codex_redmi.sh" --head "$HEAD_SHA" --phase "$PHASE" --mode benchmark --tree "$TREE" --out-root "$OUT_ROOT"
Benchmark is four arms × two backends × four cases = 32 groups, with three repeats each and the existing nofinish/last-200-frame protocol. The reducer retains the previous lowest-mean-repeat selection and reports p50/p99, CPU, mapped views and available memory evidence. No timing delta is a pass/fail threshold. Failed replay, missing evidence, short samples or pin drift invalidate a group and exclude it from paired metrics.
All arms explicitly use MOBILEGL_IPC_STAGE_MB=256, the authorized P5b exit profile for the 128 MiB improved-transparency upload. This leaves the production 32 MiB default unchanged. Stage mapping is 256 MiB per role view / 512 MiB summed inproc views; this is not an inferred RSS increase. Other controls are MOBILEGL_IPC_VERB_BARRIER=1, stats enabled, and transport inproc for split / monolith for pull, push and splitctl. rd12 is excluded from this matrix.
The session reboots, fixes fan level 2, checks temperature below 40000 mC, pins/checks before and after each group, and cleans up unpin/fan0/stayon-off before releasing its own lock. Each retry uses a new attempt directory; rerunning a completed mode requires a new --out-root or phase. Do not remove old evidence to rerun.
Outputs are under $OUT_ROOT/$PHASE/2f7cbe2e/{correctness,benchmark}: matrix.json, matrix.md, cleanup exit, per-group attempt/runner exit and pin evidence; benchmark also produces ab-tables.md. Per-repeat originals include result JSON, private mobilegl.log, screenshots and benchmark JSON. Missing/failed evidence cannot produce a green matrix.
Runner checks already performed
Both shell scripts pass bash -n; Python modules compile. Real Windows Git Bash dry runs produced exactly 8/32 groups, and the Windows adapter imports the existing Windows runner correctly. Nine artifact-input selftests passed on Windows and WSL (skip, SSIM, missing/truncated PNG, split proof, wrong splitctl runtime, stale directories, short benchmark). Logs: ~/w7/p5b-codex-redmi-dry-{correctness,benchmark}.log and ~/w7/p5b-codex-runner-selftest.log.