Repository navigation
Retain practice completion counts across deferred listings - #591
Chris0Jeky wants to merge 5 commits into
Conversation
|
You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard. |
|
Independent review triage: Grok4.7 high found the Android contentManifestRevision producer/consumer mismatch (HIGH). It was reproduced on a clean baseline Android build plus failing new regression, then fixed in d1458a0. Fresh read-only Sol medium review of the correction found no CRITICAL/HIGH or other actionable finding. Two unassessed LOW coverage ideas about listing clones/current predicates are not confirmed defects; declined further expansion in this bounded wave. Existing listing regressions cover the behavior being corrected. Final committed-head build:android, check:android, budget, Prettier and diff checks exit0. Code-identical correction suites pass24/24; prior current-base focused suites pass30/30. Five gzip bytes of JavaScript headroom remain, so combined integration needs a new exact-head budget check. Actual-browser practice/physical-device and hosted CI gates remain open. |
Preserve all eight PR files by their exact existing blobs. The comparison from prior base 2758d91 to main cde5f56 contains no overlapping paths. This refreshes combined source evidence; earlier CI is not promoted to current-head acceptance. Actual pending/failed deferred browser acceptance and unchanged byte budgets remain merge gates.
… saves Keep all reviewed production code unchanged. Exercise pending, rejected, retried and reloaded official definitions at both viewport widths through the real app-owned adapter and native IndexedDB. Preserve the synthetic saved record at every phase. Add exact-head source archive/hash receipts and full unchanged-budget verification before native acceptance. Python compilation passes locally; browser execution remains a CI gate.
|
Current-main acceptance is now complete for the previously missing practice/deferred seam at 36a7e9b, tree 2678ea9e556d9509857a25fae8cdb60ed6483ca9. The eight original PR blobs, including all reviewed production changes and the Android correction, are unchanged. Main cde5f56 was merged without overlap; the only added files are the browser proof and its read-only workflow. Exact-head run 37692449381 passes pinned npm verify: 1,317 Node tests, zero failures/skips, then Quiet Wing 581,847 + 29 assertions. All original ceilings pass: application gzip 134,939, net shell 1,411,599, total initial transfer 204,279. No threshold change. The new native Chromium 154 proof passes eight phase/viewport cases: pending, failed, retried and reloaded official definitions at 390px/1280px. It calls the actual emitted application's owned practice adapter, holds and aborts the real deferred script request, and uses native IndexedDB. The Observatory remains 1 completed / 55 total in all eight cases, and the full synthetic stored record stays equal to the original after every phase. This is actual app-owned adapter acceptance, not a claim about visible Castle panel layout, physical devices, service-worker lifecycle or deployment. Downloaded artifact 11513329619 SHA-256 3bf1c9fc1cc268b11e7526c6d87598947d2be5586e8b1c35a57f3c76d6ebc3a2 and all captured source hashes were independently verified; clean build and native identities match the exact head. Existing Grok/Sol review and triaged LOW findings remain applicable because reviewed production bytes/risk boundaries are unchanged. No inline unresolved review threads were returned. Remaining broad CI jobs must finish before merge; HUMAN_TODO.md and physical/TalkBack gates remain open. |
1 participant
Product behavior
Historical practice completions count against registered deferred listings while their full chunk is pending or failed; loaded definitions retain exact-body matching. Content identity binds startup plus deferred definition bytes in both web and derived Android receipts/checks.
Refs #389. The route pending/loading timeout behavior already existed and was not rewritten. Save formats, IDs/revisions, lazy loading and numerical budget ceilings remain unchanged.
Current-main integration
Head 36a7e9b, tree 2678ea9e556d9509857a25fae8cdb60ed6483ca9, based on main cde5f56.
The eight original PR files were retained by their exact Git blobs while merging nonoverlapping current-main changes. The only new files are
tests/browser_practice_deferred.pyand.github/workflows/practice-deferred.yml. Reviewed production bytes and risk boundaries are unchanged.Executed exact-head evidence
Dedicated Practice deferred acceptance run 37692449381 passes:
npm run verify, clean web/Android builds, 1,317 Node tests with zero failures/skips, then Quiet Wing 581,847 + 29 assertions;The native test holds and aborts the actual deferred script request, invokes the emitted application's owned practice adapter and uses real IndexedDB. Observatory completion stays 1 of 55 through all eight cases, with the exact synthetic stored record preserved after every phase. This closes the previously missing actual app-owned adapter acceptance, not visible Castle-layout or physical-device acceptance.
Downloaded artifact 11513329619, ZIP SHA-256 3bf1c9fc1cc268b11e7526c6d87598947d2be5586e8b1c35a57f3c76d6ebc3a2, was independently checked. Captured input hashes match its source archive; clean build/native identities match the exact head. Broader repository CI is tracked separately and must finish before merge.
Review and historical proof
Independent Grok review found the Android startup-only content-hash mismatch. The correction was red-proved and reviewed by a fresh read-only Sol medium lens with no remaining actionable blocker. Two LOW coverage ideas were explicitly declined on the thread, not silently dropped. The current refresh does not modify those reviewed production files, so it does not invent another review loop. No unresolved inline review thread was returned at the integration checkpoint.
Earlier focused evidence remains valid within its scope: six suites 30/30, corrected Android/identity/content/budget selection 24/24. The new exact-head receipt above supersedes the old NOT-verified statement about browser adapter pending/failure coverage. Five application gzip bytes of headroom remain; every future integration must re-prove the unchanged limit.
Not claimed / remaining boundaries
Do not merge before all required current-head CI and the existing aging gate are satisfied. No deployment, physical Android/TalkBack, full visible Castle-panel review or service-worker lifecycle acceptance is inferred from the new adapter test. HUMAN_TODO.md remains authoritative and unchanged. The earlier refused Muse host-ownership launch is not counted as a worker or review.