Skip to content

ci: inspect iOS launch rejection in an isolated cloud run - #7157

Closed
antuz123 wants to merge 433 commits into
sprintfrom
codex/r5-ios-only-diagnostic-scheduling-20261024
Closed

antuz123 wants to merge 433 commits into
sprintfrom
codex/r5-ios-only-diagnostic-scheduling-20261024

Conversation

@antuz123

@antuz123 antuz123 commented Oct 24, 2026 •

Copy link
Copy Markdown
Collaborator

State: CLOSED_SUPERSEDED_NOT_ACCEPTED

Superseded by #7164, published source a23d9a9c92cb4cdddca03e07844a722805f74f24. This draft's complete source history is incorporated in that candidate.

This administrative closure grants no native, physical, security, combined R5 or release acceptance. All prior failures, NOT RUN items, handoffs, source-specific evidence and physical-scope decisions below remain unchanged. Continue work in #7164.


Cloud iOS startup still fails in the owned LAUNCH_APP command before the PIN assertion. The earlier installed-byte proof precedes Maestro's clearState copy/uninstall/reinstall operation. This continuation measures the remaining installed bundle after failure and preserves the failed startup result.

Head 7c00dc22937f94111298350a767466354f239ba9; tree ba043ab4acab7cfad078deb4f08fc3143d918a47. Full reviewed #7155/c85 ancestry is retained. Seven-path delta against frozen bea.

Behavior and selection

  • On startup failure, query only the owned current simulator bundle; compare its typed inventory, executable bytes, signature, ad-hoc identity and effective entitlements with the built input. Recheck both inventories and the final owned container before retaining fixed observations.
  • Retain at most eight allowlisted Apple error-domain/int32 code pairs from the owned failed launch command. No raw descriptions, paths, logs or identifiers are exported.
  • The optional keeper/ios-diagnostic-only modifier, together with keeper/native-diagnostic, skips only this diagnostic workflow's Android job. The modifier alone starts no app job. Default iOS-then-Android behavior, source/privacy/installed-byte guards and canonical/full acceptance requirements remain unchanged.
  • This selection is for an iOS-only investigation. Android on this source will be SKIPPED, with no transferred Android pass or acceptance waiver.

Evidence and limits

Exact-source 392 Python tests PASS, 28 diagnostic/draft/source-identity Node controls PASS, and both native workflow actionlint checks PASS (shellcheck/pyflakes unavailable). Diff/clean-source checks PASS; app/native/dependency inputs and shared/canonical/full-acceptance worker workflows are unchanged against bea. Independent post-replacement review cleared37 focused controls and the late typed-built-inventory drift canary; independent scheduling review cleared28 controls and all eight label combinations.

Previous #7154/run40078818062 is terminal FAILURE on frozen bea. iOS build/signature/pre-UI bytes passed; LAUNCH_APP failed with FBS/simctl markers. Android build/startup/persistence/worker probes passed, then its first product sample failed waiting for Start New after the host Skip acknowledgement. #7156 safely recovered those Android facts; actual tap delivery and both underlying causes remain unknown.

No repair, retry, re-sign, product-flow/deadline change or production application change is included. A conditional copy hazard is source-confirmed; an actual bundle mutation or signing cause is not yet established. Native execution on this source has not run at PR creation.

Cloud native execution

Selected iOS-only run40084560886/attempt1 is COMPLETED / FAILURE. iOS job 120844708599: dependencies, disposable driver readiness and build PASS; startup failed at recorded LAUNCH_APP sequence2. Its sole parsed error pair is FBSOpenApplicationServiceErrorDomain / 1, which does not establish the underlying denial cause. Later iOS persistence, worker and product phases were skipped. Android job 120852015637 actually SKIPPED as selected; it receives no acceptance credit.

The post-failure collector returned UNKNOWN at INSTALLED_CONTAINER_POLICY; replacement bytes, signature, identity and effective entitlements were not measured. A source review explains a diagnostic admission mismatch: pinned Maestro reinstalls from a timestamped $bundleId-<milliseconds>.app name, while the reused initial-install path policy requires hexa_keeper_dev.app. This is a diagnostic defect, not evidence of the application's launch-denial cause. The original failed outcome remains primary; no guard was relaxed or app repaired in this run.

Retained artifact 12557780480, 7,871bytes, ZIP SHA256 56ce0392d9578fdd2e0808197a2d77c9f83e5983f2f89c9d821bce19ae0d9dff. Official referenced platform workflow SHA ee60cd2627d3a32962b158fc6a00434021f8a0d1, refs/pull/7157/merge. Source remains frozen. No Keychain acceptance, installed trust, physical acceptance or release authority.

Codex local handoff

State: NEEDS_LOCAL
Base: sprint
Head branch: codex/r5-ios-only-diagnostic-scheduling-20261024
Head commit: 7c00dc2
Goal and acceptance criteria: One coordinator-selected source-bound iOS cloud diagnostic window; preserve failures and recover exact post-Maestro bundle/code observations.
Changes made: Seven .github/documentation paths described above, retaining reviewed c85 ancestry.
Cloud checks: 392PythonPASS;28NodePASS;2actionlintPASS; independent package reviewsCLEAR; unchanged app/native/dependency/shared-worker fencesPASS.
Required local checks: IOS, ANDROID, HARDWARE
Local checks requested: This diagnostic selection runs cloud iOS only; Android and physical/full-product obligations remain open separately. No personal Mac.
Risks / known gaps: Cause remains unknown; original bea failures preserved. No installed trust, full software matrix, hardware acceptance, shared-branch promotion, release signing/upload/distribution or deployment authorization.

redkeep-agents and others added 30 commits October 13, 2026 03:48
redkeep-agents and others added 22 commits October 22, 2026 23:31
Source-review baseline only. Tree is exactly cd9d50a; second parent e343310 is preserved without producer rewriting. Its selected app and canonical collection changes are applied in separate reviewed descendants A and B. This baseline is not a native or release candidate.
Source-review join only. Tree is exactly B642fc14d1f2c3bd3451fe257642b01c6d1c07c95; second parent044ebb6ea2d465f6a70e9116e3feac5a604cd696 is preserved. Selected external thirteen-file work is exposed separately in C1 Android markers and C2 Maestro snapshot descendants. This join is not a native or release candidate.

Copy link
Copy Markdown
Collaborator Author

Coordinator-selected iOS-only cloud diagnostic — 24 October 2026, 10:05 UTC.

Freeze exact 7c00dc22937f94111298350a767466354f239ba9, tree ba043ab4acab7cfad078deb4f08fc3143d918a47. Reviewed c85 post-Maestro observation ancestry is retained. Root independently ran392Python controls,28diagnostic/draft/source-identity controls and two actionlint checks: PASS. Independent package reviews CLEAR. App/native/dependency inputs and shared/canonical/full-acceptance worker workflows are unchanged against bea.

Previous full native run40078818062 is terminal FAILURE; both retained artifacts are immutable. Android's finite readout40084255557 is terminal SUCCESS for bounded observations only. The separate private Linux dependency check is terminal SUCCESS. Fresh complete global app-repository snapshot has zero queued runs and nine in-progress contract/CodeQL/CI/backend/provenance workflows, with no active native app window.

The live PR is draft at the frozen source and currently has only keeper/ios-diagnostic-only; modifier-only native scheduling is inactive. Add keeper/native-diagnostic once to select one IOS job. Android for this diagnostic must be SKIPPED; that supplies no Android pass or waiver. Keep source frozen until terminal; no duplicate window, cancellation, repair/re-sign/launch retry or source promotion.

Report actual original failure plus read-only post-replacement inventory/signature/entitlement and numeric error-code facts. Root causes remain unknown until supported. Integration8b56, R4/1d8, earlier native sources, and private security proposals remain separate. No installed trust, hardware, release signing/upload/distribution, merge or deployment authority.

antuz123 commented Oct 24, 2026 •

Copy link
Copy Markdown
Collaborator Author

Frozen iOS-only cloud diagnostic — terminal result

Source 7c00dc22937f94111298350a767466354f239ba9, tree ba043ab4acab7cfad078deb4f08fc3143d918a47 remains frozen.

Run40084560886/attempt1 is COMPLETED / FAILURE. iOS job 120844708599: dependencies, driver readiness and build PASS; startup failed at recorded LAUNCH_APP sequence2. The sole parsed error pair is FBSOpenApplicationServiceErrorDomain / 1; underlying cause remains UNKNOWN. Later iOS persistence, workers and product phases were skipped. Android job 120852015637 actually SKIPPED as selected; no Android pass transfers to this source.

Post-failure replacement observations are UNKNOWN at INSTALLED_CONTAINER_POLICY. Bytes, signature, identity and entitlements were not measured. Pinned Maestro source confirms that reinstall uses a timestamped $bundleId-<milliseconds>.app name, while the diagnostic reused an initial-install policy requiring hexa_keeper_dev.app. This explains the diagnostic rejection; it does not explain the app launch denial. A separate minimal diagnostic correction is being prepared without modifying this frozen source or its canonical acceptance guard.

Artifact 12557780480: 7,871bytes, ZIP SHA256 56ce0392d9578fdd2e0808197a2d77c9f83e5983f2f89c9d821bce19ae0d9dff. No launch retry, signing repair, acceptance relaxation, source advancement, personal Mac, release signing/upload/deployment/merge. Parallel Android #7158 remains a separate source and run.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants