fix(e2e): harden iOS simulator diagnostics - #100
Conversation
|
Final exact-head verification is complete for
Residual hosted macOS/CoreSimulator long-tail risk is documented in Issue #99; this PR intentionally adds no retry, device rebuild, fixed sleep, or broader timeout. |
|
@codex review |
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 3e74467748
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| } | ||
| publishCandidate(candidate, stagingRoot, summary, state); | ||
| } | ||
| writeFileSync(join(stagingRoot, "publication-summary.json"), `${JSON.stringify(summary)}\n`); |
There was a problem hiding this comment.
When failure evidence reaches MAXIMUM_FILES or approaches MAXIMUM_TOTAL_BYTES, this unconditional write adds an uncounted file and bytes after the limits have already been enforced. The published artifact can therefore exceed the hard file-count or total-byte bounds promised by ADR 0045, while filesPublished and totalBytes still imply that the limits were respected; reserve and account for the summary within both budgets.
Useful? React with 👍 / 👎.
leon-zym
changed the title
1 participant
Summary
Scope and conclusion
PR #98's product commits are not supported as the cause. The observed failures drift across CoreSimulator app services, Photos services, and Maestro/XCUITest startup while the same acceptance transaction can also pass.
This PR completes the diagnostic and fail-closed foundation for #99. It does not claim that same-host build → E2E on the standard hosted
macos-26runner is statistically stable. That execution-plane decision is tracked in #101.The exact P0-head full run below is deliberately retained as a failed sample: the fresh Simulator passed guest health, app installation, and photo seeding, then Maestro's XCTest driver failed to start. The observation classified the failure as
driver-startup, preserved cleanup evidence, and exposed severe post-build host contention instead of retrying the run green.Verification
pnpm verify0d9443c53d4050ad2714eea852cfd11f07938bdd: failed at Maestro driver startup after guest health/install/seed passed; schema v2 recordedexecution:true,maestro:false,recorder:true,telemetry:false,failureBoundary:driver-startup; cleanup verified the owned Simulator absentFollow-up
#101 owns first-sample stability: build a sealed exact-SHA acceptance package, execute it on a fresh test host, and evaluate the accepted x86_64/macOS 15 execution plane with a staged, no-retry sample budget. This keeps later infrastructure experiments out of the already large #100 diff.
Closes #99