test(native-profile): bound hosted macOS busy-probe case

[Ingwannu]

Summary

• give the deterministic busy-process rejection case its own 10-second Bun harness budget
• document why hosted macOS full-suite setup can cross the default deadline
• leave the product's SQLite lock deadline and CODEX_BUSY behavior unchanged

Closes #1563

Verification

# repeated five times under the requested 2-core limit
taskset -c 0,1 bun test --isolate tests/native-profile-manager.test.ts --test-name-pattern 'normal switch rejects a busy native Codex process'
bun run typecheck
bun run privacy:scan
git diff --check

Focused result: 5/5 passes. Typecheck and privacy scan are green.

Checklist

• The change is limited to the flaky test harness budget.
• The product timeout and runtime behavior are unchanged.
• The hosted-macOS reason is recorded next to the test.
• No GUI or wording surface changed.

Summary by CodeRabbit

• Tests
  • Increased rollback and busy-process test timeouts to 10 seconds for more reliable test execution.
  • Added comments explaining the timeout rationale.

[github-actions]

✅ Deterministic PR hygiene checks passed.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: ASSERTIVE

Plan: Pro Plus

Run ID: 7810479e-d58e-4efa-b58c-3a68a00be2a3

📥 Commits

Reviewing files that changed from the base of the PR and between c1be34d and bef2869.

📒 Files selected for processing (1)

• tests/native-profile-manager.test.ts

---

📝 Walkthrough

Walkthrough

The rollback and busy-process tests now document setup timing on hosted macOS. Both test timeouts increase from Bun’s default to 10 seconds.

Changes

Native profile test stability

Layer / File(s)	Summary
Native profile test timeout adjustments tests/native-profile-manager.test.ts	Comments document encrypted-vault, SQLite, and shared fixture setup delays on hosted macOS. The rollback and busy-process tests use 10-second timeouts.

Estimated code review effort: 1 (Trivial) | ~3 minutes

Mergeability Score: ⚪ Minimal · up to bef28

This change only gives a flaky hosted-macOS test a bounded 10-second harness budget while leaving product timeout and runtime behavior unchanged; no actionable merge-blocking risk remains.

Possibly related PRs

• lidge-jun/opencodex#1507: Adjusts timeout handling in the same test file.
• lidge-jun/opencodex#1071: Modifies timing-sensitive native-profile tests.

Suggested reviewers: lidge-jun, wibias, luvs01

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly identifies the native-profile test timeout change for the hosted macOS busy-process case.
Linked Issues check	✅ Passed	The test file raises the busy-process harness timeout to 10 seconds and preserves the product lock deadline and CODEX_BUSY behavior required by issue #1563.
Out of Scope Changes check	✅ Passed	The changes are limited to documented test-harness timeouts in tests/native-profile-manager.test.ts and contain no unrelated product or GUI changes.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch agent/fix-1563-macos-timeout

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[Ingwannu]

The first exact-head run did not fail the changed native-profile case. GitHub annotated the unrelated existing tests/server-auth.test.ts:1802 WebSocket terminal timeout; all Linux shards and the other gates passed. I have rerun only the failed jobs once to distinguish that hosted-macOS flake from this one-line timeout fix. I will not mark this merge-ready unless the rerun is green.

[Ingwannu]

Correction: this environment's GitHub CLI does not support selecting only failed jobs, so the prior rerun command was rejected and changed nothing. I have now started one full exact-head rerun of run 31659392694; the original annotated failure remains the unrelated tests/server-auth.test.ts:1802 WebSocket terminal timeout.

[Ingwannu]

The exact-head macOS rerun exposed the same harness-boundary issue in the adjacent rollback verification: the product assertions completed correctly, but hosted macOS took 7.8 s and Bun's 5 s default expired. I added a documented 10 s budget only to that rollback regression; the product's lock and recovery deadlines are unchanged. Local validation on exact head bef2869c732adc9302a9bdacf408ee585d654a1e: focused case 5/5 under a two-core limit, full native-profile-manager suite 49/49, typecheck, privacy scan, and diff check green. Fresh exact-head CI is now running.

[Ingwannu]

Exact-head CI is now fully green at bef2869c732adc9302a9bdacf408ee585d654a1e, including the macOS full suite and all Linux shards/gates. The product lock/recovery behavior remains unchanged; this PR only gives the two hosted-macOS-sensitive transaction regressions a harness budget above the product's own deadline. This is my PR, so I am leaving approval and merge to an independent maintainer.

[Wibias]

No code blocker from my review. This stays scoped to the test harness: the 10s budget is documented next to the affected cases, while the product lock/process behavior remains unchanged. The exact head also went green across the repository CI matrix, including the hosted macOS suite that reproduced the flake.

Please rebase onto current dev and rerun CI; assuming that stays green, this is good to merge.