test(responses): make Windows shutdown spill fixtures deterministic

[lidge-jun]

Summary

Windows full-suite job 101339545421 exposed two shutdown-spill fixtures that depended on unmeasured filesystem timing. The stable-tail ordering case could enter an unmocked synchronous ACL fallback; the reserved-budget case controlled only the ACL clock and charged real disk time to its logical budget.

Make publication ordering explicit with path-specific gates, controlled timers, complete ACL mocks and cleanup. Use a shared logical spill/ACL clock for the reserve test while retaining its native drain timer, exact deadline-decrease checks, fallback proof and replay assertions. Production shutdown/ACL behavior and the separate deadline/watchdog tests are unchanged.

This independently verified foundation will be inserted below A's #3683 → #3686 → #3690 → #3692 → #3694 stack. The original Windows failure logs are preserved; it is not attributed to the SSE runtime fix.

Verification

• Failed Windows evidence: https://github.com/lidge-jun/opencodex/actions/runs/33978547130/job/101339545421 (3048 pass,39 skip,2 failures,1 unhandled error).
• Independent plan audit passed; git diff --check passed.
• No local test suite, typecheck or build executed, per owner instruction.
• Product candidate b2f24cada35ff998096cf7f3c83b95eef4ee55af: independent final fixture review PASS.
• Hosted Windows/Bun1.4.0 proof passed:143 tests,0failures; controlled fallback/clock cases passed; broken drain and reset-deadline variants each failed the exact intended assertion; retained watchdog controls passed. Typecheck and source restoration passed.
• Windows proof https://github.com/lidge-jun/opencodex/actions/runs/33981229052 executes a snapshot whose runtime/test/docs/dependency trees match the product candidate; its only extra files are the separately audited temporary verifier workflow/script.
• Full exact-product-head CI https://github.com/lidge-jun/opencodex/actions/runs/33981268532 remains required before admin merge.
• Any temporary hosted-Windows probe workflow lives only on a separate verification branch and will not be included in this PR.

Checklist

• Scope stays focused and avoids unrelated cleanup.
• Docs or release notes were updated when needed.
• Security-sensitive changes were reviewed for secrets, auth, and unsafe defaults.

Summary by CodeRabbit

• Bug Fixes

  • Improved determinism for Windows shutdown and spill-handling verification.
  • Stabilized checks for drain completion, fallback budget usage, timing, and replay behavior.
  • Added safeguards to ensure asynchronous failures are captured and cleanup completes reliably.

• Documentation

  • Added planning and review documentation covering Windows verification procedures, fixture behavior, audit steps, and validation criteria.

[coderabbitai]

Caution

Review failed

The pull request is closed.

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: ASSERTIVE

Plan: Team

Run ID: b6cf24fb-efb0-41ef-b366-8cac00d2354e

📥 Commits

Reviewing files that changed from the base of the PR and between 9115b17 and b2f24ca.

📒 Files selected for processing (4)

• devlog/_plan/260906_a_windows_fixture/000_plan.md
• devlog/_plan/260906_a_windows_fixture/010_fixture_plan.md
• devlog/_plan/260906_a_windows_fixture/011_review_resolution.md
• tests/responses/responses-state.test.ts

---

📝 Walkthrough

Walkthrough

The change stabilizes two Windows spill shutdown tests. It replaces call-count and wall-clock synchronization with target-based gates, controlled clocks, fake timers, and explicit promise cleanup. Planning documents define the test-only verification and Windows proof procedures.

Changes

Windows spill fixture

Layer / File(s)	Summary
Verification scope and evidence plan devlog/_plan/260906_a_windows_fixture/000_plan.md, devlog/_plan/260906_a_windows_fixture/010_fixture_plan.md	Defines the test-only scope, verification roles, Windows evidence procedure, remote execution fallback, and landing criteria.
Stable-tail shutdown fixture tests/responses/responses-state.test.ts, devlog/_plan/260906_a_windows_fixture/011_review_resolution.md	Gates ACL calls by spill temp path, freezes clocks, uses fake timers and native setImmediate, tracks drain and flush settlement, and joins promises during cleanup.
Reserved-budget shutdown fixture tests/responses/responses-state.test.ts, devlog/_plan/260906_a_windows_fixture/010_fixture_plan.md	Records per-target ACL timing, controls the spill clock, verifies reserved-budget decrements, and confirms replay without releasing the asynchronous gate.

Estimated code review effort: 3 (Moderate) | ~25 minutes

✨ 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 codex/a-00-windows-fixtures

---

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.}

[github-actions]

✅ Deterministic PR hygiene checks passed.

[lidge-jun]

리뷰 · 우선순위 58 / 80

이 PR은 Windows full-suite에서 깨진 shutdown spill 픽스처 두 개를 디스크 타이밍에 의존하지 않게 고칩니다. A 런타임 스택(#3683 → #3686 → #3690 → #3692 → #3694) 아래에 깔 기초 층(0)입니다. 지금 dev의 tests/responses/responses-state.test.ts는 ACL/drain 경로를 이미 다루지만, stable-tail 케이스는 unmocked sync ACL fallback에 들어갈 수 있고, reserved-budget 케이스는 ACL 시계만 통제하고 논리 budget에 실제 디스크 시간을 섞을 수 있습니다. 이 변경은 path-specific gate, controlled timer, 완전한 ACL mock/cleanup, 공유 logical spill/ACL clock으로 publication ordering과 deadline 감소를 결정적으로 만듭니다. 제품 shutdown/ACL 동작과 별도 watchdog 테스트는 건드리지 않는다고 합니다.

검증은 실패한 Windows job 증거, 호스티드 Windows/Bun 부분 증명, exact-product-head CI 요구를 본문에 적어 두었습니다. owner 지시로 로컬 suite는 안 돌렸습니다. #3683이 이 브랜치를 base로 쓰도록 스택이 잡혀 있어, A 트레인 랜딩 전에 먼저 넣는 순서가 맞습니다. types/config 분할과 무관합니다.

플래키 Windows verifier가 있으면 위층 SSE/WS/recovery PR의 CI가 전부 노이즈가 됩니다. 픽스처만 고치는 작은 표면이라 리뷰 부담도 낮습니다. draft이고 full exact-head CI가 아직 게이트입니다.

경로 tests/responses/responses-state.test.ts stable-tail - temp path gate + fake timers + stub-swap 기록으로 mid-drain publication 순서를 고정합니다. 실디스크 latency가 drain timer를 울리지 않게 합니다.

경로 같은 파일 reserved-budget - spill/ACL에 공유 logical clock, native drain timer는 유지, deadline 감소·fallback·replay assert 유지.

경로 plan 260906_a_windows_fixture - 근거/리뷰 해상도 문서입니다. 머지 블로커는 아니고 제품 코드 변경은 테스트 파일입니다.

경로 임시 verifier workflow - 별도 검증 브랜치에만 두고 이 PR에 넣지 말라고 본문이 못 박습니다. 맞습니다.

메인테이너의 판단이 필요한 지점

• A 스택 머지 전에 이 PR을 반드시 먼저 dev에 넣을지, #3683 base로만 두고 한 번에 올릴지
• Windows full-suite 초록을 이 PR 단독 게이트로 요구할지

너의 추천
A 스택 기초로 exact-head Windows CI 초록·draft 해제 후 먼저 머지를 추천합니다. 그다음 #3683 base를 dev로 옮기거나 ff로 쌓으세요.

이 댓글은 grok-bot이 작성했습니다

[lidge-jun]

Owner-authorized admin merge after independent fixture/security review and exact-head verification.

At b2f24ca every functional producer in full cross-platform run 33981268532 passed: all Linux shards, both macOS shards, the unsharded macOS control, all six Windows shards, type/privacy gates and platform smoke jobs. Only the deterministic aggregation job remained queued at the final check; its success/skipped allowlist was checked directly. No test execution or failure was bypassed.

Actual hosted Windows causal verification also passed in run 33981229052 on an identical runtime/test tree: 143 passing tests, bounded fallback/clock controls, exact negative-control assertions, retained watchdogs, source restoration and typecheck. The temporary verifier workflow is not part of this merge.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-09-05T18:13:50.611840Z	b2f24ca	Draft marked ready

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" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.