Isolate Claude CLI gating tests

[steipete]

Summary

• isolate TTYCommandRunnerActiveProcessRegistry behind a task-local state override so shutdown-fence tests cannot reject unrelated concurrent Claude login launches
• serialize the complete Linux PlatformGatingTests suite, completing the partial Claude CLI gating serialization from Serialize flaky Claude CLI platform-gating cases #2311
• make Claude auth-status timeout overrides task-local and widen only test fixture deadlines that proved load-sensitive; production defaults and behavior are unchanged
• harden the named Kiro pipe-deadline assertion without changing the probe implementation

Root cause

TTYCommandRunnerEnvTests is serialized only within its own suite. Its shutdown-fence cases mutate the process-global TTYCommandRunnerActiveProcessRegistry.isShuttingDown flag, while ClaudeLoginRunnerTests concurrently call TTYCommandRunner.run. When the fence wins, beginLaunch() rejects the Claude fixture launch and all three login tests receive launchFailed/empty output instead of their expected auth link or exit result.

The exact concurrent reproducer was:

swift test --filter 'ClaudeLoginRunnerTests|TTYCommandRunnerEnvTests'

It failed before the fix and now passes all 31 tests. The isolated registry state is task-local; the one test that crosses to a detached Thread explicitly captures its scoped state.

The Linux gating failure is a neighboring isolation class: #2311 serialized only the parameterized logged-out test's argument cases, not that test against the rest of PlatformGatingTests. Suite-level serialization closes that gap, while task-local auth timeout control keeps slow arm64 fake-CLI fixtures inside a test-only budget.

Verification

• swift test --filter ClaudeLoginRunnerTests
• swift test --filter 'ClaudeLoginRunnerTests|TTYCommandRunnerEnvTests' — 31 tests passed
• arm64 Docker PlatformGatingTests — 10 consecutive focused runs passed
• swift build -c release — passed in 218.64s
• make check — SwiftFormat clean; SwiftLint 0 violations in 1,577 files
• three consecutive make test runs on the final debug implementation:
  • 60/60 first-pass groups, 719 selections, 0 retries/timeouts, 692.8s
  • 60/60 first-pass groups, 719 selections, 0 retries/timeouts, 575.6s
  • 60/60 first-pass groups, 719 selections, 0 retries/timeouts, 477.9s
• autoreview: no accepted/actionable findings

swift test -c release was also attempted after making the new seams configuration-independent. It compiled past this patch and then hit existing unrelated DEBUG-only OAuth test seams in ClaudeOAuthDelegatedRefreshLinuxTests; the production release build is green.

No changelog edit here; this is an internal test-infrastructure fix.

[clawsweeper]

Codex review: needs maintainer review before merge. Reviewed July 19, 2026, 12:30 AM ET / 04:30 UTC.

Summary
The PR refactors TTY active-process tracking to support task-local test isolation, serializes Linux platform-gating tests, and widens test fixture timeouts without changing production defaults.

Reproducibility: yes. at source level: the PR identifies a concrete concurrent path where shutdown-fence tests mutate shared registry state while Claude login fixtures launch processes. The provided focused command is a plausible reproducer, but it was not rerun during this read-only review.

Review metrics: 2 noteworthy metrics.

• Patch surface: 8 files affected; 266 additions, 149 deletions. Most of the patch is a contained test-state refactor plus focused fixture and suite changes, rather than a production feature change.
• Validation claimed: 3 full-suite runs; 31-test focused reproducer; 10 arm64 focused runs. The PR provides unusually concrete flake-validation evidence, pending confirmation from the exact-head CI matrix.

Merge readiness
Overall: 🐚 platinum hermit
Proof: 🌊 off-meta tidepool
Patch quality: 🐚 platinum hermit
Result: ready for maintainer review.

Overall follows the weaker of proof and patch quality, so missing proof can cap an otherwise strong patch.

Rank-up moves:

• Allow the in-progress macOS and Linux checks to complete on the current head before merging.

Risk before merge

• [P1] The exact branch still has in-progress macOS and Linux CI jobs, so the intended reduction in load-sensitive failures is not independently confirmed by the current check matrix yet.

Maintainer options:

1. Decide the mitigation before merge
   Retain the task-local registry isolation as the root-cause fix, keep Linux serialization limited to the flaky test suite, and merge after the exact-head macOS/Linux matrix confirms the reported focused and full-suite behavior.
2. Pause or close
   Do not merge this PR until maintainers decide whether the risk is worth taking.

Next step before merge

• [P2] No discrete automated repair is identified; this maintainer-authored patch needs normal review and completion of the existing CI matrix.

Security
Cleared: The diff changes internal process-test state and test fixtures only; no dependency, workflow, secret, packaging, or privilege boundary change is evident.

Review details

Best possible solution:

Retain the task-local registry isolation as the root-cause fix, keep Linux serialization limited to the flaky test suite, and merge after the exact-head macOS/Linux matrix confirms the reported focused and full-suite behavior.

Do we have a high-confidence way to reproduce the issue?

Yes, at source level: the PR identifies a concrete concurrent path where shutdown-fence tests mutate shared registry state while Claude login fixtures launch processes. The provided focused command is a plausible reproducer, but it was not rerun during this read-only review.

Is this the best way to solve the issue?

Yes. Task-local test state addresses the shared mutable-state cause more directly than broad serialization, while suite serialization remains limited to the separately flaky Linux gating fixture group.

AGENTS.md: found and applied where relevant.

Codex review notes: model internal, reasoning high; reviewed against 242469b5b23e.

Label changes

Label changes:

• add P2: This is a bounded reliability improvement for flaky test infrastructure, with no stated production behavior change.
• add rating: 🐚 platinum hermit: Overall readiness is 🐚 platinum hermit; proof is 🌊 off-meta tidepool and patch quality is 🐚 platinum hermit.
• add status: 👀 ready for maintainer look: ClawSweeper has no concrete contributor-facing blocker left for this PR. Not applicable: This is maintainer-authored test-infrastructure work, so the external contributor proof gate does not apply; the PR body nevertheless supplies focused and full-suite after-fix validation results.

Label justifications:

• P2: This is a bounded reliability improvement for flaky test infrastructure, with no stated production behavior change.
• rating: 🐚 platinum hermit: Overall readiness is 🐚 platinum hermit; proof is 🌊 off-meta tidepool and patch quality is 🐚 platinum hermit.
• status: 👀 ready for maintainer look: ClawSweeper has no concrete contributor-facing blocker left for this PR. Not applicable: This is maintainer-authored test-infrastructure work, so the external contributor proof gate does not apply; the PR body nevertheless supplies focused and full-suite after-fix validation results.

Evidence reviewed

What I checked:

• Task-local registry isolation: The PR replaces the mutable process-global registry internals with a State object and adds a testing-only task-local override, so shutdown-fence tests can use isolated state rather than reject unrelated Claude fixture launches. (Sources/CodexBarCore/Host/PTY/TTYCommandRunner.swift, a3cac151f9c0)
• Scoped auth-status deadline override: The auth-status timeout override is task-local and only used by test fixtures; the normal isLoggedIn timeout remains the production fallback when no override is scoped. (Sources/CodexBarCore/Providers/Claude/ClaudeCLIAuthStatusProbe.swift, a3cac151f9c0)
• Related merged containment: Merged PR Serialize flaky Claude CLI platform-gating cases #2311 serialized only the parameterized platform-gating test cases. This PR moves serialization to the Linux suite and addresses the separate shared TTY registry race described in its reproduction. (TestsLinux/PlatformGatingTests.swift, ad7ee2f807db)
• Provided real-run validation: The PR body reports the focused concurrent reproducer passing all 31 tests, ten arm64 Docker platform-gating runs, a release build, make check, and three clean full-suite runs; the exact GitHub CI matrix is still in progress. (a3cac151f9c0)

Likely related people:

• steipete: The repository owner authored the current focused test-infrastructure change and is the clearest routing point for its intended concurrency boundary. (role: likely follow-up owner; confidence: medium; commits: a3cac151f9c0; files: Sources/CodexBarCore/Host/PTY/TTYCommandRunner.swift, Sources/CodexBarCore/Providers/Claude/ClaudeCLIAuthStatusProbe.swift)
• Chipagosfinest: Authored the merged Linux platform-gating containment change that this PR explicitly completes at suite scope. (role: recent adjacent contributor; confidence: high; commits: ad7ee2f807db; files: TestsLinux/PlatformGatingTests.swift)

What the crustacean ranks mean

• 🦀 challenger crab: rare, exceptional readiness with strong proof, clean implementation, and convincing validation.
• 🦞 diamond lobster: very strong readiness with only minor maintainer review expected.
• 🐚 platinum hermit: good normal PR, likely mergeable with ordinary maintainer review.
• 🦐 gold shrimp: useful signal, but proof or patch confidence is still limited.
• 🦪 silver shellfish: thin signal; proof, validation, or implementation needs work.
• 🧂 unranked krab: not merge-ready because proof is missing/unusable or there are serious correctness or safety concerns.
• 🌊 off-meta tidepool: rating does not apply to this item.

Shiny media proof means a screenshot, video, or linked artifact directly shows the changed behavior. Runtime, network, CSP, and security claims still need visible diagnostics.

How this review workflow works

• ClawSweeper keeps one durable marker-backed review comment per issue or PR.
• Re-runs edit this comment so the latest verdict, findings, and automation markers stay together instead of adding duplicate bot comments.
• A fresh review can be triggered by eligible @clawsweeper re-review comments, exact-item GitHub events, scheduled/background review runs, or manual workflow dispatch.
• PR/issue authors and users with repository write access can comment @clawsweeper re-review or @clawsweeper re-run on an open PR or issue to request a fresh review only.
• Maintainers can also comment @clawsweeper review to request a fresh review only.
• Fresh-review commands do not start repair, autofix, rebase, CI repair, or automerge.
• Maintainer-only repair and merge flows require explicit commands such as @clawsweeper autofix, @clawsweeper automerge, @clawsweeper fix ci, or @clawsweeper address review.
• Maintainers can comment @clawsweeper explain to ask for more context, or @clawsweeper stop to stop active automation.

[steipete]

Merged as 663099e8f9fb2680dabeaa811e844dd6090ea9aa after exact-head CI completed green.

Root cause: TTYCommandRunnerActiveProcessRegistry.isShuttingDown was process-global. The shutdown-fence cases in TTYCommandRunnerEnvTests set that flag while ClaudeLoginRunnerTests concurrently entered TTYCommandRunner.run; beginLaunch() then rejected the unrelated Claude fixture launch, producing launchFailed/empty output and a missing auth link. Suite-level .serialized never isolated those two separate suites.

Fix: the active-process registry now resolves through a task-local isolated State in tests, while its default remains the shared production state. The detached-thread shutdown case explicitly captures its scoped registry state because task locals do not propagate to a raw Thread. PlatformGatingTests is serialized at suite scope, completing #2311's partial parameter-case serialization. Claude auth-status timing uses a task-local test override; the remaining timing changes are fixture-only, including the named Kiro deadline assertion. Production defaults are unchanged.

Verification:

• swift test --filter ClaudeLoginRunnerTests passed.
• swift test --filter 'ClaudeLoginRunnerTests|TTYCommandRunnerEnvTests': 31 tests passed concurrently. This command failed before the registry isolation and passed after it.
• arm64 Docker focused PlatformGatingTests: 10 consecutive runs passed.
• swift build -c release: passed in 218.64s.
• make check: SwiftFormat clean; SwiftLint reported 0 violations in 1,577 files.
• Three consecutive make test runs on the final debug implementation each passed 60/60 groups and all 719 selections with 0 first-pass failures, 0 retries, and 0 timeouts: 692.8s, 575.6s, and 477.9s.
• Final local autoreview and branch autoreview both reported no accepted/actionable findings.
• Exact-head CI run 29673366301 on a3cac151f9c05b9a33822e9116a7363b6b63dd36: 9 passed, 0 pending, 0 failed, 0 cancelled. This includes build-linux-cli (linux-arm64, ubuntu-24.04-arm), Linux x64, Linux musl, both macOS test shards, lint, and security checks.

swift test -c release was additionally attempted. It compiled past this patch's configuration-independent test seams, then stopped on pre-existing unrelated DEBUG-only OAuth seams in ClaudeOAuthDelegatedRefreshLinuxTests; the production release build above is green.