test: harden codex manager cli isolation - #321

What Changed

• hardened test/codex-manager-cli.test.ts isolation by resetting shared mocks and reapplying auth/browser/server stubs after vi.resetModules()
• pinned non-interactive stdin defaults and added a regression for the closed-stdin manual callback guard so worker reuse cannot persist login state incorrectly
• fixed test/wait-utils.test.ts fake-timer sleep mocking so virtual time advances instead of spinning into worker OOM

Risk Level

• low
• test-only changes; no runtime or product code changed

Rollback Plan

• revert PR #321

Validation

• node_modules\\.bin\\vitest.cmd run --pool=threads --maxWorkers=1 test/codex-manager-auth-commands.test.ts test/codex-manager-cli.test.ts
• node_modules\\.bin\\vitest.cmd run --pool=threads --maxWorkers=1 test/wait-utils.test.ts
• node_modules\\.bin\\vitest.cmd run --pool=threads --maxWorkers=1 test/codex-manager-cli.test.ts
• npm run lint
• npm run typecheck
• npm run build
• split-suite fallback passed all 221/221 files and 3165/3165 tests
• monolithic npm test -- --pool=threads --maxWorkers=1 still ends in a Vitest worker OOM near the end, but no functional test failures remained before the runner died

note: greptile review for oc-chatgpt-multi-auth. cite files like lib/foo.ts:123. confirm regression tests + windows concurrency/token redaction coverage.

Greptile Summary

this pr hardens test isolation in the codex manager cli suite and fixes an oom in the wait-utils test. all three problems it fixes are real and the solutions are sound.

• mock leak fix (codex-manager-cli.test.ts): six mocks (promptQuestionMock, planOcChatgptSyncMock, applyOcChatgptSyncMock, runNamedBackupExportMock, exportNamedBackupMock, detectOcChatgptMultiAuthTargetMock, normalizeAccountStorageMock) were not reset between tests; queued mockResolvedValueOnce values could bleed across cases. all are now reset in beforeEach.
• stdin descriptor fix (codex-manager-cli.test.ts): readableEnded and destroyed are now captured at module load, restored in afterEach, and pinned to false in beforeEach via setOpenStdinState(). this prevents a previous worker's closed-stdin state from short-circuiting manual-callback tests.
• oom fix (wait-utils.test.ts): the old sleep mock returned immediately without advancing fake time, so sleepWithCountdown's while (Date.now() < endTime) guard never moved forward. the fix injects await vi.advanceTimersByTimeAsync(ms) inside the mock, which advances fake Date.now() and lets the loop terminate correctly.
• new test: "skips manual callback prompting when stdin is already closed in non-tty mode" covers the readableEnded: true early-exit path through lib/codex-manager.ts:1252; the sibling destroyed: true branch (same || condition) has no coverage yet.

Confidence Score: 5/5

• safe to merge — all changes are test-only, fix documented leaks/oom, and add a new targeted assertion
• both changed files are test-only; the fixes address concrete, reproducible problems (leaking mocks, closed-stdin short-circuit, fake-timer oom). the only remaining gap is the destroyed: true branch, which is a follow-up p2 — not a blocker
• no files require special attention

Important Files Changed

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[beforeEach] --> B[vi.resetModules + all mockReset]
    B --> C[restoreTTYDescriptors]
    C --> D[setOpenStdinState\nreadableEnded=false, destroyed=false]
    D --> E[import auth/browser/server modules\nreset + default implementations]
    E --> F[test body runs]
    F --> G{stdin state?}
    G -- readableEnded=true --> H[skip promptQuestion\nskip browser open\nskip server start\nexit 0]
    G -- open --> I[normal manual OAuth flow]
    F --> J[afterEach: restoreTTYDescriptors]

This is a comment left during a code review.
Path: test/codex-manager-cli.test.ts
Line: 5420-5466

Comment:
**missing `destroyed: true` branch coverage**

the production code at `lib/codex-manager.ts:1252` gates on `input.readableEnded || input.destroyed`. this new test covers `readableEnded: true` (with `destroyed` left `false`), but there is no test for the `destroyed: true` path (with `readableEnded` left `false`). consider adding a sibling test that calls:

```ts
Object.defineProperty(process.stdin, "destroyed", {
  value: true,
  configurable: true,
});
```

and re-asserts the same "no prompt / no exchange" expectations. without it, a regression that drops the `destroyed` check would go undetected.

How can I resolve this? If you propose a fix, please make it concise.

_{Reviews (2): Last reviewed commit: "test: cover closed stdin manual callback..." | Re-trigger Greptile}