test(gap): de-flake test_gap_fs_fd_2749 — it raced the libuv threadpool

[proggeramlug]

test_gap_fs_fd_2749 is flaky, and it has been silently red-lighting unrelated PRs. It took out #6397, #6398 and #6399 today — each of which had it as their only untriaged failure — and it cost a full bisect before the oracle itself became the suspect.

The test races the libuv threadpool

The three callback-form fd mutators are dispatched concurrently:

fs.ftruncate(fd, 1, (err) => cbCode("ftruncate closed cb", err));
fs.futimes(fd, 1, 2, (err) => cbCode("futimes closed cb", err));
fs.fchown(fd, 0, 0, (err) => cbCode("fchown closed cb", err));

Those land on the libuv threadpool and their callbacks fire in completion order, which is not deterministic across three different syscalls.

The gap harness compares perry's output against a freshly run node oracle — and the oracle itself is unstable. Measured directly:

	distinct outputs
node --experimental-strip-types, 40 runs	4
perry, 40 runs	1

Node emits fchown closed cb before futimes closed cb on roughly one run in six:

   ftruncate closed cb: EBADF ftruncate
-  futimes closed cb: EBADF futime      <- ~84% of runs
-  fchown closed cb: EBADF fchown
+  fchown closed cb: EBADF fchown       <- ~16% of runs
+  futimes closed cb: EBADF futime

Perry is deterministic (always submission order), so it disagrees with the oracle whenever the oracle takes the minority path. That is a coin flip on every PR, and it is why the failure looked like a regression: it correlated with nothing.

The fix

The test is about the error surface (err.code / err.syscall) of each fd mutator, not about libuv's scheduling. Buffer the three callback results and flush them in a fixed order once all three have landed.

Every assertion is preserved, and the emitted 10 lines are byte-for-byte what they were — only the print order is pinned.

Verification

	before	after
node, 60 runs	4 distinct	1 distinct
perry, 40 runs	1 distinct	1 distinct
perry vs node	flaky	byte-identical
expected output	—	unchanged (same 10 lines)

Deliberately left alone

While restructuring I briefly sequenced the promisified forms by calling main() from the last callback, and that surfaced a separate, real perry bug: awaiting a promisify(fs.ftruncate) rejection from inside an fs callback yields undefined undefined instead of EBADF ftruncate. That is not this PR's business — filing it separately with a repro rather than folding an unrelated runtime bug into a test de-flake. main() stays at top level here, exactly as before.

Summary by CodeRabbit

• Tests
  • Stabilized callback error-reporting test output by enforcing a consistent result order.
  • Ensured callback results are collected before being displayed, reducing variability across runs.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: 8b2447cf-5db0-4b72-8ae3-4ced7fbebf87

📥 Commits

Reviewing files that changed from the base of the PR and between 5a233c5 and e36420e.

📒 Files selected for processing (1)

• test-files/test_gap_fs_fd_2749.ts

---

📝 Walkthrough

Walkthrough

The callback-form test now buffers all three fd-mutator results and prints them in a fixed order after every callback has completed, removing output variability caused by callback completion order.

Changes

Callback output determinism

Layer / File(s)	Summary
Buffer and order callback results test-files/test_gap_fs_fd_2749.ts	Defines CB_ORDER, stores callback results in cbLines, waits for all callbacks, and prints results according to the fixed order.

Estimated code review effort: 2 (Simple) | ~10 minutes

Possibly related issues

• ci: test_gap_fs_fd_2749 is flaky on CI (passes 5/5 locally) — the last untriaged gap failure, randomly reds the required conformance gate #6396: The deterministic callback output addresses the test flakiness and CI timing sensitivity described by this issue.

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly states the test de-flake and matches the change to stabilize test_gap_fs_fd_2749.
Description check	✅ Passed	The description covers the summary, fix, and verification, though it omits the template's related issue and checklist sections.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

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.}