test: deflake test-runner-coverage

[christianaurichzm]

test-runner-coverage is marked flaky on Windows, Linux s390x and AIX. Every recorded failure is the same assertion:

AssertionError [ERR_ASSERTION]: The expression evaluated to a falsy value:

  assert(!findCoverageFileForPid(result.pid))

Cause

The two subtests that pass NODE_V8_COVERAGE: tmpdir.path leave a coverage-<pid>-<ts>-<n>.json file behind, and tmpdir.refresh() only runs once, at the top of the file. The nine later checks spawn without NODE_V8_COVERAGE, so their coverage goes to the private mkdtemp directory that setupCoverage() creates and cleanup() removes, never to the test tmpdir. findCoverageFileForPid() keys only on the pid, so a match can only come from that stale file, once the OS hands the same pid to a later child.

Reproduction

Forcing pid reuse inside a pid namespace reproduces it on Linux x64. Pushing the cursor past RESERVED_PIDS makes the kernel allocate cyclically within [300, pid_max), a window this test wraps through several times:

unshare -rpf --mount-proc bash -ec '
  echo 350 > /proc/sys/kernel/pid_max
  cat /proc/sys/kernel/pid_max          # verify the requested value took effect
  for i in $(seq 1 320); do /bin/true; done
  out/Release/node test/parallel/test-runner-coverage.js'

	result
without the patch	40/40 runs failed
with the patch	0/40 runs failed

tools/test.py --repeat=24 -J parallel/test-runner-coverage passes on a normal Linux x64 run either way.

Fix

Refresh the tmpdir from an after() hook on each subtest that dumps into it. The hook rather than a trailing call matters: if an assertion in one of those subtests fails, a trailing cleanup is skipped and the stale file turns every later check into a second, misleading failure. Same forced pid reuse, with the failure injected into the last dumping subtest:

	cascading failures
cleanup at end of subtest	20/20 runs
cleanup in after()	0/20 runs

This also drops the PASS, FLAKY entries added in #65562 as a stopgap "while the stale coverage-file behavior is investigated"; with the stale file gone that failure path is removed, and keeping the entries would hide a regression
on those platforms.

Needs a full CI run to reach the affected configurations. The GitHub Actions Windows coverage job sets NODE_V8_COVERAGE for the test step, so it does not exercise this path.

Fixes: #55154

[nodejs-github-bot]

Review requested:

• @nodejs/test_runner

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 90.11%. Comparing base (f9ab994) to head (4d8c8c3).
⚠️ Report is 10 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main   #65728      +/-   ##
==========================================
+ Coverage   90.07%   90.11%   +0.03%     
==========================================
  Files         754      754              
  Lines      256395   256395              
  Branches    48494    48523      +29     
==========================================
+ Hits       230947   231044      +97     
+ Misses      16563    16479      -84     
+ Partials     8885     8872      -13     

see 32 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[nodejs-github-bot]

CI: https://ci.nodejs.org/job/node-test-pull-request/76872/

[trivikr]

Stress tests:

• main: https://ci.nodejs.org/job/node-stress-single-test/883/
• pr: https://ci.nodejs.org/job/node-stress-single-test/884/