[JarryShaw]

• Searched for similar pull requests
• Followed the coding style (make pylint, make mypy, make isort)
• make test passes, and a test case covers the change
• Changelog entry: N/A — centralised in docs(changelog): shared 1.5.0 changelog — long-lived, merges last (#610, #616, #617, #618, #620) #657

What is the purpose of your pull request?

• ci — workflows or build tooling

---

Description of your pull request and other information

Established: each missing pid is a real xdist worker (4 controller children, cpu_count 4), and it died after reporting its results, which is why every test passed. xdist ignores a worker's EOF after workerfinished. No log has a traceback. A process with no data still writes a file, so nothing ran its atexit save to completion. Not established: what killed it. execnet SIGKILLs after 10 s, but a measured save takes 0.24 s, so a crash during finalisation is the other candidate. Nothing in the #1409–#1412 diff spawns or kills a process.

Ruled out: runner image/Python (hits on both 20260927/3.14.7 and 20261004/3.14.8); test count (first hit ran 3195, the same as a clean run); chdir (data path is absolute); a test touching coverage (none does).

Changes:

1. tests/conftest.py: pytest_sessionfinish saves a worker's coverage before shutdown.
2. .github/coverage.toml: sigterm = true.
3. unit-tests.yml: the guard prints each missing pid's cmd and atexit fate, and coverage-processes.log is uploaded as an artifact. Doc citations are renumbered.

tests/project: 404 passed, 1 skipped. The 5 new tests fail without the fix.

Closes #1423

[github-actions]

Coverage: 89.43% (unit tier, Python 3.14, 7eef5df56, Unit Tests run success)

Package	Statements	Missed	Branches	Partial	Cover
pcapkit (top level)	104	4	20	4	93.55%
pcapkit/const	18791	968	2352	869	90.90%
pcapkit/corekit	2038	89	634	25	94.76%
pcapkit/dumpkit	234	2	84	2	98.74%
pcapkit/foundation	2637	98	928	48	94.78%
pcapkit/interface	112	7	40	5	92.11%
pcapkit/protocols	16504	219	4420	183	98.04%
pcapkit/toolkit	539	74	168	3	84.58%
pcapkit/utilities	429	4	122	4	98.55%
pcapkit/vendor	4409	2342	1006	157	43.25%

Per-file detail: the coverage-html artifact of this run.

[JarryShaw]

Verdict: GOOD TO GO at 7eef5df56 (Sonnet cross-review; the author ran on Opus).

• Real xdist run: coverage 7.16.2, pytest-xdist 3.8.0, in a throwaway venv. Each pid writes one .coverage.*, with no DataError. Arcs executed after the session-finish save (in pytest_unconfigure) still land in the worker's file.
• Reproduction: SIGKILLing a worker at pytest_unconfigure leaves its file in place and the guard reports missing []. On the base conftest.py the same kill gives missing ['24103'], which is ci(unit-tests): Python 3.14 coverage guard intermittently finds an xdist worker with no data file #1423's failure.
• No-op elsewhere: plain pytest and pytest -n 2 without coverage leave no stray files and raise no errors.
• sigterm = true: it only installs a save-and-re-raise handler while coverage runs, and nothing in the repo sends SIGTERM.
• Workflow: the existing guard strings are unchanged, and the new upload step mirrors its siblings (upload-artifact@v7, !cancelled() && matrix.coverage).
• Docs: all 11 renumbered citations land on the content their prose describes, and none are left stale.
• Tests: tests/project plus the tier-guard and support-helper modules give 544 passed, 2 skipped.
• Note, not blocking: combine dedupes by the content hash embedded in the filename (coverage/data.py, hash_for_data_file). After this hook's early save, a later append keeps the old name, so two workers with byte-identical data at session end would dedupe even if their later arcs differ. That realistically needs two workers that ran no tests. CI's current Combined 30 files, skipped 36 is true duplicates, so there is no under-count today.