test(tiers): keep the xdist probe out of recursive collection (#1756)

[JarryShaw]

Please follow the guide below

• You will be asked some questions, please read them carefully and answer honestly

• Put an x into all the boxes [ ] relevant to your pull request (like that [x])

• Use Preview tab to see how your pull request will actually look like

• 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. Narrow runs only, listed below.

• Added a changelog entry under docs/source/changelog/ and regenerated CHANGELOG.md, if the change is user-visible: 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?

Tick the commit type your subject line carries.

• fix — corrects a defect
• feat — adds a feature
• perf — changes performance, not behaviour
• refactor — changes neither behaviour nor performance
• test — tests only
• docs — documentation only
• ci — workflows or build tooling
• release — bumps the version or rolls up a distribution
• chore — anything else

---

Description

Closes #1756.

On #1755's GraalPy run, a crashed worker turned the whole session into pytest exit 2. The mechanism is now measured, and it corrects the issue's guess:

• Nothing is aggregated across workers. XdistSubprocessTests.setUp writes the probe tests/protocols/test_zzz_xdist_guard_probe_unit.py and removes it only through addCleanup.
• When gw10 died mid-test, the probe stayed on disk. xdist's replacement worker collected tests/ again, and the collection-time audit in tests/conftest.py found the probe's read and set session.shouldfail.
• The controller then raised Interrupted once the other workers drained.

Fix: tests/conftest.py adds collect_ignore = [f'protocols/{PROBE_MODULE_NAME}']. pytest consults collect_ignore only while it recurses into a directory, never for a path named on the command line. So:

• no recursive session, including a replacement worker, collects a stale probe;
• the probe's own nested run, which names the probe's path, still audits it;
• real unit-tier reads are guarded as before, and SuiteIsCleanTests still scans the path on disk.

Test: ProbeWorkerCrashTests, in tests/test_tier_guard_xdist.py.

• It runs the real probe test in a throwaway pytest -n 2 session, where the worker calls os._exit once the probe is on disk.
• It asserts that the crash happened, that the probe was left behind, and that there is no Interrupted and the exit is 1.
• On main's conftest.py it fails, with the CI-identical message. Without xdist it skips, like its neighbours.

Tests: the two tier-guard modules plus test_unit_tests_queue.py give 152 passed, 3 skipped (the xdist classes skip in the repo venv). In an xdist venv, both tier-guard modules under -n 2 give 117 passed.

[JarryShaw]

Verdict: GOOD TO GO at eb3253d34. A Sonnet cross-review checked it (author: Opus).

• Mechanism, re-derived and reproduced without a crash: a hand-placed stale probe plus pytest -n 2 --dist load tests/protocols on main's conftest.py exits 2 with the CI-identical Interrupted line. The CI log fits: gw10's nested probe run is in the monitor at 04:04:48, node down follows at 04:04:50, and the interrupt comes when the session drains at 04:08:20.
• Reproduced independently with SIGKILL on the worker: on main, exit 2 with Interrupted; with the fix, exit 1 and only the crashed test fails.
• The guard still bites: a real unit-tier module reading a generated capture still errors serially (exit 4) and still interrupts under -n 2 (exit 2).
• The new test fails on main for the intended reason ('Interrupted' unexpectedly found), after its crash and stale-probe assertions pass.
• Gates pass, including test_tier_guard_xdist.py at -n 3, repeated six times with no flake.
• Non-blocking: XdistSubprocessTests.setUp checks probe_path.exists() before taking the lock (:176-179 versus :189). That pre-existing race is filed as test(tiers): XdistSubprocessTests checks for a leftover probe before taking the lock #1758.
• My own check: both tier-guard modules plus test_unit_tests_queue.py give 152 passed and 3 skipped, with pcapkit.__file__ printed inside the PR's worktree.

[github-actions]

Coverage: 89.59% (unit tier, Python 3.15, eb3253d34, Unit Tests run success)

Package	Statements	Missed	Branches	Partial	Cover
pcapkit (top level)	104	3	20	3	95.16%
pcapkit/const	18862	954	2368	868	91.06%
pcapkit/corekit	2461	69	780	35	96.54%
pcapkit/dumpkit	276	3	100	3	98.40%
pcapkit/foundation	2915	164	1016	34	93.59%
pcapkit/interface	138	7	52	5	93.68%
pcapkit/protocols	17184	224	4574	189	98.08%
pcapkit/toolkit	872	177	298	8	77.18%
pcapkit/utilities	512	2	158	4	99.10%
pcapkit/vendor	4417	2342	1010	157	43.38%

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