[JarryShaw]

Part of #1052: --verbose from A, plus C, E, G and H. B, D and F follow in sibling PRs.

• 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?

• 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
• chore — anything else

---

Description of your pull request and other information

• E: engine-tests is now one job per engine. Each job loops over 3.10–3.15, one venv per Python version, and every cell gets its own log group, annotations and summary row. 3.15 runs as a separate step with continue-on-error. No required check names an engine job.
• G: a new changes job classifies the PR diff. On a docs-only PR the four test legs are skipped and project-tests runs instead. required-checks accepts skipped only in that case.
• H/A/C: unittest-ordering no longer runs on PRs. Its run step gets --verbose and a 20-minute step cap. The apt steps are capped at 5 minutes, with one bounded retry; the slowest of 120 observed runs took 34s.
• tests/project/test_unit_tests_queue.py (30 tests) runs the gate shell, the classifier (against a scratch merge commit), the cell loops and the apt retry, and checks that project-tests runs every test that reads docs/ or Markdown.

Unit Tests jobs per code PR push: 62 listed / 61 run before, 24 listed / 21 run after (docs-only, by construction: 4). On b00096a59, each engine job took 2.2–3.8 min, all 35 checks passed, and the Python 3.14 leg needed one re-run: the first attempt was killed by a runner shutdown at 41 min.

[JarryShaw]

Cross-review verdict on 442ebcd4f: NEEDS CHANGES (ran on Sonnet; author Opus)

1. A docs-only PR can break a test that never runs. tests/corekit/test_sentinel_exports_unit.py:210 reads docs/source/contributing/conventions/sentinel-convention.rst. Change NULL to NULLX there and that module fails, yet the project leg that project-tests runs still passes 530 of 530. The PR would classify as docs-only, skip test, and pass required-checks. Fix: run every test that reads docs/ on docs-only PRs, or count docs/source/contributing/conventions/ as code, and correct the stated justification.
2. Paths with spaces are misclassified. The classifier splits paths on whitespace (unit-tests.yml:1012, diff.stdout.split()), so a root-level file named x.md docs/y reads as docs. Use git diff -z and split on NUL.
3. 8 of 19 mutations survive test_unit_tests_queue.py. The survivors: the OWN_PR/workflow check, the HEAD^1 base, the empty-diff-is-code fallback, || true on the engine cell, the failed[] exit, the apt retry count, the apt step cap, and the whitespace split.

Confirmed:

• All 36 old engine cells match the 36 new ones on pass/skip counts (main run 37407908831 against PR run 37410433253).
• A failing cell fails the step. The 3.15 continue-on-error applies to that step only.
• All four workflow_call callers pass gate-only: true.
• The required contexts are unchanged.
• No new ${{ }} appears in run:.
• Every docs line reference matches.
• Merging with ci(coverage): report test coverage from a workflow (#1063) #1064: unit-tests.yml merges cleanly; workflows.rst (6 hunks) and releasing.rst (2) conflict on line citations.

[JarryShaw]

Cross-review verdict on 1f075fb94: NEEDS CHANGES (ran on Sonnet; author Opus)

• Blocking (mechanical): the new project-tests step adds 10 lines, and the docs citations below it were not renumbered. Each is 10 lines off:

  Location	Cites	Should be
  releasing.rst:49	unit-tests.yml:1065-1134	1075-1144 (gate)
  workflows.rst:347 and :365	1238	1248
  workflows.rst:370	1137	1147
  workflows.rst:374	1136-1157	1146-1167
  workflows.rst:384	1237-1291	1247-1301
  workflows.rst:390	1240	1250

  I confirmed it myself: gate sits at 1075 and required-checks at 1247.

• All three earlier points are fixed:

  • The corekit docs reader now runs on docs-only PRs. A mutated sentinel-convention.rst exits 1 there.
  • The classifier splits on NUL, so x.md docs/y now reads as code.
  • 30 executing tests now catch all 8 earlier surviving mutations.

• No missed docs reader found across tests/, pcapkit/ and util/.

• Non-blocking mutation survivors: continue-on-error on either project-tests step, the dpkg --configure -a removal, and required-checks dropping one leg.

[github-actions]

Coverage: 88.74% (unit tier, Python 3.14, c7a047efc, Unit Tests run success)

Package	Statements	Missed	Branches	Partial	Cover
pcapkit (top level)	104	4	20	4	93.55%
pcapkit/const	18757	1035	2342	846	90.23%
pcapkit/corekit	1874	91	578	22	94.33%
pcapkit/dumpkit	136	0	40	0	100.00%
pcapkit/foundation	2422	143	842	37	92.52%
pcapkit/interface	112	7	40	5	92.11%
pcapkit/protocols	15653	187	3942	162	98.19%
pcapkit/toolkit	487	71	144	3	84.15%
pcapkit/utilities	429	4	122	4	98.55%
pcapkit/vendor	4409	2359	1006	158	42.84%

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

[JarryShaw]

Cross-review verdict on c7a047efc: GOOD TO GO (ran on Sonnet; author Opus). It is confirmed by this head's own CI: Required checks passed is green.

• Rebase onto 62c9878e6 is clean. The only changes beyond main's own are resolved conflicts and renumbered citations in releasing.rst and workflows.rst. The test file is byte-identical to the reviewed head. ci(coverage): report test coverage from a workflow (#1063) #1064's coverage leg, steps and docs section are intact.
• Fixed across rounds:
  • the corekit docs reader now runs on docs-only PRs;
  • the classifier splits on NUL;
  • 33 executing tests catch every mutation from the earlier rounds;
  • every unit-tests.yml:N citation is pinned by content.
• Engines: 36 engine jobs became 6 Engines (<engine>) jobs, and the pass/skip counts per cell match the old jobs.
• Live CI check: the first coverage-comment.yml comment posted on this PR (88.74%, from github-actions[bot]), so ci(coverage): report test coverage from a workflow (#1063) #1064's workflow_run path works.
• Non-blocking survivors:
  • an || true on the project leg step;
  • dropping dpkg --configure -a between apt attempts.