test(protocols): keep the #1889 spin's first raise inside its try (#1897)

[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

The fix is in the test harness. examples/generators/options.py behaves correctly in both cases below.

• The CI failure, on CPython 3.13: the backward jump closing _Spin's inline loop falls outside its except TimeoutError range in the exception table. A SIGALRM handled at that jump went straight past the swallow, so the case was still TIMEOUT but swallowed stayed 0. 3.12's table has the same gap, but an alarm handled at that jump there resumes inside the try, so only 3.13 fails naturally; 3.14 and 3.15 cover the jump. The spin now waits in _Spin.wait, called from inside the try.

• A stall before the try: the deadline armed on entry, so a one-second stall before the spin delivered the raise outside it. spin_roundtrip now holds back the deadline's own first signal.alarm until wait arms it. Re-arms and the final cancel still reach the real alarm.

• New tests: one raises at the spin's first backward jump through sys.monitoring (CPython 3.12+), and one stalls before the spin. Each fails with its part of the fix reverted.

• Loops of the target test, 200 runs each:

  Interpreter	Before, idle	Before, under load	After
  3.13.16	19 failed	26 failed	0 failed
  3.14	0 failed	0 failed	0 failed

  tests/project, tests/test_support_helpers.py and the tier guard pass.

Closes #1897

[JarryShaw]

Verdict: NEEDS CHANGES at 61230f43c. The fix holds; one wording change is required. Three places overstate the 3.12 mechanism: the _Spin.wait docstring (:151), the monitoring test's docstring (:351), and this PR's body. The 3.12 exception table does leave the loop-closing jump outside the try. But a naturally handled alarm there is attributed to the jump target, which is inside. In 175 natural first alarms on the old 3.12 spin, 25 were handled at that jump and all 25 were still swallowed. Only a raise injected through sys.monitoring lands on the jump itself.

Everything else the Sonnet cross-review checked holds:

• 3.13 cause: on the old spin, all 9 alarms handled at offset 292, the closing jump, left swallowed at 0. The other 31 were handled at the call and swallowed.

• Real re-arm: removing options.py's re-arm signal.alarm(1) fails the swallow test, so the re-arm is still what the test proves.

• Settle paths: dropping either settle path fails a test.

• New tests: each fails when its part of the fix is reverted. The monitoring test skips cleanly on 3.8–3.11.

• Loop counts:

  Tree	Python	Idle	Loaded
  Old	3.13	8 of 50 failed	8 of 50 failed
  New	3.12–3.15	0 failed	0 failed

• No leaks: after every test, the alarm, the handler and the sys.monitoring tool ids come back as they were.

[JarryShaw]

Verdict: GOOD TO GO at a22a43c56. The amend changes only docstrings. With docstrings stripped, the file's AST matches the reviewed head.

A Sonnet re-check found the new wording accurate in the docstrings, the PR body and the commit message:

• 3.12's table has the gap, but an alarm handled at the jump resumes inside the try.
• Only 3.13 fails naturally.
• The monitoring test exercises the gap on 3.12 and 3.13, and is a guard that cannot fail from 3.14.

The re-check confirmed the jump offsets with dis on 3.12.15, 3.13.16 and 3.14.8. The functional findings from the review of 61230f43c still hold.

This PR's Python 3.13 CI leg gates readiness: the re-check's own ad hoc run of the monitoring test on 3.13 hung in a scratch copy and was stopped. Two nits, neither blocking:

• the _Spin.wait docstring does not say the gap is version-specific;
• "each fails with its part of the fix reverted" holds for the monitoring test on 3.12 and 3.13 only.

[github-actions]

Coverage: 90.07% (unit tier, Python 3.15, a22a43c56, Unit Tests run success)

Package	Statements	Missed	Branches	Partial	Cover
pcapkit (top level)	104	2	20	2	96.77%
pcapkit/const	19595	881	2618	867	92.11%
pcapkit/corekit	2504	70	792	36	96.54%
pcapkit/dumpkit	276	3	100	3	98.40%
pcapkit/foundation	2916	164	1016	34	93.59%
pcapkit/interface	138	7	52	5	93.68%
pcapkit/protocols	17411	224	4580	189	98.10%
pcapkit/toolkit	872	177	298	8	77.18%
pcapkit/utilities	522	2	158	4	99.12%
pcapkit/vendor	4431	2341	1018	156	43.64%

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