[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?

• fix — corrects a defect

---

Description of your pull request and other information

The remaining GraalPy-only failures from #1963, each reproduced on GraalPy 25.4.4 locally.

• Defect, fixed: _importer_stacklevel counted GraalPy's <frozen _frozen_importlib…> bootstrap frames, which its warnings.warn skips. An import-time warning was therefore attributed to <sys>:0. The new ImporterStacklevelTests cover both interpreters' frame names, and fail without the fix.
• Gates:
  • test_sentinel_py_abc_1960_unit blocks _abc only off GraalPy, because GraalPy's _py_abc imports from _abc.
  • test_compat compares only the octet and position of a UnicodeDecodeError on GraalPy, which words the error differently.
  • XdistSubprocessTests and ProbeWorkerCrashTests skip on GraalPy.
• No change needed:
  • test_extraction_no_eof and the large-declared-length field test already pass after fix(corekit): copy SeekableReader's octets and feed MultiDict a dict, and gate GraalPy's limits (#1963) #1974/fix(foundation): a record length above sys.maxsize crashes Frame.read with OverflowError on 32-bit-Py_ssize_t builds #1975.
  • The preexec_fn and SOCK_SEQPACKET modules are in the LATEST tier, so the GraalPy leg never runs them.

Not done:

• Sharding the leg in two. The matrix comes from util/python_versions.py, and a shard would rename the check.
• Junit timings put the curated tier at about 54% of the old test time. That projects to well under the leg's timeout, so re-measure after merge first.
• Found and filed separately: GraalPy never calls __del__, so Extractor.__del__ cannot close an abandoned extraction's file there.

Part of #1963.

[JarryShaw]

GOOD TO GO at f2944a2f7. A Sonnet review of 30a584e3e confirmed the stacklevel fix on GraalPy 25.4.4 and CPython 3.14.8:

• GraalPy's import stack is <frozen _frozen_importlib_external> then <frozen _frozen_importlib>, and its warnings.warn skips both.
• The new predicate never matches a CPython frame.
• test_config_unit passes on both (38 passed), and ImporterStacklevelTests fails with the fix reverted.
• Every gate keys on sys.implementation.name, and the tier guards pass.
• The CPython path still blocks _abc, and still compares the full decode message.

f2944a2f7 rewords one garbled docstring sentence and drops a count ("both rows").

The review also asked for :issue: in place of #1963 in the test docstrings and #: comments. I kept #1963, because tests/ is not rendered by Sphinx and plain #NNN is the prevailing form there.

[github-actions]

Coverage: 91.52% (unit tier, Python 3.15, f2944a2f7, Unit Tests run success)

Package	Statements	Missed	Branches	Partial	Cover
pcapkit (top level)	104	4	20	2	95.16%
pcapkit/const	19597	881	2618	867	92.11%
pcapkit/corekit	2540	71	808	37	96.54%
pcapkit/dumpkit	276	3	100	3	98.40%
pcapkit/foundation	2918	23	1016	25	98.78%
pcapkit/interface	138	4	52	3	96.32%
pcapkit/protocols	17450	226	4598	191	98.09%
pcapkit/toolkit	872	36	298	10	95.21%
pcapkit/utilities	523	2	158	4	99.12%
pcapkit/vendor	4431	1995	1018	167	51.73%

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