[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

Without the C _abc module, as on GraalPy, abc falls back to _py_abc. That rebinds _abc_negative_cache and _abc_negative_cache_version on a class during isinstance. SentinelMeta.__setattr__ refused every write, so isinstance(1, NullType) raised TypeError, which broke JSON dumping and extraction there (#1960).

It now lets exactly those two names through, and only where the type already holds them, which only _py_abc arranges. Every other write, every delete, and register on a concrete type is still refused. The docstring and the sentinel-convention page note the exception.

tests/corekit/test_sentinel_py_abc_1960_unit.py runs the checks in a child interpreter that really takes the _py_abc fallback (abc.ABCMeta is _py_abc.ABCMeta). It fails without the fix on 3.14 and 3.8.

Closes #1960

[JarryShaw]

GOOD TO GO at 3e433b8a8. A Sonnet review read _py_abc for both 3.8 and 3.14. After __new__, its only class-attribute writes are the two allowed cache names in __subclasscheck__. The invalidation counter lives on _py_abc.ABCMeta, never on a sentinel type. Deliberately poisoning the negative cache could produce a wrong "no" but never a wrong "yes". _abc_cache, _abc_registry and all other writes stay refused, and on C _abc builds both names are still refused. The test's fallback is real (abc.ABCMeta is _py_abc.ABCMeta). It fails without the fix (17 failed), and tests/corekit, the tier guard and the doc claims pass.

[JarryShaw]

NEEDS CHANGES at 3e433b8a8. Project tests fails test_docs_unresolvable_roles_1919_unit.py::DeadTargetTests::test_no_dead_target_appears. The new _PY_ABC_CACHE comment writes :meth:~abc.ABCMeta.new``, which is not in the Python inventory. A follow-up commit writes it as a literal. The review missed it because it never ran the #1919 guard.

[JarryShaw]

GOOD TO GO at a5cc70842. A Sonnet delta review confirmed the new commit only turns the ABCMeta.__new__ and abc roles in sentinels.py and sentinel-convention.rst into literals, with no code change. No other role in the PR's prose points at an unresolvable target. The #1919 and #1936 guards, the new #1960 test, the sentinel exports test and the doc-claims test pass.

[github-actions]

Coverage: 90.07% (unit tier, Python 3.15, a5cc70842, Unit Tests run success)

Package	Statements	Missed	Branches	Partial	Cover
pcapkit (top level)	104	2	20	2	96.77%
pcapkit/const	19597	881	2618	867	92.11%
pcapkit/corekit	2505	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.