[JarryShaw]

Please follow the guide below

• 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
• Added a changelog entry under docs/source/changelog/ and regenerated CHANGELOG.md

What is the purpose of your pull request?

• fix — corrects a defect

---

Description of your pull request and other information

Closes #2433. Deferred still failed to pickle under protocol 0/1 on GraalPy 25.4 even after #2424's __getstate__/__setstate__ fix, with TypeError: typing.Generic() takes no arguments from copyreg._reduce_ex.

Root cause: for protocol < 2, copyreg._reduce_ex walks cls.__mro__ for a base to compute a base-object state from, before __getstate__ is ever consulted. On CPython, Generic is a heap type whose __new__ is just the inherited, differently-bound object.__new__, so the walk passes it and stops at object (state=None). On GraalPy, Generic is itself a non-heap (built-in) type, so the walk's first check (not base.__flags__ & _HEAPTYPE) breaks on Generic directly and calls Generic(self), which raises. __getstate__/__setstate__ never run in that failure, since the crash happens before the walk's result is combined with them.

Fix: Deferred.__reduce_ex__ hard-codes the base=object, state=None outcome the walk always reached on CPython for protocol 0/1, bypassing copyreg._reduce_ex's MRO walk (and Generic with it) entirely. It also checks for, and defers to, a subclass's own __reduce__ override before building that hard-coded tuple — a regression caught in cross-review: without the check, a Deferred subclass overriding __reduce__ had it silently dropped at protocol 0/1 while protocol 2+ (via object.__reduce_ex__, unchanged) kept honouring it. Protocol >= 2 is otherwise untouched. Dropping Generic[_KT] instead was considered and rejected: test_deferred_proto_type_2417_unit.py reads Deferred.__parameters__ at runtime, which only exists because Generic is a real base, not TYPE_CHECKING-only.

Confirmed on real GraalPy 25.4.4 by cross-review: test_deferred_pickle_2423_unit.py goes from 8 failed on main to 16 passed/18 subtests on this fix, and the full tests/foundation/reassembly/ suite passes 173/854 subtests, 0 failed, on GraalPy. A new regression test (test_deferred_pickle_graalpy_2433_unit.py) reproduces the failure mode locally by monkeypatching copyreg._reduce_ex to raise the same TypeError, and a second new test pins the subclass-__reduce__ fix.

CPython: mypy 98/32 (baseline, unchanged), pylint 10.00/10, isort clean, tests/foundation/reassembly (176/890 subtests, including the two new regression tests) and tests/corekit (693/6767 subtests, unaffected by this change) pass, 100% coverage on the changed file.

[JarryShaw]

Review: NEEDS CHANGES at 1d529f02e (Opus, different model from the author). The fix itself works — the reviewer actually got GraalPy 25.4.4 running locally and confirmed it reproduces #2433's exact failure on main and is fixed on this PR. Three things to fix:

1. Mechanism docstring is wrong about the reason. It says Generic.__new__ satisfies the walk's stopping condition. Traced on real GraalPy: Generic is a non-heap (built-in) type there (__flags__ lacks _HEAPTYPE), so the walk's first check stops on it — __new__ binding is irrelevant. Reword to say that.
2. Real regression: protocol 0/1 now ignore a subclass's own __reduce__ override (object.__reduce_ex__ checks for one; this PR's protocol<2 branch doesn't), while protocol 2–5 still honour it. Inconsistent. Fix: if type(self).__reduce__ is not object.__reduce__: return self.__reduce__() before the hard-coded tuple.
3. No regression test that fails without the fix, per standing practice. The reviewer confirmed one is cheap: patch copyreg._reduce_ex to raise, pickle under protocol 0/1 — red on main, green on this PR. Add it, plus one for the __reduce__-override case in point 2.

Also: the PR body's corekit count (866/7649) is actually the combined reassembly+corekit run, not corekit alone (693/6767) — fix the label. And since GraalPy was actually run against this fix (not just reasoned about), replace "GraalPy isn't available here" with the real result before this closes #2433.

Confirmed correct and unaffected: protocol ≥2 is byte-identical to main for every object tested, old pickles made by main's __getstate__-only code still load under this PR and vice versa, keeping Generic[_KT] is justified (test_deferred_proto_type_2417_unit.py genuinely depends on it at runtime), no scope creep (3 files only), and the stated CPython test counts reproduce. mypy/pylint/isort and 3.8–3.13 not re-run (environment gap on the reviewer's side, not a finding).

[github-actions]

Coverage: 94.44% (unit tier, Python 3.15, b8ac76505, Unit Tests run success)

Package	Statements	Missed	Branches	Partial	Cover
pcapkit (top level)	104	4	20	2	95.16%
pcapkit/const	19609	133	2620	120	98.84%
pcapkit/corekit	2605	74	818	37	96.52%
pcapkit/dumpkit	282	3	102	3	98.44%
pcapkit/foundation	3070	24	1072	26	98.79%
pcapkit/interface	138	4	52	3	96.32%
pcapkit/protocols	17593	193	4678	179	98.32%
pcapkit/toolkit	924	36	334	10	95.55%
pcapkit/utilities	502	7	144	5	98.14%
pcapkit/vendor	4449	1854	1020	167	55.11%

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

[JarryShaw]

Review: NEEDS CHANGES at 641331f47 (Opus, round 2, different model from the author). The code fix itself is confirmed correct and complete — this time including a real run on GraalPy 25.4.4 (oracle/graalpython release, checksum-verified): the new test file + test_deferred_pickle_2423_unit.py go from 10 failed on main to 19 passed/26 subtests on this head, the __reduce__-override guard is called at every protocol 0–5 on both CPython and GraalPy (verified with an independently-written test, not just reading theirs), and the full tests/foundation/reassembly/ suite passes 176/890 subtests with 100% coverage retained on data.py.

Only the new test file's docstrings need fixing — numbers borrowed from the wrong module, one self-contradiction, and a count that shouldn't be there per the owner's no-counted-numbers-in-docs ruling:

• Module docstring claims "this module's tests go from 8 failed on main to 16 passed/18 subtests" and "173 tests/854 subtests" for the full suite — those are test_deferred_pickle_2423_unit.py's numbers (already correctly credited in the PR body), not this module's (which has 3 tests/8 subtests).
• Same docstring says both "confirmed directly against real GraalPy 25.4.4" and "GraalPy itself cannot be run from this environment" in the same breath — self-contradictory.
• test_protocol_0_1_survive_a_copyreg_reduce_ex_failure's docstring claims it was "confirmed on real GraalPy 25.4.4" — true now (this review just did it), but wasn't true when written.

Fix: drop the counts (owner ruling, no counted numbers in docs/prose) and the "cannot be run" sentence; state plainly that it's confirmed on GraalPy 25.4.4 now that it has been.

Non-blocking, noted for awareness only: test_protocol_2_and_above_never_call_copyreg_reduce_ex only exercises protocol 2 despite its name saying "2 and above"; a subclass whose __getstate__ returns {} now gets __setstate__({}) called at protocol 0/1 where copyreg used to skip it — consistent with protocol 2's existing behavior, not a regression.

[JarryShaw]

Review: GOOD TO GO at b8ac76505 (Opus, round 3, different model from the author). Confirmed the diff from round 2's head is exactly the claimed docstring-only change in the new test file (parsed code and # comments identical; only two docstrings changed) — the other three files are byte-identical to round 2. All three round-2 findings fixed: no counted numbers remain, the GraalPy-availability self-contradiction is resolved, and the "confirmed on GraalPy" claim is now correctly attributed to a cross-reviewer's actual run rather than stated pre-verification. Full tests/foundation/reassembly/ suite re-run clean: 176/890 subtests, 100% coverage on data.py, confirmed against the PR head (not the shared checkout).

Together with rounds 1–2's findings (both fixed and verified, including a real run on GraalPy 25.4.4), this PR is ready to merge.