[JarryShaw]

• Searched for similar pull requests

• Followed the coding style — ran pylint and mypy on the seven files (no findings on touched lines); isort not run

• make test passes, and a test case covers the change — legs foundation, protocols --exclude protocols/internet, protocols/internet, plus tests/project

• Changelog entry — N/A — centralised in docs(changelog): shared 1.5.0 changelog — long-lived, merges last (#610, #616, #617, #618, #620) #657

• fix — corrects a defect

---

Closes #1364

ProtocolBase.register and the Internet/Link/Transport/SCTP/Frame/PCAPNG overrides now compare the argument with the stored entry before resolving a ModuleDescriptor: the stored entry, or an equal descriptor, is a silent no-op that keeps it; a different entry still warns; invalid input still raises. Extractor.register_dumper was already fixed by #1376 and is pinned only.

Probe, TCP.register(21, TCP.__proto__[21]) (same for IPv4, Ethernet, SCTP, Frame, PCAPNG): before — 1 warning, entry becomes the class; after — 0 warnings, entry is the stored descriptor.

Tests: new tests/protocols/test_registry_stored_descriptor_unit.py, 6 passed + 28 subtests; on origin/main 8 failed (SCTP alone reverted: 1). Legs: 522 / 1059 / 566 tests, 0 failures; tests/project 399 passed, 1 skipped.

[JarryShaw]

Verdict: GOOD TO GO at 584d5b02a. Reviewed by Sonnet; the author ran on Opus.

• Repro: on main, all seven registrars (Internet, Link, Frame, PCAPNG, SCTP, Transport via TCP/UDP, ProtocolBase) warn and swap the stored descriptor for the class. On this head there are 0 warnings, and __proto__[code] is the original, still-unimported descriptor.
• Semantics: an equal copy of the stored descriptor is a silent no-op that keeps the stored object. A class over a stored descriptor still warns, which matches the docstrings. A descriptor over a stored class is silent, as on main. Nothing diverges from fix(protocols): parser registrars warn only when the entry really changes (#1362) #1400.
• Invalid input: a plain tuple equal to a stored descriptor still raises RegistryError, as do non-class inputs.
• Coverage of registrars: the dumper, engine, reassembly and traceflow registries already keep shipped descriptors (question(registry): should a built-in engine, reassembly or traceflow overridden by name be restorable through the public registrar? #1363/fix(registry): foundation registrars restore the built-in a name shipped with (#1363) #1376). Re-registering all 18 is silent and keeps identity, and no other registrar has this pattern.
• Tests: 8 of the new module's tests fail with the 7 source files reverted. protocols --exclude protocols/internet gives 1059 tests with 0 failures, and foundation 522 with 0 failures.
• Nits, not blocking: the module-level wrappers such as register_transtype still call .klass, so going through them still imports the class (the stored entry stays a descriptor). The guard is also repeated in seven places, and one PCAPNG docstring line is overlong.

[github-actions]

Coverage: 89.43% (unit tier, Python 3.14, 584d5b02a, Unit Tests run success)

Package	Statements	Missed	Branches	Partial	Cover
pcapkit (top level)	104	4	20	4	93.55%
pcapkit/const	18791	968	2352	869	90.90%
pcapkit/corekit	2038	89	634	25	94.76%
pcapkit/dumpkit	234	2	84	2	98.74%
pcapkit/foundation	2637	98	928	48	94.78%
pcapkit/interface	112	7	40	5	92.11%
pcapkit/protocols	16518	219	4434	183	98.04%
pcapkit/toolkit	539	74	168	3	84.58%
pcapkit/utilities	429	4	122	4	98.55%
pcapkit/vendor	4409	2342	1006	157	43.25%

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