[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
• Added a changelog entry — N/A, centralised in chore(docs): centralise changelog entries for unreleased versions #2483

What is the purpose of your pull request?

• 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
• const — regenerates constant or registry tables, members keeping their numeric values

---

Description of your pull request and other information

Part of #2511. The LinuxSLL, FRAD and NETLINK maps are covered by #2510.

• register_ipx, register_sctp_chunk, register_sctp_parameter and register_sctp_cause are added to foundation.registry.protocols. Each follows the existing wrappers: a # NOTE: <map> tag, _check_code and a target check, the classmethod call, and _register_name for protocol maps. They are exported and documented alongside the others.
• registry-protocol.rst has a new Registration Wrappers section. It states the two-part shape and the exception for inherited classmethods (owner's option 2 on refactor(registry): every class-level registry map gets its foundation.registry wrapper #2511).
• ext.rst: the SCTP and ICMP extension-point rows now name the wrappers.
• A new test pins every class-level map to its wrapper and fails without this change. Its excused table lists the maps other streams own.

Loopback's wrapper is deferred to the link-layer section, which #2510 is editing. Its int keys raise a question that is posted here once the review lands.

[JarryShaw]

Cross-review (Sonnet) at d5d81d620: NEEDS CHANGES. What is here is correct: the inventory is complete, the four wrappers match their siblings in shape, exports and docs, and the tests fail without the change. But the owner's option 2 on #2511 assigns more to this PR than it delivers:

• IPX needs its own register, because the inherited Internet.register is TransType-typed. Loopback needs both a register classmethod and a register_loopback wrapper.
• The Registration Wrappers section has to state the inherited-classmethod exception. As written, it contradicts IPX as shipped.
• EXCUSED attributes Loopback to refactor(link): dispatch LinuxSLL's non-EtherType ARPHRD values through its own registry map #2510, which does not own it.
• The coverage test pins only the wrapper half, so it cannot catch a missing classmethod.

Fix round dispatched. It is pushed once queued runs drop below 20. Loopback's int keys raise the open question posted on #2510.

[JarryShaw]

The fix round is done locally (e451c70cb, not pushed yet). It adds IPX.register, Loopback.register, register_loopback, _check_code(kind=None) and a classmethod-half coverage pin, plus the option-2 and naming text. Its review asks for one more small round, now running: the naming rule as written is false for register_apptype, and IPX lost a junk-code pin. Decisions:

1. IPX's __proto_key__.
   • (a) No pairing, as on main. Any key is accepted through code=, including TransType.UDP, which then keys Packet.NCP.
   • (b) (Packet, False), like Internet and LinuxSLL. That breaks 1.5.0's accepted {IPX: 4}.
   • (c) (Packet, True), with register_ipx taking ints. It keeps 1.5.0's bare-int keys, rejects other enums and bools, and mirrors SCTP.
     Recommendation: (c), which is what the branch does.
2. The naming rule's exact wording (your refactor(link): dispatch LinuxSLL's non-EtherType ARPHRD values through its own registry (#2510) #2527 Q4 ruling). Read literally, it fails for register_pcap/register_pcapng, and for register_apptype, which takes bare ints but writes both TCP's and UDP's maps. Options: (a) state the rule with those named exceptions; (b) rename them, which is breaking. Recommendation: (a).

[JarryShaw]

1. C
2. A

[JarryShaw]

Owner chose 1(c) and 2(a). IPX declares __proto_key__ = (Packet, True) and register_ipx takes ints, which keeps 1.5.0's bare-int keys and rejects other enums and bools, as the branch already does. The naming rule is stated with its exceptions named (register_pcap, register_pcapng, register_apptype, and any other the check finds). The second fix round, now running, writes exactly that.

[JarryShaw]

Fix rounds at 5b2833f59, cross-reviewed (Sonnet, round 2): GOOD TO GO.

• IPX.register and Loopback.register mirror Link/Internet.register with per-key warnings, and register_loopback is added. Removing either classmethod fails the tests.
• The naming rule's exceptions (register_pcap, register_pcapng, register_apptype) and the protocol-wide registrars are stated on registry-protocol.rst. An AST pass over every wrapper matches them.
• A None/str mutation fails the new subtests, and every added line in protocols.py, ipx.py and loopback.py is covered.

Non-blocking: the "tagged # NOTE" wording does not fit three wrappers, the two new classmethods have no versionadded, and one import is out of isort order.