[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 chore(docs): centralise changelog entries for unreleased versions #2483

What is the purpose of your pull request?

• refactor — changes neither behaviour nor performance

---

Description of your pull request and other information

The owner's option 1 on #2593 and #2594. A make() field that holds an enum takes the member or an int, resolved with the enum's get. The X_default/X_namespace/X_reversed keyword groups and _make_index's inspection of the caller's frame are removed from every make() and _make_*, SCTP and 802.11 included.

• Resolver: Protocol._make_index(value, Enum, *, bits, name) checks the code as _make_uint does, then returns the member. A name, a bool wider than one bit, or a code that does not fit raises ProtocolError. This settles refactor(protocols)!: range-check the registry codes, timestamp counts and byte fields the makers still mask #2676's first bullet: IPv4(protocol=256) and IPv6_Route(next=256) raise where they wrote 0x00. The enum's own range is not applied: a code that fits its field resolves to the member get gives, which minting registries create, or to an int where get refuses. Captures with unassigned codes therefore still rebuild.
• Strings: callers holding a string use the new public Protocol.resolve_index(name, default=None, *, namespace, reversed=False), which pep.rst describes.
• Docs: classes that shipped the keywords get versionchanged and versionremoved notes.
• Option-list codes in MH, IPv4, HIP, TCP, DCCP, SCTP and PCAP-NG go through the same check (fix(protocols): range-check option-list codes in MH, IPv4, TCP, DCCP, SCTP and PCAP-NG makers #2693).
• Guard: a test fails if a make() regains a group, or if _make_index inspects a frame.

Breaking: the shipped keywords are removed with no deprecation, per the owner's ruling. NGAP criticality no longer takes a str. HTTPv1(status=404) writes 404 Not Found, where it used to write 404 ; an unassigned status writes its registry phrase.

Every leg passes, apart from the CI-script failures main has too, with the same ids. Shipped captures parse identically to main.

Closes #2594
Closes #2693

[JarryShaw]

NEEDS CHANGES at 22fb952d3, from a sharded Sonnet review with an Opus check of each finding. The core holds: no groups or frame inspection remain, every _make_index site refuses out-of-range values, a bool and a name, and captures rebuild exactly. What has to change:

1. Regression: _make_index also applies the enum's own range, so HIP transport mode, HIT suite list, MH UPN reason and FBM fb_type refuse codes that fit their wire field. Keep only the width check plus get.
2. Option-list codes are still masked in MH and IPv4 (and in TCP, DCCP, SCTP and PCAP-NG, per fix(protocols): range-check option-list codes in MH, IPv4, TCP, DCCP, SCTP and PCAP-NG makers #2693). Range-check them the way HOPOPT does. This folds fix(protocols): range-check option-list codes in MH, IPv4, TCP, DCCP, SCTP and PCAP-NG makers #2693 into this PR.
3. Prose: a code no member holds comes back as a minted member where the registry mints, and as an int only where get refuses. Fix this in _make_index, registry-protocol.rst and the body. Drop the timed context in resolve_index, pep.rst, the dumpkit NOTE and the test docstrings.
4. Tests: PCAPIO(protocol='ETHERNET') has no test; add one, with an unknown name. Type HTTP/2 settings as Enum_Setting | int.

[JarryShaw]

Pushed the fix round 22fb952d3→e7bc62ce5, rebased onto 57a6c366d. Every finding from the sharded review is addressed:

• _make_index drops the enum-range check, so HIP Transport, HITSuite, MH UPN and FBM codes build up to their field width.
• Option-list codes are range-checked in every family. This closes fix(protocols): range-check option-list codes in MH, IPv4, TCP, DCCP, SCTP and PCAP-NG makers #2693, and MH _make_msg_be now takes BindingError.
• The prose is corrected: a minting registry returns a member, and only an enum whose get refuses returns an int. Timed wording is removed.
• New: a PCAPIO link-type-name test, and HTTP/2 settings typed Enum_Setting | int.

A Sonnet re-review is running.

[JarryShaw]

NEEDS CHANGES at e7bc62ce5, for prose only, from a Sonnet re-review. Every code fix holds:

• the four fields build up to their field width;
• all 114 _make_index sites refuse out-of-width values, -1, True and names;
• option-list codes are range-checked in every family, in both branches;
• shipped captures rebuild byte-exactly.

Timed words remain in four test docstrings and comments: test_hip_list_argument_str_names_2394_unit.py ("before #2594" twice, "before #2394") and test_protocol_base_unit.py ("the old keyword group"). Restate each as what holds now.

[JarryShaw]

GOOD TO GO at 3d1dea6f1. This amend changes comments and docstrings only; I checked that all seven changed files have identical syntax trees once docstrings are stripped. The four flagged test docstrings, and a few other timed phrases in the touched files, now state what holds. The re-review confirmed every code fix:

• the four fields build up to their field width;
• all 114 _make_index sites refuse out-of-width values, -1, True and names;
• option-list codes are checked in every family;
• captures rebuild byte-exactly.

Merging this first will make the other protocol PRs cheaper to rebase.