[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 under docs/source/changelog/ and regenerated CHANGELOG.md, if the change is user-visible

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
• chore — anything else

---

Description of your pull request and other information

Closes #1042. MP_CAPABLE now reads and makes all five RFC 8684 §3.1 lengths: 4, 12, 20, 22 and 24.

• Schema: skey is present when length > 4 and rkey when length > 12. New fields are dl_len (UInt16, > 20) and checksum (2 octets, > 22), named to match DSS.
• Reader: the guard now accepts (4, 12, 20, 22, 24).
• Maker: takes dl_len and checksum, which **kwargs used to drop silently. It picks the length from how many of skey/rkey/dl_len/checksum are non-None. Passing a field without the one before it raises ProtocolError.
• Data model change callers can notice: MPTCPCapable gains dl_len and checksum, and skey becomes Optional[int]. Direct constructors must pass both new fields, and to_dict() gains two keys.
• Tests: a parse test and a make→read test for each length. 4/22/24 failed on main and 12/20 failed only on the missing dl_len. The [OptNo message-count tripwire goes from 32 to 33. No EXPECTED_FAILURES entry covered MP_CAPABLE.

make pylint/isort were not run in full. mypy reports nothing new in the touched regions.

[JarryShaw]

Cross-review verdict on f1f0f2179: GOOD TO GO (ran on Sonnet; author Opus).

• Schema matches RFC 8684 Figure 4, checked against the RFC text: skey if length > 4, rkey if > 12, then dl_len (16 bits) and checksum (16 bits, optional), giving lengths 22 and 24. Not checking flag A against the length is defensible: Length decides which fields are present.
• Reader: 4, 12, 20, 22 and 24 parse, with each field None below its length. Lengths 3, 5, 11, 13, 19, 21, 23, 25 and 28 each raise ProtocolError: TCP: [OptNo 30] invalid format.
• Maker: of all 16 combinations of set and unset fields, only the five contiguous prefixes build, and each round-trips byte-exact. The other 11 raise. Passing only skey still builds the 20-octet form (examples/generators/options.py:448).
• Tests: the new tests fail on main (13 failed). The [OptNo count is 33, measured, so the 32→33 pin update is right. All MPTCP modules and test_option_roundtrip_unit pass.
• Changelog: changelog_md.py --check is in step. The pin is 164 against a measured 164.

Labelled breaking. The data model MPTCPCapable.__init__ now requires dl_len and checksum positionally, so an existing constructor call raises TypeError. skey also becomes Optional[int], and to_dict() gains two keys. The changelog entry already says so.

Non-blocking: dl_len=70000 wraps silently, and the checksum length is not checked. Both are library-wide field behaviour, and DSS does the same.

[JarryShaw]

Cross-review verdict on a785d9bb7: GOOD TO GO (ran on Sonnet; rebase re-check).

git range-diff against the reviewed head f1f0f2179 shows exactly one changed hunk: the process.rst:104 pin, 164→165. The code and tests are byte-identical. On the new head, the count measures 165 and changelog_md.py --check is in step. test_tcp_mptcp_capable_length_unit gives 18 passed, test_conventions_doc_claims 37 passed, and test_changelog_md 60 passed. The earlier verdict and the breaking label stand.