[JarryShaw]

• Searched for similar pull requests
• Followed the coding style — mypy and isort --check-only are clean on the three TCP modules; pylint adds no new messages on changed lines (its existing W0613/C0301 baseline remains)
• make test passes, and a test case covers the change. I ran the new module, the TCP/MPTCP modules, test_option_roundtrip_unit, tests/project, and run_unittest_leg.py protocols --exclude protocols/internet. I did not run the full suite.
• Changelog entry — N/A, added centrally

What is the purpose of your pull request?

• fix — corrects a defect

---

Closes #1211
Closes #1218
Closes #1219
Refs #1223. The TCP repro there, 020405b4 00aabbcc, now round-trips because the post-EOOL octets are kept. PaddingField is unchanged.

Before → after, on c9676c563, comparing from_data(TCP(raw).info) with raw: header octet 12 5e, POC-SP 0a033f00, QS 1b08834012345679, EOOL 00010101 020405b4 at DO 7, and UTO 1c048005 all used to come back mismatched. Now all are byte-exact.

Breaking: these data-class fields are now required: Flags.reserved, PartialOrderServiceProfile.reserved, QuickStartResponse.reserved and .nonce_reserved, and UserTimeout.granularity. The TCP offset, POC-SP profile, QS flags and nonce BitField dicts also need a reserved key. Data_TCP gains padding whenever options are present. make() gains padding and reserved. padding is re-emitted only while the options still end with EOOL. If the EOOL is edited out, the padding is dropped and the Data Offset is recomputed. A User Timeout keeps its parsed unit only while the value still fits in it, and switches unit otherwise. Six existing test modules built the old shapes by hand, so I updated them. One of them, test_tcp_mptcp_join_flag_ordering_unit, greps make()'s source, and its substring is now _make_tcp_options(options.

Tests: the new tests/protocols/transport/test_tcp_roundtrip_unit.py has 11 passed and 25 subtests passed. Against origin/main it gives 26 failed and 5 passed. The new edited-data tests (EOOL removed, NOP and EOOL removed, every option removed, UTO unit switch) give 8 failed against the first revision, 1fd285e7a. All 1233 TCP frames across the 22 captures still rebuild byte-exactly. test_option_roundtrip_unit gives 6 passed and 362 subtests passed, with no tcp entry in EXPECTED_FAILURES. The ordering leg ran 909 tests with 0 failures, and tests/project gives 389 passed and 1 skipped.

[JarryShaw]

Cross-review verdict on 1fd285e7a: NEEDS CHANGES (Sonnet reviewed; author Opus). One edited-data regression, which I re-derived myself.

padding goes stale when the EOOL is removed. I parsed a header with MSS, NOP, EOOL plus 6 padding octets. It rebuilds byte-exactly. Then I removed the options and called make() from _make_data's output, which raises ProtocolError: TCP: padding given without an End of Option List option. main built it. This is the same edited-data shape that IPv4 #1344 had to fix today. Drop the carried padding when the option list no longer ends in EOOL, and add tests for removing the EOOL and for removing every option.

Minor, same class: editing a parsed UTO with G clear to 40000 s raises timeout value too large: 0, where main auto-switched to minutes. The message also prints the timeout argument rather than the value.

Held up:

• fix(tcp): reserved bits in the TCP header and in POCSP/Quick-Start options are written back as zero #1211: RFC 9293 §3.1 bits 4–6 are reserved and bit 7 is NS, so nothing is double-counted. All 16 octet-12 patterns are byte-exact (14 mismatch on main), as are POC-SP and QS-Response at their extremes.
• fix(tcp): bytes after EOOL are dropped and the Data Offset is recomputed, shrinking the header #1218: non-zero, repeated and option-like octets after EOOL all round-trip. A fresh make() pads once.
• fix(tcp): User Timeout option loses its granularity bit on rebuild #1219: the RFC 5482 G bit holds for 7fff/ffff/8000/0000/003c.
• Revert proof and leg: the revert proof is 22/4. The leg is 907 tests, 0 failures. All TCP fixture frames rebuild.

[JarryShaw]

Round-2 verdict on 72615ea54: GOOD TO GO. I verified the delta from 1fd285e7a myself: 2 files, +100/−36.

The blocker is fixed: padding is written back only while the options still end in EOOL, and Data Offset recomputes otherwise. I parsed MSS, NOP, EOOL plus 6 padding octets (which rebuilds byte-exactly), then rebuilt it with options removed:

removed	Data Offset	options written
EOOL	7	020405b401010100 (realigned)
NOP + EOOL	6	020405b4
everything	5	none

UTO units: seconds stay seconds while they fit in 15 bits and minutes stay minutes only for whole minutes, so 40000 s with G clear becomes 666 min (1c04829a). The error now prints the real value. The parsed 1c048001, 1c040001, 1c047fff and 1c04ffff all still rebuild exactly.

Tests: the new module fails 8 tests against the previous head and 26 against origin/main. The protocols --exclude protocols/internet leg runs 909 tests with 0 failures.

[github-actions]

Coverage: 88.94% (unit tier, Python 3.14, 72615ea54, Unit Tests run success)

Package	Statements	Missed	Branches	Partial	Cover
pcapkit (top level)	104	4	20	4	93.55%
pcapkit/const	18777	1032	2348	843	90.27%
pcapkit/corekit	1930	89	602	22	94.59%
pcapkit/dumpkit	189	1	68	1	99.22%
pcapkit/foundation	2442	102	852	39	94.38%
pcapkit/interface	112	7	40	5	92.11%
pcapkit/protocols	15993	207	4168	178	98.05%
pcapkit/toolkit	495	73	150	1	83.57%
pcapkit/utilities	429	4	122	4	98.55%
pcapkit/vendor	4409	2359	1006	158	42.84%

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