[JarryShaw]

• Searched for similar pull requests
• Followed the coding style — mypy on the 11 changed modules: no new errors vs main (311 both); pylint/isort not run
• make test passes, and a test case covers the change — ordering legs protocols/internet 579, protocols --exclude protocols/internet 1082, corekit 633: all OK; tests/project 404 passed, 1 skipped
• Changelog entry — N/A — centralised in docs(changelog): shared 1.5.0 changelog — long-lived, merges last (#610, #616, #617, #618, #620) #657

What is the purpose of your pull request?

• fix — corrects a defect

---

Description of your pull request and other information

Closes #1223. Closes #1329 (its PadN remainder; the rest merged in #1345).

PaddingField now packs the octets it was given (zeros when absent; a value of the wrong width is truncated or zero-filled, as before, so fresh builds are unchanged); Schema.pack delegates to it. Each site carries the octets through reader, data model and make(): HOPOPT/IPv6-Opts PadN pad and QS report unused, IPv6-Frag reserved_octet, IPv6-Route SR0/Type 2 reserved and RPL padding, MH PadN option/sub-option data, HTTP/2 padding ×3, OSPF crypto reserved, and reserved on the 12 HIP parameters listed here. HIP and SCTP trailing alignment padding is kept as a padding attribute only when non-zero.

Breaking: the data models above gain required fields; five existing test modules that build them by hand were updated.

Before → after, e.g. HOPOPT PadN 3b000104aabbccdd → 3b00010400000000 → exact; SCTP DATA …61ffffff → …61000000 → exact. New tests: tests/protocols/test_padding_kept_as_captured_unit.py, tests/protocols/internet/test_mh_padn_contents_unit.py (both fail on main: 32 + 3 failures). TCP 020405b4 00aabbcc already round-trips on main; kept as a guard.

Left for follow-up (other owners): IPv4 Quick-Start report and option padding (ipv4.py), L2TP offset pad (link/), PCAP-NG SPB and option padding (misc/pcapng.py reader/maker needed), and ~37 MH reserved PaddingField sites not listed on these issues.

[github-actions]

Coverage: 89.47% (unit tier, Python 3.14, 03265271b, 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	2043	89	636	25	94.77%
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	16657	219	4484	183	98.06%
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.

[JarryShaw]

Verdict: NEEDS CHANGES at 3edd77fab (Sonnet cross-review; the author ran on Opus). One blocking regression, which I reproduced:

• L2TPv2 wrong-width padding is now zero-filled. L2TPv2(offset=3, padding=X, payload=b'xx').data:

  X	main	3edd77fab
  aa	…0003aa00007878 (zero-padded)	…0003000000 7878
  aabbccddee	…0003aabbcc7878 (truncated)	…0003000000 7878

  ConditionalField-wrapped PaddingFields used to pack through BytesField (struct Ns). The new PaddingField.pack returns zeros for any wrong width, which contradicts link/l2tpv2.py:283 ("zero-filled or truncated to offset octets"). The fix is to keep the truncate/zero-pad semantics for a non-empty wrong-width value, plus a pinning test.

• Non-blocking:

  • Consider warning on a wrong-width value.
  • The new data-model fields are required, so PadOption(type=…, length=4) and IPv6_Frag(...) without them now raise TypeError. The makers also read them inconsistently: some use getattr(..., b''), some read strictly. Defaults or tolerant reads, chosen consistently.

• Everything else held:

  • Byte-exact: HOPOPT PadN, IPv6-Frag, MH PadN, HTTP/2 DATA, SCTP DATA and HIP reserved/trailing were all lossy on main and are exact here.
  • Captures: 120 of 1614 frames gain only all-zero keys (additive), and all 4858 layers rebuild exactly.
  • Tests: the five edited test modules only supply the new fields.
  • Legs: protocols/internet 579, the rest of protocols 1082, corekit 633 and project 591, all with 0 failures.

The worker is addressing the regression; the label is now review: needs-changes.

[JarryShaw]

Verdict: GOOD TO GO at 03265271b (round 2, Sonnet cross-review; the author ran on Opus).

• L2TPv2 regression is fixed, re-derived by me: L2TPv2(offset=3, padding=X, payload=b'xx').data matches main for aa, aabbcc and aabbccddee. The reviewer also matched b'', None and the HTTP/2 wrong-width cases. The new pin test fails on round 1's pack.
• Tolerant reads: all 21 maker reads use getattr(..., <zero default>), and the HIP defaults are per width. A stand-in that lacks the new fields rebuilds with zeros. The data-model constructors still require the new fields; Info (corekit/infoclass.py:150-201) generates __init__ without defaults, so this is consistent with the !.
• No warning on wrong width: accepted. Captured padding always has the field's width, so only make() callers reach that branch, and L2TP documents silent truncate/zero-fill there.
• Regression re-check:
  • The byte-exact rows (HOPOPT PadN, IPv6-Frag, MH PadN, HTTP/2, SCTP, HIP) are still exact.
  • Across 23 captures, 1614 frames gain only additive zero-valued keys, and make_samples md5s are identical to main.
  • Legs: protocols/internet 579, the rest of protocols 1084, corekit 633 and project 591, all with 0 failures. pytest on tests/project gives 404, which is the same suite counted differently.
• Leftover PaddingField sites that are not in fix(corekit): non-zero padding octets are zeroed on rebuild #1223/fix(mh): reserved fields, BU/BA flag bits and PadN contents are written back as zero #1329 are tracked in fix(corekit): PaddingField sites outside #1223 still zero-fill their padding on rebuild #1437.