[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?

• fix — corrects a defect

---

Description of your pull request and other information

SCTP's capture cuts now go through Schema's built-in short read, with no per-protocol override.

• fix(transport): the SCTP schema on its own repacks a final chunk cut inside its padding at full width #2680: SCTP's padding_length no longer caps the padding at the octets left. A chunk, parameter or cause cut inside its padding now reads short. Schema.unpack records the cut on the innermost item, and pack() writes back the octets read at every level. Before, the SCTP schema packed 34 octets in as 36, and DATAChunk 22 as 24.
• fix(protocols): handle a cut option area in Schema.unpack and drop SCTP's unpack override #2685: Schema.unpack's OptionField hand-back gives back no more of a cut option area's remainder than it read. It records the cut on the option field only when nothing is given back. ChunkListField reports its remainder against the declared area, as other option fields do. This fixes the same defect seen at schema level in TCP.
• fix(transport): rebuild an SCTP packet cut short of its envelope's length (#2673) #2679's Schema_SCTP.unpack override is removed. fix(transport): rebuild an SCTP packet cut short of its envelope's length (#2673) #2679's tests pass unchanged without it, and a non-seekable stream with a length parses again.
• fix(protocols): a declared length running past the data is accepted and zero-filled on rebuild #1325's overrun guard asks a new hook, OptionField._accepts_short_read. It refuses by default. SCTP's PaddedOptionField accepts only a short read confined to an item's padding, which RFC 9260 leaves out of the Length.

Cut sweeps over every padding offset (bytes, streams, IPv4 and IPv6), standalone chunk schemas, and a TCP option area cut at every offset all fail on main. Every leg passes, apart from the CI-script failures main also has. Shipped captures are identical to main.

Closes #2680
Closes #2685

[JarryShaw]

NEEDS CHANGES: wrong layer. Schema.unpack already records a field cut by the end of the data (_note_short_read), and pack() cuts back to the octets read (_cut_short_read). That covers PaddingField too. The record never fires for SCTP because the schema module's padding_length caps the padding at the octets left, so a cut padding reads as complete; on pack, the length is recomputed at full width. On main, DATAChunk.unpack on 22 octets packs 24 with no record.

The rework removes the cap so the built-in records the cut, which also fixes a chunk schema on its own. It drops this PR's separate padding check in the SCTP.unpack override, and it audits other padding helpers that cap the same way.

[JarryShaw]

Pushed the rework, 4c4b0a324→c4e9af739. This PR now closes #2680 and #2685. SCTP capture cuts go through Schema's built-in short read:

• the padding cap is removed;
• Schema.unpack's OptionField hand-back is fixed;
• fix(transport): rebuild an SCTP packet cut short of its envelope's length (#2673) #2679's Schema_SCTP.unpack override is removed;
• fix(protocols): a declared length running past the data is accepted and zero-filled on rebuild #1325's guard gains a refuse-by-default _accepts_short_read hook.

A Sonnet re-review is running.

[JarryShaw]

NEEDS CHANGES at c4e9af739 (tests only), from a Sonnet re-review. The fix is correct, and the review found no regression:

• 1696 SCTP cut rows rebuild exactly; 56 fail on main.
• A TCP option-area cut rebuilds exactly.
• In a 903,528-row schema sweep across nine protocols, 7,222 rows go from inexact to exact and none get worse.
• The merged tree passes 198 modules.

The scoped hook is the right shape: it declares a policy and leaves recording to the built-in. Its scoping is not pinned, though:

1. If PaddedOptionField accepts any short read, every test still passes, so an SCTP parameter or cause whose Length runs past its chunk would no longer be refused. Pin that refusal (runs past the end of the data).
2. Pin the record on TCP cuts at 20–22 octets, which land before the EOOL is consumed, so dropping the hand-back note fails the new modules and not only fix(transport): an SCTP packet cut short by the capture does not rebuild as captured #2673's.
3. If you can, kill the surviving min() (kept = consumed), no-hand-back-note and max(area, 0) mutants.

[JarryShaw]

GOOD TO GO at acf8920a4, rebased onto 57a6c366d. This round changed tests only: pcapkit/'s patch-id matches the reviewed c4e9af739. The re-review's gaps are closed:

• Hook scoping is pinned: an SCTP parameter or cause whose Length runs past its chunk is still refused.
• The hand-back record is pinned by TCP cuts at 20–23 octets.
• The surviving mutants are now killed: the min() and no-note branches, by an over-reporting test option.
• max(area, 0) stays as it is: that mutant is equivalent, since the only ChunkListField sizes its area with max(..., 0) already.

[github-actions]

Coverage: 94.42% (unit tier, Python 3.15, acf8920a4, Unit Tests run success)

Package	Statements	Missed	Branches	Partial	Cover
pcapkit (top level)	104	4	20	2	95.16%
pcapkit/const	21062	177	2914	158	98.56%
pcapkit/corekit	2794	74	890	37	96.77%
pcapkit/dumpkit	281	3	100	3	98.43%
pcapkit/foundation	3507	27	1218	32	98.75%
pcapkit/interface	138	4	52	3	96.32%
pcapkit/protocols	22637	154	5928	138	98.96%
pcapkit/toolkit	961	26	342	8	96.62%
pcapkit/utilities	513	7	144	5	98.17%
pcapkit/vendor	5590	2286	1280	214	55.75%

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