[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

A list of schema items cut short of its declared length was zero-filled to that length (#2686), and items were then read on through the zeros. A framed IGMPv3 Report cut inside its records therefore listed records the capture never held, and a cut HTTP/2 SETTINGS frame listed 0:0 settings.

• ListField stops before any item that would start past the captured octets. The item the cut lands inside still reads its missing octets as zeros, and the cut stays a quiet short read. The packet length < 0 warning from a trailing zero item is gone.
• _zero_filled_items gives the zeros left unread back to the padding budget.
• HTTPv2._make_http_settings pads a cut frame's settings with zeros to its declared Length, so parse → rebuild stays byte-exact.
• Sweep of all six lists of schema items: only IGMPv3 records and HTTP/2 settings produced phantoms.
• Both new test modules fail on base (106 and 43 failures) and pass with the fix. All plain-unittest legs pass.
• Not changed: lists of plain values (e.g. GroupRecord.sources) still read uncaptured items as zeros.

Closes #2756

[JarryShaw]

NEEDS CHANGES at 71f5a46d7 (Sonnet cross-review, reproduced locally)

• Regression: an edited cut SETTINGS frame no longer recomputes its Length. Take a frame cut at 10 octets, set settings to one pair in its to_dict(), and call HTTPv2.from_data. Base gives Length 6 (15 octets); this head pads back to Length 18. The __short_read__ gate in _make_http_settings pads any list, so it should pad only when the settings are still the pairs the captured octets begin.
• Held up: every boundary cut (no zero-only item, the item the cut lands inside kept); the ledger give-back and the raise path; byte-exact rebuild of unedited cuts; no regression across the six schema-item lists; the adapted tests; the tier markers; the docs.
• CI on this head is cancelled until the fix round lands.

[JarryShaw]

GOOD TO GO at 5b967c748 (Sonnet re-review, and my own probe)

• Round-1 regression fixed. An edited cut SETTINGS frame recomputes its Length again. At cuts 10, 12 and 17, one pair gives 6 and none gives 0. Changing values or reversing the pairs also recomputes it. Unedited cuts 0–17 still rebuild byte-exact at 18, from both info and to_dict().
• How the padding is decided. It happens only when the settings still pack to the captured packet[:keep], which is always the settings payload. A cut in the frame header (type/stream) counts as no settings captured.
• Known limit. An "edit" that leaves the settings byte-identical to the capture keeps the declared Length. It can't be told apart from no edit.
• Tests. The new EditedFrameTests fails on 71f5a46d7 and passes here. The 2756, 2686, 2749 and IGMP modules pass, along with the short-read and schema round trips. Project tests, Lint and deploy-pages pass.

[github-actions]

Coverage: 94.45% (unit tier, Python 3.15, 5b967c748, Unit Tests run success)

Package	Statements	Missed	Branches	Partial	Cover
pcapkit (top level)	104	4	20	2	95.16%
pcapkit/const	20918	175	2908	156	98.57%
pcapkit/corekit	2895	76	930	37	96.84%
pcapkit/dumpkit	286	3	102	3	98.45%
pcapkit/foundation	3505	27	1216	32	98.75%
pcapkit/interface	138	4	52	3	96.32%
pcapkit/protocols	22792	155	5910	127	99.00%
pcapkit/toolkit	961	26	342	8	96.62%
pcapkit/utilities	559	7	150	5	98.31%
pcapkit/vendor	5590	2286	1280	214	55.75%

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