[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 #1041. _read_http_header refused a header with no CRLF, i.e. every message with zero field lines; _test_start_line mirrored it, so the dispatcher said unknown HTTP version too. Both now read it as a start line with zero fields.

That check was also what kept the HTTP/2 preface (header PRI * HTTP/2.0) out of HTTP/1. It is now refused by name (_HTTP2_PREFACE_HEADER, derived from http._HTTP2_PREFACE), same HTTP: invalid format; a regression test pins it in the parser, predicate and dispatcher.

New tests: headerless request, headerless response with body, make→read round trip, preface guard. The first three fail on main. Two existing tests that asserted the old ProtocolError as proof of delegation now assert the built bytes. Changelog pin 162 → 163.

[JarryShaw]

Cross-review verdict on 0ab5320fa: GOOD TO GO (ran on Sonnet; author Opus).

• Headerless messages parse. GET / HTTP/1.0\r\n\r\n and HTTP/1.0 200 OK\r\n\r\nbody now parse directly and through HTTP. On main they raised invalid format and unknown HTTP version.
• The HTTP/2 preface is still refused, with the same exception types and messages as main. This holds for the full, truncated and corrupt-tail preface across HTTPv1, _test_start_line and HTTP. The only change is that __cause__ is now None.
• Disclosed widening, judged acceptable. Whitespace variants such as PRI * HTTP/2.0\r\n\r\n now parse as HTTP/1 version 2.0. The dispatcher's byte-exact 24-octet preface check still catches every real preface first. main already parsed PRI * HTTP/2.0 with fields as HTTP/1.
• The predicate and the parser agree on ~45 crafted inputs and 40,000 fuzzed ones. The one disagreement is an orphan continuation line, which main has too.
• No consumer assumes at least one header field. The edited tests still test what their names say.
• New tests: they fail on main (6 + 1) and pass on the head.
• Changelog: changelog_md.py --check is in step, and the pin is 164 (measured 164).

Not breaking: no signature or pack path changes.

Nits: test_httpv1_only_raises_protocol_error still labels the preface entry "no CRLF" under the converted sites, and no test pins the whitespace variants.

Follow-up filed as #1050: from_data fails on any empty-bodied message. That is pre-existing on main, and I reproduced it.

[JarryShaw]

Cross-review verdict on 82ec132a8: GOOD TO GO (ran on Sonnet; rebase re-check).

The diff over pcapkit/ and tests/ matches the reviewed head 0ab5320fa exactly: identical git patch-id --stable and the same three files. Only 1.5.0.rst, CHANGELOG.md and the process.rst:104 pin changed. The count is 165, matching the pin, and changelog_md.py --check is in step. The #1038 and #1041 entries each appear once, with no conflict markers. test_http_unit gives 64 passed, test_conventions_doc_claims 37 passed and test_changelog_md 60 passed. The earlier verdict stands.