Hacktoberfest 2026：维护者为十月标记出来的 issue，仍然开放、适合新手。 浏览 Hacktoberfest issue

A 304 with content-length is accepted or rejected by framing alone

未关闭
  • 哪里不对？
#1,330 0 条评论 0 个 reaction 已指派 0 人 在 GitHub 查看

维护者通常 4 天内回复

还没有人认领这个 Issue。

评估

难度
4/5
预计耗时
3-5 天
新手友好度
45/100
Issue 类型
缺陷
描述清晰度
基本清楚
活跃度
活跃
技术栈
python
领域
networking

调研方向

Start in src/h2/stream.py at _initialize_content_length (line 1369), receive_headers (line 1113), send_headers (line 909), and remotely_pushed (line 1052), using the standalone probe to reproduce the reported cases. First resolve whether 204/304 should set the expected length to 0 or None, then cover the exemption and HEADERS-frame check together. Done means the 17 reported assertions pass with uv run --frozen pytest tests/ -q at 100% coverage.

由索引模型根据 Issue 内容生成。

描述

While working on #1328 I found two more holes in the same function, and they turn out to be two halves of one problem. I have not touched anything — the measurements are in a standalone probe, and I would rather agree on the direction first.

A 304 or 204 carrying content-length is accepted or rejected depending on framing

_initialize_content_length (src/h2/stream.py:1369) implements exactly one exemption, self.request_method == b"HEAD" (:1375). RFC 9113 § 8.1.1 names three messages that are "defined as having no content": 204, 304, and the response to HEAD, and it says a message defined to have no content "MAY have a non-zero content-length header field". RFC 9110 § 8.6 repeats it for 304 — that is how a client learns the size of the cached entity. H2Stream never stores :status, so the other two cannot be written as the code stands. Observable today, on master and on #1329 alike:

304 + content-length: 1234 + empty DATA with END_STREAM   -> InvalidBodyLengthError: Expected 1234 bytes, received 0
304 + content-length: 1234 + END_STREAM on the HEADERS    -> accepted ['ResponseReceived', 'StreamEnded']
204 + content-length: 1234 + empty DATA with END_STREAM   -> InvalidBodyLengthError: Expected 1234 bytes, received 0

One message, two verdicts, decided purely by which frame carried END_STREAM. That is an enforcement that rejects something the RFC permits, and it is reachable from a plain H2Connection.receive_data on a legitimate response.

The same hole #1328 fixed, in the other branch of the same if

receive_headers (:1113) — the else branch, i.e. a HEADERS block that is not trailers, calls _initialize_content_length(headers) and nothing else. So a bodyless message that ends on the HEADERS is never policed at all:

POST content-length: 15 + END_STREAM on the HEADERS, no DATA  -> accepted ['RequestReceived', 'StreamEnded']
200  content-length: 1234 + END_STREAM on the HEADERS         -> accepted ['ResponseReceived', 'StreamEnded']

content-length: 15 with nothing sent is the same lie #1328 is about, and _actual_content_length stays 0.

The obvious one-line fix is not safe on its own. Adding if end_stream: self._track_content_length(0, end_stream=True) to that branch, measured:

                                    before        after that line
POST content-length: 15, no body    accepted      InvalidBodyLengthError(15, 0)   <- intended
304  content-length: 1234, no body  accepted      InvalidBodyLengthError(1234, 0) <- new breakage
204  content-length: 1234, no body  accepted      InvalidBodyLengthError(1234, 0) <- new breakage
POST content-length: 0, no body     accepted      accepted                          <- control
GET  no content-length, no body     accepted      accepted                          <- control

So the 204/304 exemption has to land first, or with it. That is why I am asking rather than sending a patch: the exemption is a design call. Setting self._expected_content_length = 0 for 204/304 mirrors the HEAD precedent and additionally makes DATA after a 204 an error, which RFC 9110 § 6.3 implies; setting it to None only skips the comparison. I lean to 0, but that is your call, and H2Stream would need to start recording :status.

Two smaller ones in the same file, and what I am not reporting

Two smaller ones in the same file, both one-line

  • send_headers (:909) assigns self.request_method = extract_method_header(bytes_headers) on every header block, including a trailers block, which has no :method — so outbound trailers overwrite a remembered b"HEAD" with None. H2Connection.send_headers(1, HEAD) then a trailers block, then a 200 + content-length: 1234 + empty DATA + END_STREAM, raises where it should not. Guarding on method is not None fixes it.
  • remotely_pushed (:1052) records self._authority but never the method, so a pushed stream never gets the HEAD exemption at all. A PUSH_PROMISE with :method: HEAD followed by 200 + content-length: 1234 + empty DATA + END_STREAM raises; the identical push with :method: GET raises the same way, which is the tell that the method is simply not consulted.

Not reported

  • RST_STREAM or a local reset mid-body runs no length check, and should not: the message is abandoned and the stream is closed, so nothing later can contradict the sender. Measured, no error, and I read that as specified rather than a bug.
  • A content-length in outbound trailers is a real asymmetry — h2 will send a trailers block carrying content-length: 0, and feeding those exact bytes back into an h2 client raises ProtocolError: Received content-length header in trailer. But fixing it means changing the trailer field in four existing tests in tests/test_basic_logic.py, so I would file that separately rather than widen this.
  • A 1xx carrying content-length is accepted and then overwritten. Inconsistent with the new trailers rule, but the RFC puts that obligation on the sender and no wrong byte is ever compared, so I do not think it is worth an interop risk.

Verification

uv run --frozen pytest tests/ -q is 1670 passed at 100.00% coverage (466 statements, 98 branches in stream.py, 0 missed), on d2b43d0. Every scenario above is a runtime measurement against the real H2Connection, not a reading. I have the probe as a standalone script and the 17 assertions as a pytest file; say the word and I will open a PR with the exemption and the HEADERS-frame check together, at 100% coverage.

主要语言
Python
星标
1k
派生
197
30 天内没有已合并 PR

环境准备

这个项目没有提供开发容器、Dockerfile 或贡献指南，环境需要你自己搭建：先看它的 README，通用步骤见我们的新手贡献指南。

从这里开始

  1. 先读完整个 Issue，再读项目的贡献指南。
  2. 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
  3. Fork 仓库，在一个分支上完成修改。
  4. 提交 Pull Request，并在描述里引用这个 Issue 编号。

python-hyper/h2 的其他 Issue

查看 python-hyper/h2 的全部 Issue

相似的 Issue

更多 Python Issue

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。