[Bug]: SSE client splits data lines on U+2028, U+2029 and U+0085, so streamed events fail to parse

未關閉
  • 哪裡不對？
#1,432 0 則留言 0 個 reaction 已指派 2 人 在 GitHub 檢視

維護者通常 2 天內回覆

@rohityan 已經在處理了。

開始於 2026年11月7日。

  • #1433 來自 @devinhuang2001 —— 未關閉

評估

這個 Issue 還沒有評估資料。

描述

What happened?

parse_sse_stream in src/a2a/client/transports/http_helpers.py reads the stream with response.aiter_lines() (line 78). httpx's line decoder follows str.splitlines(), so it also ends a line at U+2028 (LINE SEPARATOR), U+2029 (PARAGRAPH SEPARATOR) and U+0085 (NEL). SSE line ends are only CR, LF and CRLF, and the docstring on line 73 says the parser conforms to the SSE spec.

The server writes these three characters raw: json_utils.dumps uses ensure_ascii=False (src/a2a/utils/json_utils.py:19). sse-starlette frames on CR/LF only, so it sends one data: line. The client splits that line into several. json.loads fails on the first piece, and the stream ends with A2AClientError('Invalid JSON-RPC response in SSE stream').

Any string field in a streamed event can hold these characters, for example message text, artifacts or status messages that quote a document. Resubscribing sends the same task snapshot, so that task stays unreadable over streaming. Non-streaming GetTask is not affected. The same parser is used by the JSON-RPC and REST transports and by the v0.3 compat transports.

This looks like a regression from #1132, which removed httpx-sse. httpx-sse 0.4.3 split SSE lines on CR/LF only (_splitlines_sse).

Repro (library level, using the versions pinned in uv.lock: sse-starlette 3.4.4, httpx 0.28.1)

import asyncio, json
import httpx
from sse_starlette.sse import ServerSentEvent

async def main():
    data = json.dumps({"text": "before
after"}, ensure_ascii=False)
    frame = ServerSentEvent(data=data).encode()
    print(frame.count(b"data:"))  # 1: the server sends one data line
    async def body():
        yield frame
    resp = httpx.Response(200, content=body())
    lines = [l async for l in resp.aiter_lines()]
    print(len(lines))             # 3: the client sees the data line split in two
    json.loads(lines[0][len("data: "):])

asyncio.run(main())

Expected: one data: line that parses as JSON. Observed: JSONDecodeError.

Possible fixes:

  • Split SSE lines on CR, LF and CRLF only, with a small line decoder over aiter_text().
  • Also, or instead: escape U+2028, U+2029 and U+0085 in json_utils.dumps.

Related: #1213, the end-of-stream flush in the same parser.

No existing tests cover these characters in SSE data.

Relevant log output
json.decoder.JSONDecodeError: Unterminated string starting at: line 1 column 10 (char 9)
Code of Conduct
  • I agree to follow this project's Code of Conduct

Drafted with Claude Code. I reviewed the code paths and ran the repro.

主要語言
Python
星號
2.2k
分支
532
平均合併
3 天 2 小時
30 天內合併 PR
46

環境準備

從這裡開始

  1. 先讀完整個 Issue，再讀專案的貢獻指南。
  2. 在 Issue 下留言說明你要接手 —— 這能避免兩個人做同樣的事。
  3. Fork 儲存庫，在一個分支上完成修改。
  4. 送出 Pull Request，並在描述裡引用這個 Issue 編號。

a2aproject/a2a-python 的其他 Issue

查看 a2aproject/a2a-python 的全部 Issue

相似的 Issue

更多 Python Issue

把新 issue 寄到你的電子郵件信箱

精選適合新手參與的 GitHub issue 摘要。