[SayHelloToWorld]

Fixes #53255

AI-assisted, auto-generated by the vLLM issue bot (Copilot Agent).
No existing open PR addresses this issue (checked before opening).

Summary

• Root cause: non-streaming path set finish_reason="tool_calls" when tool_choice=="required" without verifying any tool_calls existed.
• Fix: require non-empty tool_calls before mapping stop->"tool_calls".
• Tests: added a minimal non-streaming unit test asserting finish_reason=="stop" when tool_choice="required" produces no calls.
  What changed
• vllm/entrypoints/openai/chat_completion/serving.py
  • Guarded the non-streaming finish_reason decision with and bool(tool_calls).
• tests/entrypoints/openai/chat_completion/test_chat_error.py
  • Added test_required_tool_choice_no_calls_non_stream_finish_reason asserting the response finish_reason is "stop".
    Commit & push
• Created branch fix/issue-53255 and committed the change with Co-authored-by trailer.
• Attempted to push to origin but push failed due to lack of GitHub credentials:
  fatal: could not read Username for 'https://github.com': terminal prompts disabled
  Next steps for a human maintainer

1. Review the two-file diff (sm...

[longluubao]

Heads-up for reviewers: this overlaps #53256, which fixes the same issue and was opened ~15 minutes before this one (#53256 at 11:52:55Z, this at 12:07:37Z). The bot's note that "no existing open PR addresses this issue" was not accurate at the time it was written.

The code change is the same idea in both. #53256 additionally carries the end-to-end measurement behind the report — unpatched vs patched on the same host and image, with the tool parser off and on — showing finish_reason going from "tool_calls" to "stop" when no call is produced, and staying "tool_calls" when one is.

I'm the author of both #53255 and #53256, so I have no objection to whichever one a maintainer prefers to take; flagging it only so the same change doesn't get reviewed twice.

[SayHelloToWorld]

@longluubao Ok, I will close this one.