fix(infer): raise the intended ValueError when tool_choice names an unknown tool

[Lesereingrape]

PR type

• Bug Fix
• Feature
• Others

PR information

A swift deploy server answers HTTP 500 Internal Server Error for a request that selects a tool which is not in tools, even though the code has a message prepared for exactly that case.

ChatCompletionRequestMixin.__post_init__ filters the tool list by name:

name = self.tool_choice['function']['name']
tool = next(tool for tool in self.tools if tool['function']['name'] == name)
if tool is None:
    raise ValueError(f"Tool choice '{name}' not found in tools.")
self.tools = [tool]

next(generator) with no default raises StopIteration as soon as the generator is exhausted, so tool is never None and the ValueError below it is unreachable. __post_init__ runs while the request object is being built, i.e. before create_chat_completion (swift/pipelines/infer/deploy.py:272) gets a chance to run its own except Exception → 400 handler, so the exception escapes the handler entirely.

Measured on the current main (5256c62e), constructing the request directly:

File "swift/infer_engine/protocol.py", line 250, in __post_init__
    tool = next(tool for tool in self.tools if tool['function']['name'] == name)
StopIteration

and end to end, with a FastAPI app that registers the same create_chat_completion(request: ChatCompletionRequest, raw_request: Request) signature as deploy.py:272 and a body of {"model": "m", "messages": [...], "tools": [{"type": "function", "function": {"name": "get_weather"}}], "tool_choice": {"type": "function", "function": {"name": "no_such_tool"}}}:

• before the fix: 500 Internal Server Error, body Internal Server Error — the client is told nothing about what it sent wrong;
• after the fix: 422 with {"type": "value_error", ..., "msg": "Value error, Tool choice 'no_such_tool' not found in tools."} — the message the author wrote.

tool_choice naming a specific function is part of the OpenAI chat-completions request format that this endpoint is compatible with, so a client that misspells a tool name is a normal case, not a crafted one. A wrong name should not look like a server fault.

The fix is one argument: give next() the default the following if already expects.

tool = next((tool for tool in self.tools if tool['function']['name'] == name), None)

Tested with the new tests/general/test_protocol_tool_choice.py, which pins the failing case plus two cases that must keep working (a name that is in tools narrows request.tools to just that tool; a string tool_choice such as 'auto' leaves the list untouched):

• on main before the fix: Ran 3 tests ... FAILED (errors=1), the error being the StopIteration above;
• after the fix: Ran 3 tests ... OK, and the same route probe now returns the 422/ValueError message;
• flake8, isort --check-only and yapf --diff report nothing on either file.

Disclosure: this PR was prepared, tested and submitted by an AI agent working on behalf of the account owner.