Incorrectly infering `str` as `Never`

[Stream29]

Describe the Bug

https://github.com/Stream29/dify-official-plugins/blob/1ad4d571a4d80138a9a76419c8ac9e1074567ce6/models/anthropic/models/llm/llm.py#L774

This will produce a type error without the type: ignore:
cd models/anthropic
uv sync
uv run pyrefly check

ERROR `str` is not assignable to attribute `arguments` with type `Never` [bad-assignment]
   --> models\llm\llm.py:774:61
    |
774 | ...                   tc.function.arguments = current_tool_params
    |                                               ^^^^^^^^^^^^^^^^^^^
    |
 INFO 1 error (2 suppressed)

Sandbox Link

No response

(Only applicable for extension issues) IDE Information

No response

[javabster]

Hi @Stream29 thank you for opening an issue, unfortunately I can't seem to figure out from the code you've shared what the type of tc.function.arguments is meant to be, so I'm not able to determine if this is a bug or expected behaviour from Pyrefly. If possible could you share more context from your code? or if you're able to create a minimum reproducible example of the bug in our sandbox that would be even better: https://pyrefly.org/sandbox/

[Stream29]

@javabster I tried to make a reproduction in the sandbox. But I didn't figure out how to add pydantic as dependency in sandbox.

My codebase depends on dify-plugin which has such code: https://github.com/langgenius/dify-plugin-sdks/blob/194bb771dfc6e4991ce7f6f9b6dc6e9c8b54e99a/python/dify_plugin/entities/model/message.py#L210
, where is the incorrect point.

[javabster]

Ah ok thank-you for the extra context, looks like this could be a bug with our pydantic integration - cc @migeed-z