fix(openai): handle pydantic BaseModel as metadata

[hassiebp]

Important

Handle pydantic.BaseModel as metadata in langfuse/openai.py by converting it to a dictionary, and update tests accordingly.

• Behavior:
  • In _get_langfuse_data_from_kwargs() in langfuse/openai.py, handle metadata as pydantic.BaseModel by converting it to a dictionary using model_dump(). If not a BaseModel or dict, set metadata to an empty dictionary.
• Tests:
  • Remove test_fails_wrong_metadata() from tests/test_openai.py as metadata no longer raises TypeError for non-dict inputs.

^{This description was created by for fdc8a2b. You can customize this summary. It will automatically update as commits are pushed.}

Disclaimer: Experimental PR review

Greptile Overview

Greptile Summary

Added support for Pydantic BaseModel as metadata by converting it to dict using model_dump(). However, the implementation has a critical flaw: invalid metadata types (strings, numbers, etc.) are now silently converted to empty dict {} instead of raising a TypeError.

Key Issues:

• The else branch in the metadata validation (line 406) catches all non-dict, non-BaseModel types and silently converts them to {}
• This breaks existing error handling behavior - the original code correctly raised TypeError for invalid metadata
• The removed test test_fails_wrong_metadata verified this error behavior, but no test was added to verify BaseModel handling works correctly

Confidence Score: 1/5

• This PR is not safe to merge due to a critical logic error in metadata validation
• Score reflects a logic bug that silently converts invalid metadata to empty dict instead of raising errors, breaking existing validation behavior and removing important error feedback for users
• langfuse/openai.py requires immediate attention to fix the metadata validation logic

Important Files Changed

File Analysis

Filename	Score	Overview
langfuse/openai.py	1/5	added BaseModel handling for metadata, but else branch silently converts invalid types to empty dict instead of raising error
tests/test_openai.py	3/5	removed test for invalid metadata, but no test added to verify BaseModel metadata handling works

Sequence Diagram

sequenceDiagram
    participant User
    participant OpenAI Client
    participant _get_langfuse_data_from_kwargs
    participant Langfuse

    User->>OpenAI Client: create(metadata=BaseModel())
    OpenAI Client->>_get_langfuse_data_from_kwargs: kwargs with metadata
    _get_langfuse_data_from_kwargs->>_get_langfuse_data_from_kwargs: Check if metadata is dict
    alt metadata is BaseModel
        _get_langfuse_data_from_kwargs->>_get_langfuse_data_from_kwargs: metadata.model_dump()
    else metadata is invalid type
        _get_langfuse_data_from_kwargs->>_get_langfuse_data_from_kwargs: metadata = {} (silent conversion)
    end
    _get_langfuse_data_from_kwargs->>Langfuse: return langfuse_data with metadata
    Langfuse-->>User: tracked generation

Loading

[greptile-apps]

_{2 files reviewed, 1 comment}

_{Edit Code Review Agent Settings | Greptile}