Callable classes can't be registered as tools as takes-RunContext detection fails

[bitnahian]

Initial Checks

• I confirm that I'm using the latest version of Pydantic AI
• I confirm that I searched for my issue in https://github.com/pydantic/pydantic-ai/issues before opening this issue

Description

Consider the example code presented.

Ideally, this should return true so that we can support Callable Classes.

Feel free to reject this if it's an anti-pattern.

Example Code

from pydantic_ai import RunContext
from pydantic_ai._function_schema import _takes_ctx


# Create a concrete implementation for testing
class TestProcessor:
    def __call__(self, ctx: RunContext, messages: list):
        return messages

    def process(self, messages):
        return messages

    def post_process(self, messages):
        return messages


if __name__ == "__main__":
    processor = TestProcessor()

    result = _takes_ctx(processor)
    print(f"_takes_ctx result: {result}") # returns False

Python, Pydantic AI & LLM client version

Python 3.11.12
pydantic-ai 1.0.6

[DouweM]

@bitnahian I agree we should support that. If you have a fix in mind, can you submit a PR please?

[bitnahian]

Nice. I'll try to have something up soon.

[bitnahian]

Hey @DouweM, if I'm being honest, I opened up this issue because I was trying to implement Callable Class history processors and not tools. Although, I did have a crack at it but implementing this for tools is like opening another can of worms.

For instance, most references to tool functions rely on the Callable type, e.g.

def function_schema(  # noqa: C901
    function: Callable[..., Any], # <---- this assumes 
    schema_generator: type[GenerateJsonSchema],
    takes_ctx: bool | None = None,
    docstring_format: DocstringFormat = 'auto',
    require_parameter_descriptions: bool = False,
) -> FunctionSchema:
    """Build a Pydantic validator and JSON schema from a tool function.

    Args:
        function: The function to build a validator and JSON schema for.
        takes_ctx: Whether the function takes a `RunContext` first argument.
        docstring_format: The docstring format to use.
        require_parameter_descriptions: Whether to require descriptions for all tool function parameters.
        schema_generator: The JSON schema generator class to use.

    Returns:
        A `FunctionSchema` instance.
    """
    if takes_ctx is None:
        takes_ctx = _takes_ctx(function)

    config = ConfigDict(title=function.__name__, use_attribute_docstrings=True) # <---- assumes a user defined function
    config_wrapper = ConfigWrapper(config)
    gen_schema = _generate_schema.GenerateSchema(config_wrapper)
    errors: list[str] = []

Callable classes don't have the __name__ attribute. Surprisingly, type checkers like pyright and mypy don't complain about this either. See python/mypy#14392

Anyway, vast majority of the code in pydantic-ai relies on the __name__ attribute being available https://github.com/search?q=repo%3Apydantic%2Fpydantic-ai%20function.__name__&type=code

I could implement this as part of the current changes I have or we could scope this to only updating the _takes_ctx implementation first as a building block for supporting Callable classes.

[DouweM]

@bitnahian Feel free to implement this just for history processors to start.