kernel_function_from_method.py

# Copyright (c) Microsoft. All rights reserved.

import inspect
import logging
from collections.abc import Callable
from inspect import isasyncgen, isasyncgenfunction, isawaitable, iscoroutinefunction, isgenerator, isgeneratorfunction
from typing import Any

from pydantic import ValidationError

from semantic_kernel.exceptions import FunctionExecutionException, FunctionInitializationError
from semantic_kernel.filters.functions.function_invocation_context import FunctionInvocationContext
from semantic_kernel.functions.function_result import FunctionResult
from semantic_kernel.functions.kernel_function import KernelFunction
from semantic_kernel.functions.kernel_function_metadata import KernelFunctionMetadata
from semantic_kernel.functions.kernel_parameter_metadata import KernelParameterMetadata

logger: logging.Logger = logging.getLogger(__name__)


class KernelFunctionFromMethod(KernelFunction):
    """Semantic Kernel Function from a method."""

    method: Callable[..., Any]
    stream_method: Callable[..., Any] | None = None

    def __init__(
        self,
        method: Callable[..., Any],
        plugin_name: str | None = None,
        stream_method: Callable[..., Any] | None = None,
        parameters: list[KernelParameterMetadata] | None = None,
        return_parameter: KernelParameterMetadata | None = None,
        additional_metadata: dict[str, Any] | None = None,
    ) -> None:
        """Initializes a new instance of the KernelFunctionFromMethod class.

        Args:
            method (Callable[..., Any]): The method to be called
            plugin_name (str | None): The name of the plugin
            stream_method (Callable[..., Any] | None): The stream method for the function
            parameters (list[KernelParameterMetadata] | None): The parameters of the function
            return_parameter (KernelParameterMetadata | None): The return parameter of the function
            additional_metadata (dict[str, Any] | None): Additional metadata for the function
        """
        if method is None:
            raise FunctionInitializationError("Method cannot be `None`")

        if not hasattr(method, "__kernel_function__") or method.__kernel_function__ is None:
            raise FunctionInitializationError("Method is not a Kernel function")

        # all these fields are created when the kernel function decorator is used,
        # so no need to check before using, will raise an exception if not set
        function_name = method.__kernel_function_name__  # type: ignore
        description = method.__kernel_function_description__  # type: ignore
        if parameters is None:
            parameters = [KernelParameterMetadata(**param) for param in method.__kernel_function_parameters__]  # type: ignore
        if return_parameter is None:
            return_parameter = KernelParameterMetadata(
                name="return",
                description=method.__kernel_function_return_description__,  # type: ignore
                default_value=None,
                type_=method.__kernel_function_return_type__,  # type: ignore
                type_object=method.__kernel_function_return_type_object__,  # type: ignore
                is_required=method.__kernel_function_return_required__,  # type: ignore
            )

        try:
            metadata = KernelFunctionMetadata(
                name=function_name,
                description=description,
                parameters=parameters,
                return_parameter=return_parameter,
                is_prompt=False,
                is_asynchronous=isasyncgenfunction(method) or iscoroutinefunction(method),
                plugin_name=plugin_name,
                additional_properties=additional_metadata if additional_metadata is not None else {},
            )
        except ValidationError as exc:
            # reraise the exception to clarify it comes from KernelFunction init
            raise FunctionInitializationError("Failed to create KernelFunctionMetadata") from exc

        args: dict[str, Any] = {
            "metadata": metadata,
            "method": method,
            "stream_method": (
                stream_method
                if stream_method is not None
                else method
                if isasyncgenfunction(method) or isgeneratorfunction(method)
                else None
            ),
        }

        super().__init__(**args)

    async def _invoke_internal(
        self,
        context: FunctionInvocationContext,
    ) -> None:
        """Invoke the function with the given arguments."""
        function_arguments = self.gather_function_parameters(context)
        result = self.method(**function_arguments)
        if isasyncgen(result):
            result = [x async for x in result]
        elif isawaitable(result):
            result = await result
        elif isgenerator(result):
            result = list(result)
        if not isinstance(result, FunctionResult):
            result = FunctionResult(
                function=self.metadata,
                value=result,
                metadata={"arguments": context.arguments, "used_arguments": function_arguments},
            )
        context.result = result

    async def _invoke_internal_stream(self, context: FunctionInvocationContext) -> None:
        if self.stream_method is None:
            raise NotImplementedError("Stream method not implemented")
        function_arguments = self.gather_function_parameters(context)
        context.result = FunctionResult(function=self.metadata, value=self.stream_method(**function_arguments))

    def _parse_parameter(self, value: Any, param_type: Any) -> Any:
        """Parses the value into the specified param_type, including handling lists of types."""
        if isinstance(param_type, type) and hasattr(param_type, "model_validate"):
            try:
                return param_type.model_validate(value)
            except Exception as exc:
                raise FunctionExecutionException(
                    f"Parameter is expected to be parsed to {param_type} but is not."
                ) from exc
        elif hasattr(param_type, "__origin__") and param_type.__origin__ is list:
            if isinstance(value, list):
                item_type = param_type.__args__[0]
                return [self._parse_parameter(item, item_type) for item in value]
            raise FunctionExecutionException(f"Expected a list for {param_type}, but got {type(value)}")
        else:
            try:
                if isinstance(value, dict) and hasattr(param_type, "__init__"):
                    return param_type(**value)
                return param_type(value)
            except Exception as exc:
                raise FunctionExecutionException(
                    f"Parameter is expected to be parsed to {param_type} but is not."
                ) from exc

    def gather_function_parameters(self, context: FunctionInvocationContext) -> dict[str, Any]:
        """Gathers the function parameters from the arguments."""
        function_arguments: dict[str, Any] = {}
        for param in self.parameters:
            if param.name is None:
                raise FunctionExecutionException("Parameter name cannot be None")
            if param.name == "kernel":
                function_arguments[param.name] = context.kernel
                continue
            if param.name == "service":
                function_arguments[param.name] = context.kernel.select_ai_service(self, context.arguments)[0]
                continue
            if param.name == "execution_settings":
                function_arguments[param.name] = context.kernel.select_ai_service(self, context.arguments)[1]
                continue
            if param.name == "arguments":
                function_arguments[param.name] = context.arguments
                continue
            if param.name in context.arguments:
                value: Any = context.arguments[param.name]
                if (
                    param.type_
                    and "," not in param.type_
                    and param.type_object
                    and param.type_object is not inspect._empty
                ):
                    try:
                        value = self._parse_parameter(value, param.type_object)
                    except Exception as exc:
                        raise FunctionExecutionException(
                            f"Parameter {param.name} is expected to be parsed to {param.type_object} but is not."
                        ) from exc
                function_arguments[param.name] = value
                continue
            if param.is_required:
                raise FunctionExecutionException(
                    f"Parameter {param.name} is required but not provided in the arguments."
                )
            logger.debug(f"Parameter {param.name} is not provided, using default value {param.default_value}")
        return function_arguments