core[patch]: Support injected tool args that are arbitrary types (#27045

)

This adds support for inject tool args that are arbitrary types when
used with pydantic 2.

We'll need to add similar logic on the v1 path, and potentially mirror
the config from the original model when we're doing the subset.

libs/core/langchain_core/utils/pydantic.py

@@ -266,7 +266,7 @@ def _create_subset_model_v2(
     fn_description: Optional[str] = None,
 ) -> type[pydantic.BaseModel]:
     """Create a pydantic model with a subset of the model fields."""
-    from pydantic import create_model
+    from pydantic import ConfigDict, create_model
     from pydantic.fields import FieldInfo
 
     descriptions_ = descriptions or {}
@@ -278,7 +278,10 @@ def _create_subset_model_v2(
         if field.metadata:
             field_info.metadata = field.metadata
         fields[field_name] = (field.annotation, field_info)
-    rtn = create_model(name, **fields)  # type: ignore
+
+    rtn = create_model(  # type: ignore
+        name, **fields, __config__=ConfigDict(arbitrary_types_allowed=True)
+    )
 
     # TODO(0.3): Determine if there is a more "pydantic" way to preserve annotations.
     # This is done to preserve __annotations__ when working with pydantic 2.x

libs/core/tests/unit_tests/test_tools.py

@@ -2090,3 +2090,18 @@ class FooSchema(BaseModel):
 
     with pytest.raises(NotImplementedError):
         assert tool.invoke("hello") == "hello"
+
+
+def test_injected_arg_with_complex_type() -> None:
+    """Test that an injected tool arg can be a complex type."""
+
+    class Foo:
+        def __init__(self) -> None:
+            self.value = "bar"
+
+    @tool
+    def injected_tool(x: int, foo: Annotated[Foo, InjectedToolArg]) -> str:
+        """Tool that has an injected tool arg."""
+        return foo.value
+
+    assert injected_tool.invoke({"x": 5, "foo": Foo()}) == "bar"  # type: ignore