refactor: let TypeAdapter handle all types uniformly, skip framework params

Remove the Pydantic model special case from _preprocess_args so
TypeAdapter handles primitives, enums, Pydantic models, and containers
in one path. Skip framework-managed params (_ignore_params) to avoid
validating injected values like credential=None.

Author: yuvrajangadsingh

src/google/adk/tools/function_tool.py

@@ -132,7 +132,11 @@ def _preprocess_args(
     validation_errors = []
 
     for param_name, param in signature.parameters.items():
-      if param_name not in args or param.annotation is inspect.Parameter.empty:
+      if (
+          param_name not in args
+          or param.annotation is inspect.Parameter.empty
+          or param_name in self._ignore_params
+      ):
         continue
 
       target_type = param.annotation
@@ -150,27 +154,8 @@ def _preprocess_args(
       if args[param_name] is None and is_optional:
         continue
 
-      # Pydantic models: keep existing graceful-failure behavior
-      if inspect.isclass(target_type) and issubclass(
-          target_type, pydantic.BaseModel
-      ):
-        if not isinstance(args[param_name], target_type):
-          try:
-            converted_args[param_name] = target_type.model_validate(
-                args[param_name]
-            )
-          except Exception as e:
-            logger.warning(
-                "Failed to convert argument '%s' to Pydantic model %s: %s",
-                param_name,
-                target_type.__name__,
-                e,
-            )
-        continue
-
-      # Validate and coerce all other annotated types using TypeAdapter.
-      # This handles primitives (int, float, str, bool), enums, and
-      # container types (list[int], dict[str, float], etc.).
+      # Validate and coerce all annotated types using TypeAdapter.
+      # This handles primitives, enums, Pydantic models, and container types.
       try:
         adapter = pydantic.TypeAdapter(target_type)
         converted_args[param_name] = adapter.validate_python(

tests/unittests/tools/test_function_tool_pydantic.py

@@ -184,17 +184,18 @@ def test_preprocess_args_with_mixed_types():
   assert processed_args["user"].age == 40
 
 
-def test_preprocess_args_with_invalid_data_graceful_failure():
-  """Test _preprocess_args handles invalid data gracefully."""
+def test_preprocess_args_with_invalid_data_returns_error():
+  """Test _preprocess_args returns validation error for invalid Pydantic data."""
   tool = FunctionTool(sync_function_with_pydantic_model)
 
   # Invalid data that can't be converted to UserModel
   input_args = {"user": "invalid_string"}  # string instead of dict/model
 
-  processed_args, _ = tool._preprocess_args(input_args)
+  _, errors = tool._preprocess_args(input_args)
 
-  # Should keep original value when conversion fails
-  assert processed_args["user"] == "invalid_string"
+  # Should return a validation error for the LLM to self-correct
+  assert len(errors) == 1
+  assert "user" in errors[0]
 
 
 def test_preprocess_args_with_non_pydantic_parameters():