fix: Ignore AsyncGenerator return types in function declarations

For Vertex model backend, we send response back. This doesn't work for streaming tools that the return type is AsyncGenerator. So the fix here is to ignore the return type when it's AsyncGenerator.

We can't distinguish streaming vs non-streaming tool with AsyncGenerator though as LiveRequestQueue is optional in streaming tool.

Adds an `ignore_response` option to `build_function_declaration` to skip including the return type in the function declaration. This is enabled for tools that return `AsyncGenerator`, as the model does not yet support understanding these return types, while streaming tools can still handle them. Also, removes redundant return statements in `_get_mandatory_params`.

PiperOrigin-RevId: 794392846

Author: hangfei

contributing/samples/langchain_structured_tool_agent/agent.py

@@ -15,6 +15,8 @@
 """
 This agent aims to test the Langchain tool with Langchain's StructuredTool
 """
+from __future__ import annotations
+
 from google.adk.agents.llm_agent import Agent
 from google.adk.tools.langchain_tool import LangchainTool
 from langchain.tools import tool
@@ -23,11 +25,13 @@
 
 
 async def add(x, y) -> int:
+  """Adds two numbers."""
   return x + y
 
 
 @tool
 def minus(x, y) -> int:
+  """Minus two numbers."""
   return x - y
 
 

src/google/adk/agents/llm_agent.py

@@ -109,7 +109,8 @@
 
 
 async def _convert_tool_union_to_tools(
-    tool_union: ToolUnion, ctx: ReadonlyContext
+    tool_union: ToolUnion,
+    ctx: ReadonlyContext,
 ) -> list[BaseTool]:
   if isinstance(tool_union, BaseTool):
     return [tool_union]

src/google/adk/flows/llm_flows/base_llm_flow.py

@@ -73,7 +73,7 @@ async def run_live(
       invocation_context: InvocationContext,
   ) -> AsyncGenerator[Event, None]:
     """Runs the flow using live api."""
-    llm_request = LlmRequest()
+    llm_request = LlmRequest(live_connect_config=types.LiveConnectConfig())
     event_id = Event.new_id()
 
     # Preprocess before calling the LLM.
@@ -373,7 +373,9 @@ async def _run_one_step_async(
         yield event
 
   async def _preprocess_async(
-      self, invocation_context: InvocationContext, llm_request: LlmRequest
+      self,
+      invocation_context: InvocationContext,
+      llm_request: LlmRequest,
   ) -> AsyncGenerator[Event, None]:
     from ...agents.llm_agent import LlmAgent
 

src/google/adk/flows/llm_flows/basic.py

@@ -57,30 +57,31 @@ async def run_async(
     if agent.output_schema and not agent.tools:
       llm_request.set_output_schema(agent.output_schema)
 
-    llm_request.live_connect_config.response_modalities = (
-        invocation_context.run_config.response_modalities
-    )
-    llm_request.live_connect_config.speech_config = (
-        invocation_context.run_config.speech_config
-    )
-    llm_request.live_connect_config.output_audio_transcription = (
-        invocation_context.run_config.output_audio_transcription
-    )
-    llm_request.live_connect_config.input_audio_transcription = (
-        invocation_context.run_config.input_audio_transcription
-    )
-    llm_request.live_connect_config.realtime_input_config = (
-        invocation_context.run_config.realtime_input_config
-    )
-    llm_request.live_connect_config.enable_affective_dialog = (
-        invocation_context.run_config.enable_affective_dialog
-    )
-    llm_request.live_connect_config.proactivity = (
-        invocation_context.run_config.proactivity
-    )
-    llm_request.live_connect_config.session_resumption = (
-        invocation_context.run_config.session_resumption
-    )
+    if llm_request.live_connect_config:
+      llm_request.live_connect_config.response_modalities = (
+          invocation_context.run_config.response_modalities
+      )
+      llm_request.live_connect_config.speech_config = (
+          invocation_context.run_config.speech_config
+      )
+      llm_request.live_connect_config.output_audio_transcription = (
+          invocation_context.run_config.output_audio_transcription
+      )
+      llm_request.live_connect_config.input_audio_transcription = (
+          invocation_context.run_config.input_audio_transcription
+      )
+      llm_request.live_connect_config.realtime_input_config = (
+          invocation_context.run_config.realtime_input_config
+      )
+      llm_request.live_connect_config.enable_affective_dialog = (
+          invocation_context.run_config.enable_affective_dialog
+      )
+      llm_request.live_connect_config.proactivity = (
+          invocation_context.run_config.proactivity
+      )
+      llm_request.live_connect_config.session_resumption = (
+          invocation_context.run_config.session_resumption
+      )
 
     # TODO: handle tool append here, instead of in BaseTool.process_llm_request.
 

src/google/adk/models/llm_request.py

@@ -14,6 +14,9 @@
 
 from __future__ import annotations
 
+from collections.abc import AsyncGenerator as ABCAsyncGenerator
+import inspect
+from typing import get_origin
 from typing import Optional
 
 from google.genai import types
@@ -22,6 +25,7 @@
 from pydantic import Field
 
 from ..tools.base_tool import BaseTool
+from ..tools.function_tool import FunctionTool
 
 
 def _find_tool_with_function_declarations(
@@ -66,13 +70,13 @@ class LlmRequest(BaseModel):
   config: types.GenerateContentConfig = Field(
       default_factory=types.GenerateContentConfig
   )
-  live_connect_config: types.LiveConnectConfig = Field(
-      default_factory=types.LiveConnectConfig
-  )
   """Additional config for the generate content request.
 
   tools in generate_content_config should not be set.
   """
+  live_connect_config: Optional[types.LiveConnectConfig] = None
+  """Live connection config.
+  """
   tools_dict: dict[str, BaseTool] = Field(default_factory=dict, exclude=True)
   """The tools dictionary."""
 
@@ -99,7 +103,23 @@ def append_tools(self, tools: list[BaseTool]) -> None:
       return
     declarations = []
     for tool in tools:
-      declaration = tool._get_declaration()
+      if self.live_connect_config is not None:
+        # ignore response for tools that returns AsyncGenerator that the model
+        # can't understand yet even though the model can't handle it, streaming
+        # tools can handle it.
+        # to check type, use typing.collections.abc.AsyncGenerator and not
+        # typing.AsyncGenerator
+        is_async_generator_return = False
+        if isinstance(tool, FunctionTool):
+          signature = inspect.signature(tool.func)
+          is_async_generator_return = (
+              get_origin(signature.return_annotation) is ABCAsyncGenerator
+          )
+        declaration = tool._get_declaration(
+            ignore_return_declaration=is_async_generator_return
+        )
+      else:
+        declaration = tool._get_declaration()
       if declaration:
         declarations.append(declaration)
         self.tools_dict[tool.name] = tool

src/google/adk/tools/_automatic_function_calling_util.py

@@ -195,6 +195,7 @@ def build_function_declaration(
     func: Union[Callable, BaseModel],
     ignore_params: Optional[list[str]] = None,
     variant: GoogleLLMVariant = GoogleLLMVariant.GEMINI_API,
+    ignore_return_declaration: bool = False,
 ) -> types.FunctionDeclaration:
   signature = inspect.signature(func)
   should_update_signature = False
@@ -232,9 +233,11 @@ def build_function_declaration(
       new_func.__annotations__ = func.__annotations__
 
   return (
-      from_function_with_options(func, variant)
+      from_function_with_options(func, variant, ignore_return_declaration)
       if not should_update_signature
-      else from_function_with_options(new_func, variant)
+      else from_function_with_options(
+          new_func, variant, ignore_return_declaration
+      )
   )
 
 
@@ -293,6 +296,7 @@ def build_function_declaration_util(
 def from_function_with_options(
     func: Callable,
     variant: GoogleLLMVariant = GoogleLLMVariant.GEMINI_API,
+    ignore_return_declaration: bool = False,
 ) -> 'types.FunctionDeclaration':
 
   parameters_properties = {}
@@ -324,7 +328,8 @@ def from_function_with_options(
             declaration.parameters
         )
     )
-  if variant == GoogleLLMVariant.GEMINI_API:
+
+  if variant == GoogleLLMVariant.GEMINI_API or ignore_return_declaration:
     return declaration
 
   return_annotation = inspect.signature(func).return_annotation

src/google/adk/tools/agent_tool.py

@@ -15,6 +15,7 @@
 from __future__ import annotations
 
 from typing import Any
+from typing import Optional
 from typing import TYPE_CHECKING
 
 from google.genai import types
@@ -61,7 +62,9 @@ def populate_name(cls, data: Any) -> Any:
     return data
 
   @override
-  def _get_declaration(self) -> types.FunctionDeclaration:
+  def _get_declaration(
+      self, ignore_return_declaration: bool = False
+  ) -> Optional[types.FunctionDeclaration]:
     from ..agents.llm_agent import LlmAgent
     from ..utils.variant_utils import GoogleLLMVariant
 

src/google/adk/tools/application_integration_tool/integration_connector_tool.py

@@ -20,7 +20,7 @@
 from typing import Optional
 from typing import Union
 
-from google.genai.types import FunctionDeclaration
+from google.genai import types
 from typing_extensions import override
 
 from ...auth.auth_credential import AuthCredential
@@ -115,7 +115,9 @@ def __init__(
     self._auth_credential = auth_credential
 
   @override
-  def _get_declaration(self) -> FunctionDeclaration:
+  def _get_declaration(
+      self, ignore_return_declaration: bool = False
+  ) -> Optional[types.FunctionDeclaration]:
     """Returns the function declaration in the Gemini Schema format."""
     schema_dict = self._rest_api_tool._operation_parser.get_json_schema()
     for field in self.EXCLUDE_FIELDS:
@@ -126,7 +128,7 @@ def _get_declaration(self) -> FunctionDeclaration:
         schema_dict['required'].remove(field)
 
     parameters = _to_gemini_schema(schema_dict)
-    function_decl = FunctionDeclaration(
+    function_decl = types.FunctionDeclaration(
         name=self.name, description=self.description, parameters=parameters
     )
     return function_decl

src/google/adk/tools/base_tool.py

@@ -78,7 +78,9 @@ def __init__(
     self.is_long_running = is_long_running
     self.custom_metadata = custom_metadata
 
-  def _get_declaration(self) -> Optional[types.FunctionDeclaration]:
+  def _get_declaration(
+      self, ignore_return_declaration: bool = False
+  ) -> Optional[types.FunctionDeclaration]:
     """Gets the OpenAPI specification of this tool in the form of a FunctionDeclaration.
 
     NOTE:

src/google/adk/tools/crewai_tool.py

@@ -14,6 +14,8 @@
 
 from __future__ import annotations
 
+from typing import Optional
+
 from google.genai import types
 from typing_extensions import override
 
@@ -62,7 +64,9 @@ def __init__(self, tool: CrewaiBaseTool, *, name: str, description: str):
       self.description = tool.description
 
   @override
-  def _get_declaration(self) -> types.FunctionDeclaration:
+  def _get_declaration(
+      self, ignore_return_declaration: bool = False
+  ) -> Optional[types.FunctionDeclaration]:
     """Build the function declaration for the tool."""
     function_declaration = _automatic_function_calling_util.build_function_declaration_for_params_for_crewai(
         False,