fix(google-genai): remap finish_reason to tool_calls when response contains tool calls

Author: SyedShahmeerAli12

integrations/google_genai/src/haystack_integrations/components/generators/google_genai/chat/utils.py

@@ -559,10 +559,16 @@ def _convert_google_genai_response_to_chatmessage(response: types.GenerateConten
 
     usage.update(_convert_usage_metadata_to_serializable(usage_metadata))
 
+    # Remap finish_reason to "tool_calls" when tool calls are present, since Google GenAI returns
+    # "STOP" for both normal completions and tool calls (no dedicated FUNCTION_CALL finish reason).
+    mapped_finish_reason = FINISH_REASON_MAPPING.get(finish_reason or "")
+    if mapped_finish_reason == "stop" and tool_calls:
+        mapped_finish_reason = "tool_calls"
+
     # Create meta with reasoning content and thought signatures if available
     meta: dict[str, Any] = {
         "model": model,
-        "finish_reason": FINISH_REASON_MAPPING.get(finish_reason or ""),
+        "finish_reason": mapped_finish_reason,
         "usage": usage,
     }
 
@@ -675,13 +681,19 @@ def _convert_google_chunk_to_streaming_chunk(
     # Determine the effective content: tool_calls and reasoning take priority.
     effective_content = "" if tool_calls or reasoning else content
 
+    # Remap finish_reason to "tool_calls" when tool calls are present, since Google GenAI returns
+    # "STOP" for both normal completions and tool calls (no dedicated FUNCTION_CALL finish reason).
+    mapped_finish_reason = FINISH_REASON_MAPPING.get(finish_reason or "")
+    if mapped_finish_reason == "stop" and tool_calls:
+        mapped_finish_reason = "tool_calls"
+
     return StreamingChunk(
         content=effective_content,
         tool_calls=tool_calls,
         component_info=component_info,
         index=index,
         start=start,
-        finish_reason=FINISH_REASON_MAPPING.get(finish_reason or ""),
+        finish_reason=mapped_finish_reason,
         meta=meta,
         reasoning=reasoning,
     )

integrations/google_genai/tests/test_chat_generator.py

@@ -401,7 +401,7 @@ def test_live_run_with_tools_streaming(self, tools):
 
         assert isinstance(tool_message, ChatMessage), "Tool message is not a ChatMessage instance"
         assert ChatMessage.is_from(tool_message, ChatRole.ASSISTANT), "Tool message is not from the assistant"
-        assert tool_message.meta["finish_reason"] == "stop"
+        assert tool_message.meta["finish_reason"] == "tool_calls"
 
         tool_call = tool_message.tool_calls[0]
         assert tool_call.tool_name == "weather"
@@ -425,7 +425,7 @@ def test_live_run_with_toolset(self, tools):
         # Google Gen AI (gemini-2.5-flash and gemini-2.5-pro-preview-05-06) does not provide ids for tool calls although
         # it is in the response schema, revisit in future to see if there are changes and id is provided
         # assert tool_call.id is not None, "Tool call has no id"
-        assert message.meta["finish_reason"] == "stop"
+        assert message.meta["finish_reason"] == "tool_calls"
 
         assert tool_call.tool_name == "weather"
         assert tool_call.arguments == {"city": "Paris"}
@@ -559,7 +559,7 @@ def test_live_run_with_mixed_tools(self):
             assert tool_call.tool_name in ["weather", "population"]
             assert "city" in tool_call.arguments
             assert tool_call.arguments["city"] in ["Paris", "Berlin"]
-            assert first_reply.meta["finish_reason"] == "stop"
+            assert first_reply.meta["finish_reason"] == "tool_calls"
 
         # Mock the response we'd get from ToolInvoker
         tool_result_messages = []
@@ -806,7 +806,7 @@ async def test_live_run_async_with_tools(self, tools):
         assert len(tool_message.tool_calls) == 1, "Tool message has multiple tool calls"
         assert tool_message.tool_calls[0].tool_name == "weather"
         assert tool_message.tool_calls[0].arguments == {"city": "Paris"}
-        assert tool_message.meta["finish_reason"] == "stop"
+        assert tool_message.meta["finish_reason"] == "tool_calls"
 
     async def test_live_run_async_with_thinking(self):
         """

integrations/google_genai/tests/test_chat_generator_utils.py

@@ -286,7 +286,7 @@ def test_convert_google_chunk_to_streaming_chunk_tool_call(self, monkeypatch):
         assert chunk.tool_calls[0].tool_name == "weather"
         assert chunk.tool_calls[0].arguments == '{"city": "Paris"}'
         assert chunk.tool_calls[0].id == "call_123"
-        assert chunk.finish_reason == "stop"
+        assert chunk.finish_reason == "tool_calls"
         assert chunk.index == 0
         assert "received_at" in chunk.meta
         assert chunk.component_info == component_info
@@ -332,7 +332,7 @@ def test_convert_google_chunk_to_streaming_chunk_mixed_content(self, monkeypatch
         assert len(chunk.tool_calls) == 1
         assert chunk.tool_calls[0].tool_name == "weather"
         assert chunk.tool_calls[0].arguments == '{"city": "London"}'
-        assert chunk.finish_reason == "stop"
+        assert chunk.finish_reason == "tool_calls"
         assert chunk.component_info == component_info
 
     def test_convert_google_chunk_to_streaming_chunk_empty_parts(self, monkeypatch):
@@ -508,7 +508,7 @@ def test_convert_google_chunk_to_streaming_chunk_real_example(self, monkeypatch)
         assert streaming_chunk.content == ""
         assert streaming_chunk.tool_calls is not None
         assert len(streaming_chunk.tool_calls) == 6
-        assert streaming_chunk.finish_reason == "stop"
+        assert streaming_chunk.finish_reason == "tool_calls"
         assert streaming_chunk.index == 2
         assert "received_at" in streaming_chunk.meta
         assert streaming_chunk.meta["model"] == "gemini-2.5-flash"