fix(gemini): address remaining greptile review comments

- google_code_assist: log unrecognized optional_params keys for visibility
- gemini/authenticator: document timeout overshoot in OAuth loopback loop
- gemini/fallback_handler: take primary_call as Callable so async sync-setup errors
  are caught by the fallback try/except (symmetric with sync path)
- gemini/google_genai: pre-fetch OAuth via asyncio.to_thread in async
  get_auth_token_and_url to avoid blocking the event loop
- tests/google_code_assist: add missing __init__.py for parallel pytest discovery
- tests/fallback_handler: pass deferred coroutine factory to async fallback wrapper

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

litellm/llms/gemini/authenticator.py

@@ -257,6 +257,7 @@ class GeminiAuthenticator:
         deadline = time.monotonic() + loopback_timeout_s
         server.timeout = 1.0
         try:
+            # Note: deadline is checked after handle_request() returns, so the loop can overshoot by up to server.timeout (1s) — acceptable for a one-shot CLI login.
             while auth_code is None and error is None:
                 server.handle_request()
                 if time.monotonic() >= deadline:

litellm/llms/gemini/fallback_handler.py

@@ -8,7 +8,7 @@ _google_code_assist_chat = get_google_code_assist_chat()
 
 
 async def run_gemini_acompletion_with_code_assist_fallback(
-    primary_call: Awaitable[Any],
+    primary_call: Callable[[], Awaitable[Any]],
     fallback_kwargs: Dict[str, Any],
     auto_fallback_to_google_code_assist: bool = False,
 ) -> Any:
@@ -17,7 +17,7 @@ async def run_gemini_acompletion_with_code_assist_fallback(
     OAuth scope is insufficient.
     """
     try:
-        return await primary_call
+        return await primary_call()
     except Exception as e:
         if not auto_fallback_to_google_code_assist:
             raise e

litellm/llms/gemini/google_genai/transformation.py

@@ -2,6 +2,7 @@
 Transformation for Calling Google models in their native format.
 """
 
+import asyncio
 from typing import TYPE_CHECKING, Any, Dict, List, Literal, Optional, Tuple, Union, cast
 
 import httpx
@@ -11,6 +12,7 @@ from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLogging
 from litellm.llms.base_llm.google_genai.transformation import (
     BaseGoogleGenAIGenerateContentConfig,
 )
+from litellm.llms.gemini.common_utils import get_gemini_oauth_token
 from litellm.llms.vertex_ai.gemini.vertex_and_google_ai_studio_gemini import VertexLLM
 from litellm.types.router import GenericLiteLLMParams
 
@@ -200,6 +202,7 @@ class GoogleGenAIConfig(BaseGoogleGenAIGenerateContentConfig, VertexLLM):
         stream: bool,
         api_base: Optional[str],
         litellm_params: dict,
+        gemini_auth_data: Optional[dict] = None,
     ) -> Tuple[dict, str]:
         """
         Build final headers and API URL from auth components.
@@ -217,6 +220,7 @@ class GoogleGenAIConfig(BaseGoogleGenAIGenerateContentConfig, VertexLLM):
             custom_llm_provider=self.custom_llm_provider,
             api_base=api_base,
             should_use_v1beta1_features=True,
+            gemini_auth_data=gemini_auth_data,
         )
 
         headers = self.validate_environment(
@@ -291,6 +295,12 @@ class GoogleGenAIConfig(BaseGoogleGenAIGenerateContentConfig, VertexLLM):
             custom_llm_provider=self.custom_llm_provider,
         )
 
+        gemini_auth_data: Optional[dict] = None
+        if self.custom_llm_provider == "gemini":
+            gemini_api_key = self._get_google_ai_studio_api_key(dict(litellm_params))
+            if gemini_api_key is None:
+                gemini_auth_data = await asyncio.to_thread(get_gemini_oauth_token)
+
         return self._build_final_headers_and_url(
             model=model,
             auth_header=_auth_header,
@@ -300,6 +310,7 @@ class GoogleGenAIConfig(BaseGoogleGenAIGenerateContentConfig, VertexLLM):
             stream=stream,
             api_base=api_base,
             litellm_params=litellm_params,
+            gemini_auth_data=gemini_auth_data,
         )
 
     def transform_generate_content_request(

litellm/llms/google_code_assist/transformation.py

@@ -108,6 +108,16 @@ class GoogleCodeAssistConfig(VertexGeminiConfig):
                 )
                 generation_config[key] = value
 
+        # Log unmapped/unrecognized caller keys so users have visibility (do not auto-forward —
+        # they may not be valid Gemini fields).
+        for key in optional_params:
+            if key not in base_params and key not in generation_config:
+                verbose_logger.debug(
+                    "google_code_assist: dropping unrecognized optional_param '%s' "
+                    "(not produced by VertexGeminiConfig.map_openai_params)",
+                    key,
+                )
+
         vertex_request = {
             "contents": contents,
             "session_id": litellm_params.get("session_id", str(uuid.uuid4())),

litellm/main.py

@@ -3447,7 +3447,7 @@ def completion(  # type: ignore # noqa: PLR0915
             new_params = safe_deep_copy(optional_params or {})
             if acompletion is True:
                 response = run_gemini_acompletion_with_code_assist_fallback(
-                    primary_call=vertex_chat_completion.completion(  # type: ignore
+                    primary_call=lambda: vertex_chat_completion.completion(  # type: ignore
                         model=model,
                         messages=messages,
                         model_response=model_response,

tests/test_litellm/llms/gemini/test_fallback_handler.py

@@ -66,7 +66,7 @@ async def test_run_gemini_acompletion_with_code_assist_fallback_enabled():
     ):
         mock_acompletion.return_value = "fallback-ok"
         result = await run_gemini_acompletion_with_code_assist_fallback(
-            primary_call=_raise_scope_error(),
+            primary_call=_raise_scope_error,
             fallback_kwargs={},
             auto_fallback_to_google_code_assist=True,
         )