Personal AI, On Personal Devices
Python
fix(cloud): surface provider errors without choices (#974)
Haotian Zhengf714ce03c4b4
diff --git a/src/openjarvis/engine/cloud.py b/src/openjarvis/engine/cloud.pyindex 89e9102d..85a55387 100644--- a/src/openjarvis/engine/cloud.py+++ b/src/openjarvis/engine/cloud.py@@ -224,6 +224,26 @@ def estimate_cost(model: str, prompt_tokens: int, completion_tokens: int) -> floreturn input_cost + output_cost+def _first_choice_or_raise(resp: Any, *, provider: str, model: str) -> Any:+ """Return the first completion choice or surface the provider error."""+ choices = getattr(resp, "choices", None)+ if choices:+ return choices[0]++ error = getattr(resp, "error", None)+ if isinstance(error, dict):+ detail = str(error.get("message") or error)+ elif error is not None:+ detail = str(getattr(error, "message", None) or error)+ else:+ detail = ""++ message = f"{provider} returned no choices for model {model!r}"+ raise EngineConnectionError(+ message + (f": {detail}" if detail else " and no error message")+ )++def _serialize_anthropic_block(block: Any) -> Dict[str, Any]:"""Turn an Anthropic content block into a JSON-safe dict for tracing.@@ -625,7 +645,7 @@ class CloudEngine(InferenceEngine):else:raiseelapsed = time.monotonic() - t0- choice = resp.choices[0]+ choice = _first_choice_or_raise(resp, provider="OpenAI", model=model)usage = resp.usageprompt_tokens = usage.prompt_tokens if usage else 0completion_tokens = usage.completion_tokens if usage else 0@@ -968,7 +988,7 @@ class CloudEngine(InferenceEngine):t0 = time.monotonic()resp = self._openrouter_client.chat.completions.create(**create_kwargs)elapsed = time.monotonic() - t0- choice = resp.choices[0]+ choice = _first_choice_or_raise(resp, provider="OpenRouter", model=actual_model)usage = resp.usageprompt_tokens = usage.prompt_tokens if usage else 0completion_tokens = usage.completion_tokens if usage else 0@@ -1023,7 +1043,7 @@ class CloudEngine(InferenceEngine):t0 = time.monotonic()resp = self._minimax_client.chat.completions.create(**create_kwargs)elapsed = time.monotonic() - t0- choice = resp.choices[0]+ choice = _first_choice_or_raise(resp, provider="MiniMax", model=model)usage = resp.usageprompt_tokens = usage.prompt_tokens if usage else 0completion_tokens = usage.completion_tokens if usage else 0@@ -1073,7 +1093,7 @@ class CloudEngine(InferenceEngine):t0 = time.monotonic()resp = self._deepseek_client.chat.completions.create(**create_kwargs)elapsed = time.monotonic() - t0- choice = resp.choices[0]+ choice = _first_choice_or_raise(resp, provider="DeepSeek", model=model)usage = resp.usageprompt_tokens = usage.prompt_tokens if usage else 0completion_tokens = usage.completion_tokens if usage else 0diff --git a/tests/engine/test_cloud.py b/tests/engine/test_cloud.pyindex 6077e558..01fb280a 100644--- a/tests/engine/test_cloud.py+++ b/tests/engine/test_cloud.py@@ -118,6 +118,58 @@ class TestCloudEngineGenerate:assert result["usage"]["completion_tokens"] == 8+class TestCloudEngineEmptyChoices:+ @pytest.mark.parametrize(+ ("model", "client_attr", "provider"),+ [+ ("gpt-4o", "_openai_client", "OpenAI"),+ ("openrouter/test-model", "_openrouter_client", "OpenRouter"),+ ("MiniMax-M3", "_minimax_client", "MiniMax"),+ ("deepseek-v4-flash", "_deepseek_client", "DeepSeek"),+ ],+ )+ def test_empty_choices_surfaces_provider_error(+ self,+ monkeypatch: pytest.MonkeyPatch,+ model: str,+ client_attr: str,+ provider: str,+ ) -> None:+ for env_var in (+ "OPENAI_API_KEY",+ "ANTHROPIC_API_KEY",+ "GOOGLE_API_KEY",+ "OPENROUTER_API_KEY",+ "MINIMAX_API_KEY",+ "DEEPSEEK_API_KEY",+ "OPENAI_CODEX_API_KEY",+ "OPENAI_CODEX_BASE_URL",+ ):+ monkeypatch.delenv(env_var, raising=False)++ fake_resp = SimpleNamespace(+ choices=None,+ error={"message": f"{provider} request failed"},+ usage=None,+ model=model,+ )+ fake_client = mock.MagicMock()+ fake_client.chat.completions.create.return_value = fake_resp++ EngineRegistry.register_value("cloud", CloudEngine)+ engine = CloudEngine()+ setattr(engine, client_attr, fake_client)++ with pytest.raises(EngineConnectionError) as exc_info:+ engine.generate(+ [Message(role=Role.USER, content="Hi")],+ model=model,+ )++ assert provider in str(exc_info.value)+ assert f"{provider} request failed" in str(exc_info.value)++class TestOpenAIUnsupportedTemperatureRetry:"""Regression for #426.