fix: ignore empty SSE data frames in OpenAI-compatible and Google providers

src/tau_ai/google.py

@@ -500,7 +500,8 @@ def _parse_sse_line(line: str) -> str | None:
     line = line.strip()
     if not line or not line.startswith("data:"):
         return None
-    return line.removeprefix("data:").strip()
+    data = line.removeprefix("data:").strip()
+    return data or None
 
 
 def _loads_object(value: str) -> dict[str, JSONValue] | None:

src/tau_ai/openai_compatible.py

@@ -1259,7 +1259,8 @@ def _parse_sse_line(line: str) -> str | None:
     line = line.strip()
     if not line or not line.startswith("data:"):
         return None
-    return line.removeprefix("data:").strip()
+    data = line.removeprefix("data:").strip()
+    return data or None
 
 
 def _loads_object(value: str) -> dict[str, JSONValue] | None:

tests/test_sse_empty_data_frames.py

@@ -0,0 +1,87 @@
+import json
+from collections.abc import AsyncIterator
+
+import httpx
+import pytest
+
+from tau_agent import UserMessage
+from tau_ai import AssistantDoneEvent, OpenAICompatibleConfig, OpenAICompatibleProvider
+from tau_ai.google import GoogleGenerativeAIProvider
+from tau_ai.mistral import MistralConversationsProvider
+
+
+def _sse(*frames: str) -> str:
+    return "".join(f"{frame}\n\n" for frame in frames)
+
+
+def _chat_chunk(**delta: object) -> str:
+    return "data: " + json.dumps({"choices": [{"delta": delta}]})
+
+
+def _google_chunk(text: str, finish: str | None = None) -> str:
+    candidate: dict[str, object] = {"content": {"parts": [{"text": text}]}}
+    if finish is not None:
+        candidate["finishReason"] = finish
+    return "data: " + json.dumps({"candidates": [candidate]})
+
+
+def _mistral_chunk(**delta: object) -> str:
+    return "data: " + json.dumps({"choices": [{"delta": delta}]})
+
+
+async def _collect(stream: AsyncIterator[object]) -> list[object]:
+    return [event async for event in stream]
+
+
+_CONFIG = OpenAICompatibleConfig(api_key="test-key", base_url="https://example.test/v1")
+
+_CASES = {
+    "openai_compatible": (
+        OpenAICompatibleProvider,
+        _sse(
+            _chat_chunk(role="assistant"),
+            "data:",
+            _chat_chunk(content="ok"),
+            "data: [DONE]",
+        ),
+        "model-x",
+    ),
+    "google": (
+        GoogleGenerativeAIProvider,
+        _sse(_google_chunk("o"), "data:", _google_chunk("k", "STOP")),
+        "gemini-x",
+    ),
+    "mistral": (
+        MistralConversationsProvider,
+        _sse(
+            _mistral_chunk(role="assistant"),
+            "data:",
+            _mistral_chunk(content="ok"),
+            "data: [DONE]",
+        ),
+        "mistral-x",
+    ),
+}
+
+
+@pytest.mark.anyio
+@pytest.mark.parametrize("name", sorted(_CASES))
+async def test_provider_ignores_empty_data_frames(name: str) -> None:
+    provider_cls, body, model = _CASES[name]
+
+    def handler(_request: httpx.Request) -> httpx.Response:
+        return httpx.Response(200, text=body, headers={"content-type": "text/event-stream"})
+
+    async with httpx.AsyncClient(transport=httpx.MockTransport(handler)) as client:
+        provider = provider_cls(_CONFIG, client=client)
+        events = await _collect(
+            provider.stream_response(
+                model=model,
+                system="You are Tau.",
+                messages=[UserMessage(content="Say ok")],
+                tools=[],
+            )
+        )
+
+    assert isinstance(events[-1], AssistantDoneEvent), events[-1]
+    assert events[-1].message.text == "ok"