test: cover Generator dump/load + async empty-response guard (Tier C)

tests/generation/structured_output/test_generator_dump_and_errors.py

@@ -0,0 +1,59 @@
+"""Tests for Generator serialization and chat-completion error handling.
+
+These cover paths the existing structured-output tests don't: the dump/load
+round-trip and the async empty-response guard. The OpenAI client is constructed
+for real (no network) via the respx fixture, which supplies a fake API key and
+model name.
+"""
+
+from __future__ import annotations
+
+from typing import TYPE_CHECKING
+
+import httpx
+import pytest
+
+pytest.importorskip("openai")
+
+from autointent.generation import Generator
+from autointent.generation.chat_templates import Role
+
+if TYPE_CHECKING:
+    from pathlib import Path
+
+    from respx.router import MockRouter
+
+
+def _empty_choices_response() -> httpx.Response:
+    return httpx.Response(
+        200,
+        json={
+            "id": "chatcmpl-test",
+            "object": "chat.completion",
+            "created": 0,
+            "model": "gpt-test",
+            "choices": [],
+            "usage": {"prompt_tokens": 1, "completion_tokens": 0, "total_tokens": 1},
+        },
+    )
+
+
+def test_generator_dump_load_round_trip(respx_openai: MockRouter, tmp_path: Path) -> None:
+    gen = Generator(model_name="gpt-test", base_url=None, use_cache=False, temperature=0.5, max_tokens=128)
+
+    gen.dump(tmp_path / "gen")
+    loaded = Generator.load(tmp_path / "gen")
+
+    assert loaded.model_name == "gpt-test"
+    assert loaded.base_url is None
+    assert loaded.use_cache is False
+    assert loaded.generation_params == {"temperature": 0.5, "max_tokens": 128}
+
+
+@pytest.mark.asyncio
+async def test_get_chat_completion_async_raises_on_empty_choices(respx_openai: MockRouter) -> None:
+    generator = Generator(use_cache=False)
+    respx_openai.post("/v1/chat/completions").mock(return_value=_empty_choices_response())
+
+    with pytest.raises(RuntimeError, match="No response received"):
+        await generator.get_chat_completion_async([{"role": Role.USER, "content": "hi"}])