chore: import upstream snapshot with attribution

tests/cache/test_cache_integration.py

@@ -0,0 +1,45 @@
+import types
+
+import instructor
+from instructor.cache import AutoCache
+from openai.types.chat import ChatCompletionMessageParam
+from pydantic import BaseModel, Field  # type: ignore[import-not-found]
+
+
+def test_auto_cache_prevents_duplicate_provider_calls(monkeypatch):
+    _ = monkeypatch  # unused fixture for parity with other tests
+    """Ensure that AutoCache prevents duplicate provider calls via patch layer."""
+
+    class User(BaseModel):
+        name: str = Field(...)
+
+    call_counter = {"n": 0}
+
+    # Fake provider completion function mimicking minimal OpenAI chat response
+    def fake_completion(*_args, **_kwargs):  # noqa: D401, ANN001
+        call_counter["n"] += 1
+        content = User(name="cached").model_dump_json()
+        # Return minimal ChatCompletion-like object
+        return types.SimpleNamespace(
+            choices=[
+                types.SimpleNamespace(
+                    message=types.SimpleNamespace(content=content),
+                    finish_reason="stop",
+                )
+            ],
+            usage={},
+        )
+
+    # Create Instructor client using from_litellm so we go through patch stack
+    cache = AutoCache(maxsize=10)
+    client = instructor.from_litellm(fake_completion, mode=instructor.Mode.JSON)
+
+    messages: list[ChatCompletionMessageParam] = [{"role": "user", "content": "hello"}]
+
+    # First call – provider should be invoked
+    _ = client.create(messages=list(messages), response_model=User, cache=cache)
+    assert call_counter["n"] == 1
+
+    # Second call with identical inputs – should hit cache, no new provider call
+    _ = client.create(messages=list(messages), response_model=User, cache=cache)
+    assert call_counter["n"] == 1, "Cache miss – provider was called again"