fix: resolve async detection and test issues

- Fix async detection to not call client factory prematurely
- Add use_async parameter to explicitly indicate async mode
- Update test to avoid creating client during verification
- Fix test mocking to use correct module path

Author: zeke

src/replicate/_module_client.py

@@ -88,15 +88,14 @@ def _run(*args, **kwargs):
         return _load_client().run(*args, **kwargs)
 
     def _use(ref, *, hint=None, streaming=False, use_async=False, **kwargs):
+        from .lib._predictions_use import use
+        
         if use_async:
             # For async, we need to use AsyncReplicate instead
             from ._client import AsyncReplicate
-            from .lib._predictions_use import use
-
-            return use(lambda: AsyncReplicate(), ref, hint=hint, streaming=streaming, **kwargs)
-        from .lib._predictions_use import use
-
-        return use(_load_client, ref, hint=hint, streaming=streaming, **kwargs)
+            return use(lambda: AsyncReplicate(), ref, hint=hint, streaming=streaming, use_async=True, **kwargs)
+        
+        return use(_load_client, ref, hint=hint, streaming=streaming, use_async=False, **kwargs)
 
     run = _run
     use = _use

src/replicate/lib/_predictions_use.py

@@ -858,6 +858,7 @@ def use(
     *,
     hint: Optional[Callable[Input, Output]] = None,  # pylint: disable=unused-argument # noqa: ARG001 # required for type inference
     streaming: bool = False,
+    use_async: bool = False,  # Internal parameter to indicate async mode
 ) -> Union[
     Function[Input, Output],
     AsyncFunction[Input, Output],
@@ -878,8 +879,8 @@ def use(
     except AttributeError:
         pass
 
-    # Determine if this is async by checking the type
-    is_async = isinstance(client, AsyncClient) or (callable(client) and isinstance(client(), AsyncClient))
+    # Determine if this is async
+    is_async = isinstance(client, AsyncClient) or use_async
 
     if is_async:
         # TODO: Fix type inference for AsyncFunction return type

tests/test_simple_lazy.py

@@ -24,9 +24,10 @@ def test_use_does_not_create_client_immediately():
                 assert isinstance(model, Function)
                 print("✓ replicate.use() works without immediate client creation")
 
-                # Verify the client is stored as a callable (factory function)
-                assert callable(model._client)
-                print("✓ Client is stored as factory function")
+                # Verify the client property is a property that will create client on demand
+                # We can't call it without a token, but we can check it's the right type
+                assert hasattr(model, '_client_or_factory')
+                print("✓ Client factory is stored for lazy creation")
 
             except Exception as e:
                 print(f"✗ Test failed: {e}")
@@ -54,7 +55,7 @@ def track_client_creation(**kwargs):
 
     with patch.dict(os.environ, {}, clear=True):
         with patch.dict(sys.modules, {"cog": mock_cog}):
-            with patch("replicate._module_client._ModuleClient", side_effect=track_client_creation):
+            with patch("replicate._client._ModuleClient", side_effect=track_client_creation):
                 import replicate
 
                 # Create model function - should not create client yet