Use native ExpandTypeVisitor for expanding type aliases as well

[ilevkivskyi]

This used to be a source of code duplication, and also using the "native" visitor will allow us to implement new features for type aliases easily (e.g. variadic type aliases, that I am going to do in next PR).

[JelleZijlstra]

Looking at mypy-primer:

• The pytorch/vision error is here. The new error is correct; x[1] is a module and ModuleType.__file__ is str | None. Mypy got better at type inference here.
• The pandas hits are here. Very similar situation where mypy now infers a more precise type for an inspect.getmembers call.
• steam.py is here, another inspect.getmembers call. The code is buggy and the mypy error seems like a false positive, but again it looks like this PR infers more precise types. The second steam.py hit looks similar.

In conclusion, the PR does a better job of inferring types around inspect.getmembers.

[AlexWaygood]

For context, the more precise types we currently have for inspect.getmembers() in typeshed were added quite recently, in python/typeshed#9891 (by @Gobot1234, the author of steam.py). I noted in that PR thread that mypy didn't seem able to properly understand the new types we were adding (but merged the PR anyway, as it didn't seem to produce any mypy regressions, and was a clear improvement for pyright). It's great that it now can!

Could we add a test case for functions where a parameter is annotated with an alias to a generic TypeGuard?

[ilevkivskyi]

Yeah, it looks like a bug (or maybe not) in TypeTranslator, it doesn't translate the .type_guard in visit_callable_type(). Since ExpandTypeVisitor overrides it with some non-trivial logic that also includes .type_guard.accept(self) it worked there, but InstantiateAliasVisitor simply inherited the method before.

For now I will just add a test.

[github-actions]

Diff from mypy_primer, showing the effect of this PR on open source code:

vision (https://github.com/pytorch/vision) got 1.25x slower (42.3s -> 52.9s)
+ torchvision/models/_api.py:125: error: Item "None" of "str | None" has no attribute "endswith"  [union-attr]

steam.py (https://github.com/Gobot1234/steam.py)
- steam/state.py:2696: error: Need type annotation for "func"  [var-annotated]
- steam/state.py:2696: error: Argument 2 to "getmembers" has incompatible type "Callable[[Any], Any | bool]"; expected "Callable[[Any], TypeGuard[_T]]"  [arg-type]
+ steam/state.py:2696: error: Argument 2 to "getmembers" has incompatible type "Callable[[Any], Any | bool]"; expected "Callable[[Any], TypeGuard[Any | bool]]"  [arg-type]
+ steam/state.py:2697: error: Argument 1 to "get_annotations" has incompatible type "Any | bool"; expected "Callable[..., object] | type[Any] | Module"  [arg-type]
+ steam/state.py:2699: error: Incompatible types in assignment (expression has type "Any | bool", target has type "ParserCallback[ConnectionState, ProtobufMessage | Message | GCProtobufMessage | GCMessage]")  [assignment]
+ steam/state.py:2701: error: Incompatible types in assignment (expression has type "Any | bool", target has type "ParserCallback[ConnectionState, ProtobufMessage | Message | GCProtobufMessage | GCMessage]")  [assignment]
- steam/_gc/state.py:82: error: Need type annotation for "func"  [var-annotated]
- steam/_gc/state.py:82: error: Argument 2 to "getmembers" has incompatible type "Callable[[Any], bool]"; expected "Callable[[Any], TypeGuard[_T]]"  [arg-type]
+ steam/_gc/state.py:82: error: Argument 2 to "getmembers" has incompatible type "Callable[[Any], bool]"; expected "Callable[[Any], TypeGuard[bool]]"  [arg-type]
+ steam/_gc/state.py:83: error: Argument 1 to "get_annotations" has incompatible type "bool"; expected "Callable[..., object] | type[Any] | Module"  [arg-type]
+ steam/_gc/state.py:87: error: Incompatible types in assignment (expression has type "bool", target has type "ParserCallback[GCState[Any], GCProtobufMessage | GCMessage]")  [assignment]

spark (https://github.com/apache/spark)
+ python/pyspark/pandas/supported_api_gen.py:160: error: Argument 2 to "_organize_by_implementation_status" has incompatible type "dict[str, FunctionType]"; expected "dict[str, Callable[..., Any]]"  [arg-type]
+ python/pyspark/pandas/supported_api_gen.py:160: note: "Dict" is invariant -- see https://mypy.readthedocs.io/en/stable/common_issues.html#variance
+ python/pyspark/pandas/supported_api_gen.py:160: note: Consider using "Mapping" instead, which is covariant in the value type
+ python/pyspark/pandas/supported_api_gen.py:160: error: Argument 3 to "_organize_by_implementation_status" has incompatible type "dict[str, FunctionType]"; expected "dict[str, Callable[..., Any]]"  [arg-type]

materialize (https://github.com/MaterializeInc/materialize) got 1.11x faster (80.3s -> 72.5s)

[hauntsaninja]

Thank you! And thanks Jelle and Alex for looking through the primer diff