Generic types in annotations of __getattr__ in metaclasses are not deduced correctly

[dedebenui]

Environment data

• Language Server version: 2021.6.2
• OS and version: darwin x64
• Python version (and distribution if applicable, e.g. Anaconda): Python 3.9.5 (Anaconda)
• python.analysis.indexing: undefined
• python.analysis.typeCheckingMode: off

Expected behaviour

Pylance should determine the type of all 4 outputs to be int

Actual behaviour

Pylance determines that the type of output3 is Type[SomeFunctions] no matter the type of the argument to SomeFunctions.other_function.

Code Snippet / Additional information

from typing import Callable, TypeVar

T = TypeVar("T")

class FunctionCollection(type):
    def __getattr__(cls, key) -> Callable[[T], T]:
        return lambda x: x

    def function(cls, arg: T) -> T:
        return arg

    def factory(cls) -> Callable[[T], T]:
        return lambda x: x

class SomeFunctions(metaclass=FunctionCollection):
    pass


output1 = SomeFunctions.function(3)
output2 = SomeFunctions.factory()(3)
output3 = SomeFunctions.other_function(3)
output4 = SomeFunctions.__getattr__("other_function")(3)

[erictraut]

Thanks for the bug report. This will be fixed in the next release.

The problem is that the type evaluator assumed that the function returned by the metaclass __getattr__ needed to be bound to the function SomeFunctions. It therefore assumed that the first parameter (of type T) was a self parameter. The Python interpreter appears to know that the return result from __getattr__ should not be treated like other class or instance methods and therefore does not need to be bound by the caller's class, nor does it receive a self argument.

[jakebailey]

This issue has been fixed in version 2021.7.0, which we've just released. You can find the changelog here: https://github.com/microsoft/pylance-release/blob/main/CHANGELOG.md#202170-6-july-2021