Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Type inference on owner argument to __set_name__ #16796

Open
sg495 opened this issue Jan 19, 2024 · 0 comments
Open

Type inference on owner argument to __set_name__ #16796

sg495 opened this issue Jan 19, 2024 · 0 comments
Labels

Comments

@sg495
Copy link

sg495 commented Jan 19, 2024

Because of the circumstances under which __set_name__ is invoked, it should be possible to perform type inference on its owner argument, in a way which could be used to statically typecheck other components of a descriptor.

As a brief motivating example, below is a sketch of descriptor for a validated mutable attribute of integer type:

from __future__ import annotations
from typing import Any, Generic, Protocol, Self, Type, TypeVar, cast, overload

InstanceT = TypeVar("InstanceT")
InstanceT_contra = TypeVar("InstanceT_contra", contravariant=True)

class Validator(Protocol[InstanceT_contra]):

    def __call__(self, instance: InstanceT_contra, value: int, /) -> bool:
        ...

class IntAttr(Generic[InstanceT]):

    __validator: Validator[InstanceT]
    __name: str

    def __init__(self, validator: Validator[InstanceT]) -> None:
        self.__validator = validator

    def __set_name__(self, owner: Type[InstanceT], name: str) -> None:
        self.__name = name

    @overload
    def __get__(self, instance: None, _: Type[Any]) -> Self:
        ...

    @overload
    def __get__(self, instance: InstanceT, _: Type[Any]) -> int:
        ...

    def __get__(self, instance: InstanceT|None, _: Type[Any]) -> int | Self:
        if instance is None:
            return self
        return cast(int, getattr(instance, f"__{self.__name}"))

    def __set__(self, instance: InstanceT, value: int) -> None:
        if not self.__validator(instance, value):
            raise ValueError()
        setattr(instance, f"__{self.__name}", value)

class C:

    x = IntAttr(lambda self, value: self.validate(value)) # Pylance error

    def validate(self, value: int) -> bool:
        return value >= 10

In Pylance with strict typechecking rules, the definition of the x descriptor raises the following errors:

reportUnknownMemberType
  Type of "validate" is unknown 
reportUnknownLambdaType
  Return type of lambda is unknown
reportGeneralTypeIssues
  Cannot access member "validate" for type "object*"
    Member "validate" is unknown

Currently, Mypy doesn't raise any errors about the lack of lambda typing (not even for incorrect implementations, such as self.validatez(value) or self.validate(str(value))). Mypy also doesn't raise errors about incorrect lambda implementations once an explicit type hint is provided (see below), while Pylance does.
I opened a separate issue about this: #16797

Explicitly providing a hint for the InstanceT type removes the errors for Pylance:

class C:

    x = IntAttr["C"](lambda self, value: self.validate(value)) # Errors

    def validate(self, value: int) -> bool:
        return value >= 10

Performing inference on the owner argument to IntAttr.__set_name__ in the context of class C would ideally result in C being inferred as a value for InstanceT, allowing for static typecheking of the validator lambda function without the need for an explicit type hint.

Related issues for Mypy:

Sister issue for Pylance/Pyright: microsoft/pyright#7039

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
Projects
None yet
Development

No branches or pull requests

1 participant