Support for attr.has and attr.AttrsInstance

[koogoro]

Feature
Recently, mypy added a plugin for attr.fields that tries to detect whether a type is an attr.s-decorated class, and reports an error if it isn't. However, attrs has a couple ways to check or annotate that a class is an attr.s-decorated class: attr.has and attr.AttrsInstance, neither of which seem to be supported by mypy right now.

attr.AttrsInstance is a protocol defined in the attrs internal stubs (https://github.com/python-attrs/attrs/blob/main/src/attr/__init__.pyi#L74) that represents that an object of an attr.s-decorated class. Right now, the plugin doesn't recognize attr.AttrsInstance, so it will report an error when trying to call e.g. attr.fields on a type[attr.AttrsInstance].

attr.has is a method that checks whether a class is attr.s-decorated. The plugin won't recognize this as a valid way to ensure attrs version 22.2.0 updated the internal stubs for attr.has to be a TypeGuard for type[attr.AttrsInstance]: (https://github.com/python-attrs/attrs/blob/main/src/attr/__init__.pyi#L541), so supporting attr.AttrsInstance may be sufficient when using sufficiently new stubs for attrs.

Pitch

A possible use for attr.has:

T = TypeVar('T')

def deserialize(attr_cls: type[T], value:bytes) -> T:
    assert attr.has(cls)
    fields = attr.fields(cls)
    ...

A possible use for attr.AttrsInstance:

T = TypeVar('T', bound=attr.AttrsInstance)

def deserialize(attr_cls: type[T], value: bytes) -> T:
    fields = attr.fields(attr_cls)
    ...

[koogoro]

@hauntsaninja this is the thing that I was having trouble with with the new behavior of the plugin. I think it probably counts as a feature request because we didn't really support these before either, we just wouldn't error on them because we didn't check the types around attr.fields.