Behaviour of Type[Any]

[dmoisset]

Currently a Type[Any] is considered to have the same attributes as Type[object]. However, actually it probably has many attributes that could be used, especially when the Any comes from a module without stubs. Consider the following example:

from typing import Type
from some_module import Candy, Lollipop  # some_module has no stubs/annotations

class CandyFactory:
    manufactured_candy: Type[Candy]

cf = CandyFactory()
cf.manufactured_candy = Lollipop

print(cf.manufactured_candy.sugar_level)

Then running:

$ mypy --ignore-missing-imports candy.py 
candy.py:11: error: Type[Any] has no attribute "sugar_level"

Which seems to go a bit against the idea of gradual typing and intention of Any.

One possible solution is to consider that Type[Any] == Any, I think this actually would not miss anything that could be catched now, while reducing false positives.

[JukkaL]

I've wondered myself about how Type[Any] should behave, and your reasoning sounds about right. It should probably not be the same as Any with respect to subtyping/compatibility, only with respect to attribute access. Passing Type[Any] to a function that excepts an int should still be rejected (and vice versa).

[JukkaL]

I'm spelling out explicitly what should change:

For a value x of type Type[Any], mypy should allow arbitrary attribute access like x.foobar, and the result type should be Any.