Specifying int as type for self doesn't get flagged

[gvanrossum]

This doesn't give any error:

class C:
    def foo(self: int) -> None:
        pass

C().foo()

We ran into this in PY2 mode: Benjamin removed an argument from a method but forgot to remove it from the # type signature, and then realized that mypy didn't catch that. The PY2 form of signatures in # type comments doesn't require a type corresponding to self but it allows it, so his mistake effectively boiled down to the above example. Why isn't this an error?

[ddfisher]

I was recently looking at the code that does this. Here's where the type for self gets added to a method signature. It should be pretty straightforward to turn that into an explicit check that a self type is not provided, if that's how we want to solve this.

[gvanrossum]

Yeah, I recall reading that line too recently, but I drew a different conclusion -- I actually found a useful application for being able to specify a type for self. (Or, at least, I thought I had. See #1212.) Anyway, I think it's not a bug that you can specify a type for self -- I just think it's a bug that you can specify a type that doesn't have any overlap with the current class.)

[gvanrossum]

Believe it or not, this is a duplicate of #16.