Mypy doesn't ensure self types are reasonable

[ddfisher]

For example, this typechecks cleanly:

class C:
    def f(self: str) -> None:
        self.capitalize()

Related to #2193.

[ddfisher]

@elazarg

[elazarg]

Based on the above issue, this seems like a good idea... :)

[JukkaL]

This looks like a dupe of #16, but I'll leave this open since there's a PR that references this.

One thought from the discussion at #16: It may be reasonable to have a first argument (at least one not spelled as self) with an arbitrary annotation, but that function will not be a proper instance method and should only be callable within the class body. Example:

class A:
    def f(x: str) -> int:   # ok
        return int(x)

    a = f('123')  # ok

A().f()  # error, f is not a method

#16 has more discussion.

[gvanrossum]

Hm. Having fixed a bunch of issues the latest master found in both of our internal repos and typeshed, I would say that the fear of the pattern (functions without 'self' in classes) is exaggerated. The only thing we found was an interface (IMetadataProvider) in pkg_resources.py, which was meant for documentation only.

[elazarg]

IMO, ideally we would only allow use of such a function from within the class' scope. We would report an error for functions that are used from the outside (as suggested by Jukka), and also for functions that are unused from the inside.

However, personally I would vote for requiring an explicit hint for such functions, keeping the policy and implementation simpler.

[gvanrossum]

What would such an explicit hint look like?

[elazarg]

I was thinking about type: ignore... But of course it can be something else. And you can use @staticmethod I guess.

[gvanrossum]

Static methods can't be called without a class.

>>> @staticmethod
... def f(): pass
...
>>> f()
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
TypeError: 'staticmethod' object is not callable
>>>