Double underscore name mangling in classes to allow independent types

[JukkaL]

Mypy should perform name mangling in classes.

[marmistrz]

A simple example, exposing the behavior:

class foo:
    def __bar(self) -> None:
        print("xD")

foo()._foo__bar()

/tmp/test.py:5: error: "foo" has no attribute "_foo__bar"

[gvanrossum]

That looks intentional. You shouldn't be accessing private attributes like that.

[marmistrz]

@gvanrossum in production code not. But in tests you sometimes need to access the internal state of the object to validate what happened.

[JukkaL]

You can use getattr(foo(), '_foo__bar')() in tests. The main issue is allowing multiple classes in a class hierarchy use the same __foo attribute with independent types.

[marmistrz]

@JukkaL, yes, but it's less readable

[kwvanderlinde]

The fact of the matter is that foo does have the attribute _foo__bar and we all know it. The error message is a blatant lie and a source of confusion.

[gvanrossum]

Thank you for your encouragement. You should just use # type: ignore and move on. If you do this a lot, clearly something is wrong in your application or API design and mypy should not endorse that.

[ilevkivskyi]

@gvanrossum

You should just use # type: ignore and move on. If you do this a lot, clearly something is wrong in your application or API design and mypy should not endorse that.

As @JukkaL mentioned the original issue is not about allowing Foo()._Foo__foo, this issue is about allowing things like:

class B:
    __foo: int
class C(B):
    __foo: str

which is type safe but currently prohibited.

[gvanrossum]

Ouch, sorry, I read too quickly.

[rafaelcaricio]

I'm looking at this one.