Interaction between class-scoped TypeAlias and Self

[Gollum999]

Does anyone have any thoughts on whether or not this interaction between TypeAlias and Self should be allowed?

from typing import Self, TypeAlias

class Foo:
    T: TypeAlias = int
    
    def bar(self, val: Self.T) -> Self.T:
        return val + 2
        
    def baz(self, val: Foo.T) -> Foo.T:
        return val * 2


Foo().bar(4)
Foo().baz(5)

With Python 3.11 and mypy 1.7.1, this currently results in:

main.py:6: error: Name "Self.T" is not defined  [name-defined]
Found 1 error in 1 file (checked 1 source file)

It feels to me like if baz works, then bar should also work, but I'm not sure what kind of implications that would have. The PEPs for TypeAlias and Self don't really seem to describe how this should work.

[Gobot1234]

We did actually specify this in PEP 673

We reject type aliases containing Self. Supporting Self outside class definitions can require a lot of special-handling in type checkers. Given that it also goes against the rest of the PEP to use Self outside a class definition, we believe the added convenience of aliases is not worth it:

TupleSelf = tuple[Self, Self]  # Rejected

class Alias:
    def return_tuple(self) -> TupleSelf:  # Rejected
        return (self, self)

Someone even brought it up another repo after it had been submitted (which I can't find anymore). We just decided it wasn't worth the effort. I wouldn't be particularly opposed as long as it had to be bound by a class but the whole idea of accessing members on Self does worry me a little because it no longer means that it can be desugared from the original bound TypeVar approach that (every??) type checker(s) use(s) to implement them.

[Gollum999]

Ahh, thanks for the context. I do remember seeing that in the PEP, but I wasn't sure if it applied in this case since this usage of Self is still contained in the class definition. But I suppose "We reject type aliases containing Self" is pretty explicit.