[red-knot] Infer subscript expression types

[MichaReiser]

Limited support for inferring subscript expressions, e.g when the value is a tuple, or string.

def test(a: str):
	c = a[0]

[MichaReiser]

@carljm String and tuple support has already been implemented in #13562. Do you have something else in mind that goes beyond what's covered in that PR?

[AlexWaygood]

Following #13562, #13579 and #13667, I think the only thing we don't support with regard to type inference of subscript expressions is where a class does not define __class_getitem__ but its metaclass defines __getitem__. But we don't really support anything regarding metaclasses yet.

[AlexWaygood]

Oh, it looks like we might not support slices for string-literal and tuple types yet? x[1:3] should be inferred as tuple[str, str] if x has type tuple[int, str, str, int]. y[1:3] should be inferred as Literal["aa"] if y has type Literal["baab"].

[AlexWaygood]

We could probably also add special handling for subscript expressions where BytesLiteral types are on the l.h.s. and IntLiteral or BooleanLiteral types are on the r.h.s.