Add redundant-annotation warning

[grayjk]

Relates to #18540

Adds a warning for redundant-annotation where the annotation type is the same as the inferred type

Based heavily on the code from @asottile in the issue

This MR does not warn on ClassVar, dataclass, or NamedTuple

[A5rocks]

I'm not completely sure this feature is fine. I thought it was, but when bringing it up in conversation, others pointed out that redefinition might depend on whether there's an annotation. Obviously mypy doesn't implement that algorithm now, but it might be nice to eventually have it? Or at least have the possibility of having it? And this feature would mean we can't.

To be concrete, I'm referring to how pyright/ty assume a declaration with a type (a: int = 5) can't be redefined but a declaration without (a = 5) can.

[A5rocks]

Could you also test:

• x: Literal[1] = 1
• y: list[str] = []
• this:

from typing import TypeVar

def f(x: T) -> T:
  return x

x: Literal[1] = f(1)
y: list[str] = f([])

And uh, I guess it would be nice to test a case where running type checking without type context would error. (you may need to silence errors for a specific run?)... unfortunately I cannot think of any examples.

[grayjk]

And this feature would mean we can't.

If that redefinition algorithm is implemented, couldn't this flag be mutually exclusive with the redefinition flag?

Could you also test:

Tests added (commit ebf40a5). In the TypeVar test, it looks like is_same_type does not handle this. May be related to #19761

test a case where running type checking without type context would error

Does that mean a test like:

# mypy: check-untyped-defs
def f():
    return 4

def g():
    j: int = f() 

[A5rocks]

If that redefinition algorithm is implemented, couldn't this flag be mutually exclusive with the redefinition flag?

Yeah, potentially. I don't think that would be good UX...

[A5rocks]

Does that mean a test like:

No, there's been issues like #20013 where part of the issue is that we're running type inference twice (once without type context for heuristics reasons, and once with) and the one without type context is erroring.

I can't think of any small reproducer because it would be a bug in mypy I think...

[github-actions]

According to mypy_primer, this change doesn't affect type check results on a corpus of open source code. ✅

[grayjk]

When --disallow-any-generic is enabled, it requires an annotation for:

a = (None, [])

that conflicts with the warning from --warn-redundant-annotation

[A5rocks]

@JukkaL do you have opinions on how this interacts with (maybe) redefinition? I forgot what heuristics you've mentioned using for it before.