Initial support for TypeVarLike default parameter (PEP 696)

[cdce8p]

Closes #75
Ref: https://peps.python.org/pep-0696/

[Gobot1234]

Not entirely sure how I feel about these subclassing the typing versions if possible but apart from the one comment LGTM

[cdce8p]

Not entirely sure how I feel about these subclassing the typing versions if possible

I agree. When I first experimented with it, it felt strange. However, with subclassing we don't need to worry about the base implementation and avoid reimplementing that again.

[Gobot1234]

Not entirely sure how I feel about these subclassing the typing versions if possible

I agree. When I first experimented with it, it felt strange. However, with subclassing we don't need to worry about the base implementation and avoid reimplementing that again.

I hadn't considered that, although it would still probably be a good idea to sys.version_info checks to make sure that you aren't redefining the behaviour if TypeVarLikes already have the default argument

[Gobot1234]

This class should define empty slots to make TypeVar not have dict

[cdce8p]

Done. Although I'm not sure it will work. For 3.10 _TypeVarLike and for 3.11 _BoundVarianceMixin don't add __slots__, so the __dict__ for TypeVar is created anyway.

[JelleZijlstra]

If you're interested, we can fix that for 3.12. (I'd be hesitant for older versions since it may break some users.)

[cdce8p]

Seems like __slots__ was removed on purpose since it didn't have any use. I've updated the PR to remove it from TypeVar as well. Ref: python/cpython#30444

[Gobot1234]

Just realised that the default for the default argument shouldn't be None and this needs a new object that should be the default as you currently can't make a distinction between default=None and default=empty

[cdce8p]

Just realised that the default for the default argument shouldn't be None and this needs a new object that should be the default as you currently can't make a distinction between default=None and default=empty

Should be easy enough to implement. Just wondering now why None was chosen as default for __bound__. True, a TypeVar with bound=None doesn't make much sense but that can't be the only reason, is it?
https://github.com/python/cpython/blob/v3.11.0rc2/Lib/typing.py#L940-L943

[JelleZijlstra]

I expect this just wasn't considered, and it's never come up as a problem because a bound of None has no practical use.

(Then again, pyright just got a bug report asking for NewType(None) to work, so maybe someone has a use for subtypes of None.)