Additional validation for TypeVar defaults (PEP 696)

[cdce8p]

Check that default type is a subtype of bound or one of the constraint types.

Add default to nodes __match_args__. Missed that initially.

Ref: #14851

[github-actions]

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

[A5rocks]

Add default to nodes __match_args__. Missed that initially.

I've been wondering: why don't we use dataclasses? I probably missed something.

[AlexWaygood]

Add default to nodes __match_args__. Missed that initially.

I've been wondering: why don't we use dataclasses? I probably missed something.

Until recently it wasn't an option, because we supported Python 3.6, and dataclasses were introduced in Python 3.7. Slotted classes are also more performant, though, and you can't do slotted classes with dataclasses on <3.10. (Performance matters quite a lot for mypy!)

[JelleZijlstra]

Slotted classes are also more performant, though, and you can't do slotted classes with dataclasses on <3.10. (Performance matters quite a lot for mypy!)

You can, you just have to write out the __slots__ manually.

[AlexWaygood]

Slotted classes are also more performant, though, and you can't do slotted classes with dataclasses on <3.10. (Performance matters quite a lot for mypy!)

You can, you just have to write out the __slots__ manually.

Good point, I forgot about that.

[cdce8p]

I've been wondering: why don't we use dataclasses?

Might also be that "normal" classes are already quite optimized with mypyc, not sure about that though. Would certainly require some more analysis.

[cdce8p]

Could someone add the topic-pep-696 label here as well? That would make it easier to keep track.

[AlexWaygood]

Could someone add the topic-pep-696 label here as well? That would make it easier to keep track.

Done!