This one was added as part of implementingoverride earlier in 3.12, so there's no need to keep it around for compatibility.

This feels slightly strange to me, as (to me) it implies thatdataclass_transform is a function generic around the type variableT. But that's not really true; it's a non-generic function that returns an identity function. I skimmed PEP 695, and it doesn't seem to mention how such cases should be handled.

IIRC we did think about this and decided that this is indeed how such cases should be spelled (sinceCallable doesn't have a way yet to spell "generic").

I think we shouldn't do this since it confuses scoping and much of the point of PEP 695 is to clarify scoping. We can use a callback protocol to spell this.

I'd brought this up somewhere before PEP 695 was accepted, but didn't get much of a response. Presumably since it's not the most common thing and I think e.g. pyre might not even support-> Callable[[T], T]

Changed to a callback protocol

IIRC we did think about this and decided that this is indeed how such cases should be spelled (sinceCallable doesn't have a way yet to spell "generic").

Yeah, the alias makes that so much more readable, in my opinion. Thanks! Good call@gvanrossum.