gh-132493: lazy evaluation of annotations in typing._proto_hook

[felixscherz]

Hi, this is in regards to #132493. This PR avoids eager evaluation of annotations during isinstance checks for protocols.

• Issue: Avoid eagerly evaluating annotations #132493

[JelleZijlstra]

Thank you!

[JelleZijlstra]

Also add a test with a class that does fulfill the protocol.

Also, please add a NEWS entry.

[felixscherz]

I added a class that satisfies the protocol and a NEWS entry:)

[JelleZijlstra]

We can probably gain some performance by doing the issubclass() and getattr() checks before doing the annotations check, since they should be fast and for most use cases they'll return False (it's surely more common to do isinstance(<non protocol>, Protocol), not isinstance(<other protocol>, Protocol)).

[felixscherz]

I moved the issubclass() and geattr() checks up:)

[felixscherz]

With the new order of conditions we could also move the call to _lazy_annotationlib.get_annotations so we only do that in case the other two conditions are true, what do you think?
Like this:

if (
issubclass(other, Generic)
andgetattr(other, "_is_protocol", False)
andattrin_lazy_annotationlib.get_annotations(base, format=_lazy_annotationlib.Format.FORWARDREF)
):

[JelleZijlstra]

These can use helpers assertIsSubclass, assertIsInstance, assertNotIsinstance.

[felixscherz]

Updated it to use the helper methods