All commit authors signed the Contributor License Agreement.

Note that this PR also inherently fixes builtins overriding globals in this particular code path. I was previously unable to come up with a test for this without using the undocumented ForwardRef() constructor. I have since done some thinking, and maybe I am missing something obvious, but all I could come up with was an (extremely contrived) test:

from annotationlib import Format, get_annotations

class C:
	y: alias[memoryview, undef]

mv = __builtins__["memoryview"]
del __builtins__["memoryview"]
fwdref = get_annotations(C, format=Format.FORWARDREF)["y"]

__builtins__["memoryview"] = mv
generic = fwdref.evaluate(
	format=Format.FORWARDREF,
	globals={"memoryview": int},
	locals={"alias": dict}
)
self.assertNotIsInstance(generic, ForwardRef)
self.assertIs(generic.__origin__, dict)
self.assertEqual(len(generic.__args__), 2)
self.assertIs(generic.__args__[0], int)
self.assertIsInstance(generic.__args__[1], ForwardRef)

It feels very unlikely this would happen in real code, and I'm not sure if it's a good idea to test given that "Since this is an implementation detail, it may not be used by alternate implementations of Python." (https://docs.python.org/3/library/builtins.html), and given that __builtins__ seems to be either a module object or a dictionary depending on where you run the code.

Good point, I think it's fine to skip the test in that case.

Technical issue: your two PRs #138430 and #138075 will conflict with each other, and I won't be able to backport them to 3.14 until after 3.14.0 final comes out. So I'm planning to leave them open for now and land them once 3.14.0 final is out so I can sort out the merge conflicts and merge the backports without having to disrupt the RC process.

Technical issue: your two PRs #138430 and #138075 will conflict with each other, and I won't be able to backport them to 3.14 until after 3.14.0 final comes out. So I'm planning to leave them open for now and land them once 3.14.0 final is out so I can sort out the merge conflicts and merge the backports without having to disrupt the RC process.

@JelleZijlstra Just a quick reminder of these PRs - any updates now that 3.14.0 final is out?

Thanks @dr-carlos for the PR, and @JelleZijlstra for merging it 🌮🎉.. I'm working now to backport this PR to: 3.14.
🐍🍒⛏🤖

GH-140927 is a backport of this pull request to the 3.14 branch.

Thanks for the merges @JelleZijlstra!