GeezerVison

@Wulian233 I am not sure what you mean. Please see #132614 and #118761 (comment) for the motivation.

Thanks! Feel free to make the same change for more modules in this PR.

@JelleZijlstra thanks for taking a look! I've added the tests for enum, functools and email.utils modules. This should cover the modules handled in the original issue #109653. I'll stop for now.

Unless somebody else beats me to it, I can go through the rest of the modules handled in #118761, maybe next weekend.

Unless somebody else beats me to it, I can go through the rest of the modules handled in #118761, maybe next weekend.

I went through the rest of the modules from #118761 and added the respective tests.
I only skipped importlib.metadata and tomllib since it is not clear to me if the tests should live in cpython repo or in the respective upstream repositories. Leaving that discussion for a follow-up PR.

This should be good to go. @JelleZijlstra I pushed one more commit which I forgot to push with the others, apologies.

Also a merge conflict. If you don't get to it I can spend a bit more time later today fixing it up, hopefully we can still get it in by the beta. (And it's not the end of the world if we don't.)

Also a merge conflict. If you don't get to it I can spend a bit more time later today fixing it up, hopefully we can still get it in by the beta. (And it's not the end of the world if we don't.)

Done, thanks for a thorough review!

I'm getting these by searching for "import" in the source file and collecting the imports inside functions

Ah, good thinking. I've only included those modules that were specifically made lazy in the various PRs.

This got me thinking though that the current approach is somewhat suboptimal -- it would be more robust to have an allowlist of modules. Right now, a previously unused module can still be added and the tests might still pass. But this PR is definitely an improvement, LMK if you want me to open a follow-up issue for discussion.

An allowlist could make sense too, but might be harder to maintain. Feel free to open an issue discussing it. I guess my primary motivation was "I did all this work to make it so typing doesn't import annotationlib, let's make it so we don't regress by accident".