Hi, thanks for your contribution and welcome!

Can't this be simplified by complexity removing the global and removing the as PPE?

@JulienPalard @methane if you remove the global, the import will happen for each file that is compiled. I know the second import isn't as expensive as the first import, but I'm not really sure how much overhead it would be.

If you're fine with an reimport per-file, then I can change it and this will be much less complex.

Something wrong with the tests, I'll look at it later.

Fixed the tests, and other improvements as suggested.

I know the second import isn't as expensive as the first import, but I'm not really sure how much overhead it would be.

It's negligible compared to compiling Python source.

if you remove the global, the import will happen for each file that is compiled.

really? I thought compile_dir is called only once.

Oops, this dropped off my radar. Still an issue though. :)

Updated to simplify as asked. Maybe needs a rebase?

We recommend "merge master" than "rebase". We use "Squash and merge" when merging PRs.
So merge commit of "merge master" is not problem to us. And rebase hides "what is changed after last code review".

CIs are failing. It likely because some tests for compilall are added. You need to merge master
and fix them.

Yeah, I had just noticed that. I fixed the tests (had broken the ability to run unlimited workers with max_workers=0.. oops!), but haven't merged in master yet.

I was running the tests locally by executing the test file directly (since it has a main) -- it seems that a6b3ec5 broke the ability to do that by adding a relative import to the file. If the tests pass on travis I'll merge in master and see what happens.

Looks like it's all working finally.

Thanks @virtuald for the PR 🌮🎉.. I'm working now to backport this PR to: 3.6, 3.7.
🐍🍒⛏🤖

GH-10686 is a backport of this pull request to the 3.7 branch.

GH-10687 is a backport of this pull request to the 3.6 branch.