This fixes a race condition, there's no way to reliably have a unit test.

The process needs to exit between the return code check and the action itself.

I think you could write a unit test by synchronising the threads so the race condition occurs.

You can mockos.kill() so that it waits for athreading.Event before continuing, then wait in another thread for the process to die. Once it's dead, you can release the first thread by setting the event and thekill() should raise aProcessLookupError.

You can click onforce-pushed (in@FFY00 FFY00 force-pushed the FFY00:[bpo-40550](https://bugs.python.org/issue40550) branch from 649b00f to 49e7004 5 hours ago), it should land you in this link:https://github.com/python/cpython/compare/649b00f3308818d88e237083600c1955162fdadd..49e7004bc963966bfb2f264d6022ef4477f21bcc. You can see the changes perfectly. It shows you the changes just as same as if I were to push a different commit, but without dirtying up the history. This might not be a problem in simple PRs like this but it makes a difference when you have more than 1 commit. And makes a huge difference when you are changing a lot of code, since you won't be able to look at one clean commit, you will need to look at the original commit and the fix. For these reasons, I believe force pushing to be good practice.

Sorry, after writing all that I messed up the git history 😕

You can still view the correct diff here:https://github.com/python/cpython/compare/49e7004bc963966bfb2f264d6022ef4477f21bcc..3d7dfbc1c19280dd19693d8d8e0574590c0c1bee

@gpshead do you have a minute to review this?

Thanks@FFY00 for the PR, and@gpshead for merging it 🌮🎉.. I'm working now to backport this PR to: 3.8, 3.9.
🐍🍒⛏🤖

GH-23439 is a backport of this pull request to the3.9 branch.

Sorry,@FFY00 and@gpshead, I could not cleanly backport this to3.8 due to a conflict.
Please backport usingcherry_picker on command line.
cherry_picker 01a202ab6b0ded546e47073db6498262086c52e9 3.8