@ned-deily, @pablogsal: I propose to revert the change since it caused a regression:
https://bugs.python.org/issue34172#msg330955

@pablogsal is working on a fix https://bugs.python.org/issue35378 but I'm not fully happy with it. Even if it's a good change, I dislike pushing late fix just before a release. I would prefer to have more time to experiment the overall change and make sure that it's the best option.

https://bugs.python.org/issue34172 is likely a very old bug, as old as multiprocessing. So yeah, it should be fixed, but not by changing the behavior in a stable branch.