bpo-37531: Fix regrtest timeout for subprocesses

[nanjekyejoannah]

Fix regrtest timeout for subprocesses.

Naming things is the hardest thing in computer science. Interested in any reviews in a better name If any for the new function I introduced. cc @vstinner

https://bugs.python.org/issue37531

[ammaraskar]

Hey @nanjekyejoannah, did you accidentally attach the wrong bpo ticket for this? 3753 seems really old and unrelated.

[ZackerySpytz]

It seems that you have changed the file mode of several files under Lib/test. Why? I don't think those changes should be made.

[vstinner]

It seems like you modified Lib/test/test_os.py permissions somehow, please revert this change.

[nanjekyejoannah]

I have made the requested changes; please review again.

[bedevere-bot]

Thanks for making the requested changes!

: please review the changes made to this pull request.

[nanjekyejoannah]

Hey @nanjekyejoannah, did you accidentally attach the wrong bpo ticket for this? 3753 seems really old and unrelated.

Yes, thanks for catching this. I made the necessary changes.

[nanjekyejoannah]

It seems that you have changed the file mode of several files under Lib/test. Why? I don't think those changes should be made.

oh yes, I have reverted this.

[vstinner]

Overall, the change is great and seems to work as expected, but I tested it and it doesn't work as I expected:

• it doesn't report stdout/stderr on timeout
• regression: it ignores stdout/stderr on "CHILD_ERROR" whereas it was properly passed to the main process previously
• faulthandler and Popen.communicate() use the same timeout, so it's unclear which one which trigger first: I tuned timeouts to give more time to faulthandler. I prefer faulthandler since it dumps the traceback where Python was stuck, whereas Popen.communicate() timeout only returns "truncated" output with no traceback
• the timeout is reported as an exception, not as as "result"

I tried to add a commit to this PR but I wasn't allowed, so I forked this PR and created PR #15072 to add my changes on top of it.

[nanjekyejoannah]

Closing this then in favor of PR #15072