Workaround for small race condition in parallel doctest runner

[embray]

This is hard to reproduce reliably, but can happen if a worker process outlives its deadline and w.kill() is called, but the process has exited on its own by the time killpg() is called.

I've had this happen a few times on Cygwin where process shutdown tends to take a little longer.

CC: @jdemeyer

Component: doctest framework

Author: Erik Bray

Branch/Commit: 0b72ed5

Reviewer: Jeroen Demeyer

Issue created by migration from https://trac.sagemath.org/ticket/24563

[jdemeyer]

comment:2

Some of the calls to this kill method are wrapped already in a try/except Exception block. So it seems to be a feature that this would raise an exception if the worker is already dead. I would therefore prefer to just wrap the one remaining call to kill in a try/except block.

[jdemeyer]

comment:3

Maybe a better alternative solution: return a boolean from kill() to indicate whether or not os.kill() was successful in sending the signal.

[embray]

comment:4

Replying to @jdemeyer:

Maybe a better alternative solution: return a boolean from kill() to indicate whether or not os.kill() was successful in sending the signal.

+1

[embray]

comment:5

Although maybe instead of "successful in sending the signal" it should be "successful in sending the signal or is already dead". After all what I've written will allow other exceptions to be raised.

No, I changed my mind. I'll do it the way you suggested.

[sagetrac-git]

Changed commit from 8e4506e to 0b72ed5

[sagetrac-git]

Branch pushed to git repo; I updated commit sha1. New commits:

0b72ed5

Change DocTestWorker.kill to return a value whether or not it actually signalled the process.

[embray]

New commits:

0b72ed5

Change DocTestWorker.kill to return a value whether or not it actually signalled the process.

[jdemeyer]

Reviewer: Jeroen Demeyer

[embray]

comment:9

I like how nearly every patchbot is apparently broken...

[vbraun]

Changed branch from u/embray/cygwin/doctest-kill to 0b72ed5