Hi Tobias, On Thursday, 19 October 2017 19:42:59 CEST Tobias Stöckmann wrote: > On 19/10/17 04:45, Pádraig Brady wrote: > > Isn't there also a small race if SIGCHLD is received > > after wait() returns but before the sigsuspend() runs? > > great catch, that could happen indeed. As you wrote in your other mail, > this should go in as a separate patch as it fixes another issue my > original patch tried to address. > > The one reported by Thomas is an actually introduced regression, sorry > for that.
no worries about the regression, such stuff happens all the time. It took several hours to track down the issue because every time I ran timeout(1) via strace, it worked just fine. After a lot of printf debugging in timeout(1) I found out what was going on. Before switching to timeout(1), we had an similar implementation called "exec_timeout". It suffers from the same race condition that you fixed ;) Another good point for having timeout(1) in coreutils. Cheers, Thomas
