-=| Damyan Ivanov, 17.09.2018 05:09:09 +0000 |=- > -=| gregor herrmann, 17.09.2018 00:18:35 +0200 |=- > > On Sun, 16 Sep 2018 14:42:05 +0000, Damyan Ivanov wrote: > > > > > The test passes a thousand runs if the server child ignores SIGPIPE, > > > so this looks like the TLSv1.3 shutdown problem when the client makes > > > a fast SSL shutdown without waiting for the confirmation from the > > > server, then exits, then exits and the server sending its confirmation > > > gets SIGPIPEd. > > > > > > https://rt.cpan.org/Ticket/Display.html?id=126899 also talks about > > issues around SIGPIPE. > > And there's a new 2.060, maybe we should try and check from there. > > Oh, nice. Thanks for that notification. I was delving in internals > trying to avoid ignoring SIGPIPE. > > Upstream seems to have gone the other way, ignoring SIGPIPE and > documenting the situation. It it suits them, I think it is alright :) > > I'll run the test suite a hundred times and upload the new upstream > release.
No luck. The tests passed locally a hundred times, but two of them failed on the buildd (https://buildd.debian.org/status/fetch.php?pkg=libio-socket-ssl-perl&arch=all&ver=2.060-1&stamp=1537167776&raw=0). Wstat: 13 seems like the process was killed by SIGPIPE :( I can reproduce this looping over the two tests for a thousand times. I'd patch the two failing tests to ignore SIGPIPE, but I wonder if it would be even better to patch all the tests like that, to avoid the whack-a-mole game with the buildds. -- dam