nine github.com/rakudo/rakudo/pull/4256 "Make signal handler setup synchronous to avoid race conditions" On MoarVM setup of a signal handler is itself an asynchronous operation. Wait for it to complete before returning as not only our spec tests but probably every single user expects the signal handler to be in place at that point. This was already true for the JVM and JS backends. 09:55
Fixes a race condition in t/spec/S17-procasync/kill.t
nwc10 I wonder if it's the same race condition as I just hit in t/spec/S07-hyperrace/basics.rakudo.moar 09:56
(lots of threads wating on a futex, and one sleeping in code called from the JIT)
nine github.com/Raku/roast/commit/da922...f413b07d0f "Fix yet another race condition in kill.t" Most of the time the 'Output correct for Supply.merge on signals' test was not executed at all because the test program finished normally before it ever got the signal. Even the test plan did not expect that test to run. Fix by having the test program basically wait forever for that
signal to arrive.
nwc10 not investigating this yet, as I have other things I can make progress on
sena_kun nine, this won't go into the release. 10:03
nine sena_kun: I figured as much :) 10:04
But at least for the release after that kill.t won't be an issue anymore
sena_kun nine++ for caring
nine github.com/Raku/roast/commit/93676...8f0bc88c27 "Fix race condition in bug-coverage-stress.t signal test" The signal handler sometimes did not even get a chance to run as the test program finished right after signalling that it started. To make it reliable the only proper way to exit the test program must be from the signal handler.So give that handler a really long time to 12:45
fire and exit unsuccessfully if it doesn't.
Sigh...there's yet another race condition in bug-coverage-stress.t. It contains a test that does almost the same as the one in kill.t but uses react/whenever instead of act on the signal supply. And with react/whenever we do not seem to always catch the signal . 13:09
lizmat that may point to a more general issue with react / whenever? 13:14
nine no idea 13:15
I don't understand how that could make a difference
sena_kun weekly twitter.com/koto_san_kana/status/1...8622331910 16:03
I bet this is not how it works.
:S
[Tux] Rakudo v2021.03-1-gbccb2ce2d (v6.d) on MoarVM 2021.03-1-gc48fd57a3
csv-ip5xs0.788 - 0.815
csv-ip5xs-207.749 - 8.152
csv-parser25.472 - 26.261
csv-test-xs-200.380 - 0.408
test7.871 - 8.040
test-t1.911 - 1.930
test-t --race0.821 - 0.977
test-t-2031.628 - 32.223
test-t-20 --race9.889 - 9.973
16:50
lizmat weekly: twitter.com/koto_san_kana/status/1...8622331910 19:52
notable6 lizmat, Noted! (weekly)
lizmat sena_kun: weekly: foo not weekly foo
sena_kun lizmat, next release is next month, so I'll forget till then. :] 19:53
lizmat, but thanks, and thank you very much for all your efforts. :)
lizmat you're welcome!
btw, spectest was clean for me :-) 19:54