On Tue, 26 Sep 2023 12:02:52 GMT, David Holmes <dhol...@openjdk.org> wrote:
>> Ok! >> So we would have to swap the popping and state-update at least. Unless we >> insert members between them, the updates would still be observable in any >> order, I suppose, but inserting membars just to make SA happy seems a bit >> much. >> Can one of you try swapping and see if that makes the problem appear less >> often? I could not yet reproduce the bug at all. > > I managed to get it to fail 4/100 times in our CI so armed with that I will > try swapping the order. Though I have to admit I tend to agree with Chris > that the code in question seems unlikely to be executed such that this > failure mode is so repeatable. I'm not seeing that this code has much impact on the failure rate. I've tried changing the order and also have tried adding a nanosleep(). Always fails at about the same rate. ------------- PR Review Comment: https://git.openjdk.org/jdk/pull/15907#discussion_r1337736512
