On Fri, 9 Oct 2026 21:55:01 GMT, Serguei Spitsyn <[email protected]> wrote:
>> I don't know how to exactly falicilitate this. I have created some tests for >> this with diffrent gc implementations on my computer (linux-amd64) >> >> >> Config (release, 1,000 runs each) Pass rc Avg Max >> default (G1) 1000 110 2.08 s 2.83 s >> -XX:+UseSerialGC 1000 110 2.58 s 6.13 s >> -XX:+UseParallelGC 1000 110 1.85 s 2.68 s >> -XX:+UseZGC 1000 110 2.26 s 4.43 s >> -XX:+UseShenandoahGC 1000 110 1.99 s 3.50 s >> -Xshare:off 1000 110 1.97 s 2.48 s >> -XX:+UseCompactObjectHeaders 1000 110 1.92 s 2.28 s >> -XX:ActiveProcessorCount=1 1000 110 1.90 s 2.79 s >> -XX:TieredStopAtLevel=1 1000 0 0.96 s 1.25 s >> >> The '-XX:TieredStopAtLevel=1' Also confirms that the bug most likely only >> shows up due to the ciStreams.cpp 'bug' >> >> But the tests run with 'vm.flagless' if that is what you meant > >> But the tests run with 'vm.flagless' if that is what you meant > > I agree, this is good for test stability. > What I'd like to say is that we need to make sure there are no intermittent > failures. > So, at least this test has to be run 1000's times on multiple platforms. > Changing GC's is a good idea too but I leave it to you to decide. I'm not > sure we want it to test with all GC's all the time. Do you mean this should be integrated into the jtreg pipeline? or an external test to validate this? ------------- PR Review Comment: https://git.openjdk.org/jdk/pull/33191#discussion_r4243519011
