On Wed, 6 Sep 2023 05:01:38 GMT, Serguei Spitsyn <sspit...@openjdk.org> wrote:
> The fix looks okay to me. From JVMTI point of view this transition is hidden, > so no inconsistency can be observed. However, I wonder if the > AsyncGetStackTraces may have similar issue as the JFR had. Unfortunately we > do not have good test coverage to rely on. Thanks Serguei. When testing, I had one failure (that I couldn't duplicate) of serviceability/jvmti/stress/StackTrace/NotSuspended/GetStackTraceNotSuspendedStressTest.java with -Xcomp XX:CompileThreshold=100 -XX:+TieredCompilation -XX:+DeoptimizeALot. It's looking for the stack trace of the enter method but printed the stack trace of the carrier (not the virtual thread) when it failed. I'll create a bug for that as as there's a lurking issue there with the test, GetStackTrace or the non-standard/extension function GetVirtualThread. AsyncGetStackTraces doesn't know about stack chunks so it's not reliable when a virtual thread is mounted. ------------- PR Comment: https://git.openjdk.org/jdk/pull/15492#issuecomment-1712573937
