On Tue, 29 Oct 2024 16:52:45 GMT, Tom Rodriguez <ne...@openjdk.org> wrote:
> The notification that finalize is complete should be done after printing the > message because in Xcomp mode there might be a significant delay at the > println so the object hasn't actually been finalized yet. It probably should be volatile just from a normal Java perspective but I don't think it matters for the correctness of this test. I think the particular failure Graal was seeing was really because the System.runFinalization call that was added in the fix for JDK-6771058. This creates a second thread which might be moving forward while the main FinalizerThread is stalled in an Xcomp triggered compilation. In rare cases this lets the test finish before the first finalizable object has actually been finalized. Anyway, it's all super rare and I can't reproduce the failure against the latest master. I was able to reproduce and confirm the fix in local runs against earlier bits. ------------- PR Comment: https://git.openjdk.org/jdk/pull/21767#issuecomment-2450372063
