On Mon, 1 Nov 2021 04:39:27 GMT, Jakob Cornell <d...@openjdk.java.net> wrote:
> This will fix a few issues with the tests added in #5290: > > - [ ] intermittent failures > - [x] tests should use `failure` method to report problems rather than > throwing `AssertionError` I'm not sure whether this is expected to be a platform-specific bug, but I've been assuming it is since I've never seen the failure on my Linux system. I started with the Windows failure and noticed that the JDB loop mechanism doesn't seem to be working: Sending command: 2 2 up reply[0]: End of stack. reply[1]: main[1] That would seem to be the cause of the "Unexpected output" error, but I can't reproduce the behavior manually on the x86-64 Windows system I have access to. With repeat on: main[1] 2 2 up End of stack. main[1] End of stack. main[1] End of stack. main[1] End of stack. main[1] I wonder if Daniel can provide any more information that might enable me to reproduce the issue for debugging. If not, I don't think I can diagnose the issue. Another question I have is whether anybody has attempted to find the point the bug was introduced by backporting the test to earlier versions of the JDK. I think that would essentially mean reducing the test to `2 2 up` or similar, since the `repeat` command is brand new. I'm assuming the Mach 5 failure has the same underlying cause; I could try virtualizing Linux on PowerPC to try to reproduce that one, but that seems likely to take a lot of time and unlikely to turn up anything useful. ------------- PR: https://git.openjdk.java.net/jdk/pull/6182