On Thu, 3 Sep 2026 20:30:03 GMT, Viktor Klang <[email protected]> wrote:
>>> @macarte Converting the failing test to use the test infrastructure to run >>> it with -UseLSE and requiring an aarch64 uarch seems rather tractable. >>> @johan-sjolen, wdyt? >> >> The problem is that while that test fails 100% on Windows ARM64, it only >> fails on Linux less than 5% of the time and then only on certain machines >> (large number of cores) > > @macarte Personally I see it as this: If that's the best we can do, then it's > the best we can do. I'd rather have a test case that has a very decent chance > at catching a regression than no test at all. (The test will run in a lot of > CI builds over time so it will statistically fail sooner rather than later) @viktorklang-ora , @theRealAph : I've added the non-LSE variant tests to cover regressions; in doing so I also realized I had to update the accessor usage in SynchronousQueue as it uses LinkedTransferQueue. ------------- PR Comment: https://git.openjdk.org/jdk/pull/31465#issuecomment-6153002198
