> This test is failing often since 8304725 added a call to > Thread::current_in_asgct(). This can end up being called e.g. when resolving > calls, and then the OS last error value is lost. > > The test is reliable with a single warm-up call to getLastError.invoke() > before the loop. > > The test was introduced when in JDK-8292302 a change was undone that had made > JavaThread::threadObj call Thread::current_or_null_safe, as the use of TLS > upset this case of accessing last error directly. > > This new Thread::current_in_asgct() case shows that the VM will find new ways > to interfere with the last error value, or at least new VM code keeps wanting > to call Thread::current. This testcase is kind of niche usage, so it not an > argument that VM code should not be calling Thread::current. If this test > is to stay active, it needs to have this warm-up getLastError call. (If > there are more issues, it might mean removing the test.)
Kevin Walls has updated the pull request incrementally with one additional commit since the last revision: comment update feedback ------------- Changes: - all: https://git.openjdk.org/jdk/pull/13481/files - new: https://git.openjdk.org/jdk/pull/13481/files/dc91d5ab..1562317c Webrevs: - full: https://webrevs.openjdk.org/?repo=jdk&pr=13481&range=01 - incr: https://webrevs.openjdk.org/?repo=jdk&pr=13481&range=00-01 Stats: 1 line in 1 file changed: 0 ins; 0 del; 1 mod Patch: https://git.openjdk.org/jdk/pull/13481.diff Fetch: git fetch https://git.openjdk.org/jdk.git pull/13481/head:pull/13481 PR: https://git.openjdk.org/jdk/pull/13481
