Hi, On Wed, 2013-06-19 at 09:25 +0200, Thomas Schatzl wrote: > Hi, > > On Tue, 2013-06-18 at 15:53 -0700, Mandy Chung wrote: > > Hi Thomas, > > > > Thanks for the detailed analysis. > > > http://cr.openjdk.java.net/~tschatzl/8014890/webrev/ > > > > The fix looks good. I was able to reproduce the failure with the > > test case in the bug report running with jdk8 b94 fastdebug build. > > But the new regression test doesn't fail (I tried solaris-i586). > > Did it fail on your system? > > > > Yes, 100% on my developer machine. I will try to modify it to make it > fail on more machines (increase number of runs).
With 1000 runs (from 100) I can reproduce on product/win32, fastdebug+product/linux-x86_64, fastdebug+product/win64, fastdebug+product/linux-x86, fastdebug+product/solaris-x86, fastdebug +product/solaris-sparc... Still a few platforms, i.e. some fastdebug platforms, do not reproduce the failure. This With 100 runs I do not get that many platforms to fail. I will change the number of iterations to 1000 by default. When writing the test I was trying to make the test run as quickly as possible while still showing the issue. I probably went too far in the "being quick" direction. I updated the webrev (just the change for the number of iterations from 100 to 1000 in the test program). Its runtime is still only a few seconds. Thomas