The attached test program causes the VM to abort with a SEGV around 30%
of the time on Intel systems with the -client option.
The source fine must be compiled with the "-source 1.5" option. It requires
a J2SE build with the JSR-166 library changes which can be found under:
/net/polgar/u/martin/src/jsr166/images (other systems)
The failure usually occurs in java.util.concurrent.ConcurrentHashMap$Segment.get(), but I don't see anything
obviously wrong with the generated code for this method. It may just be
using a value that was corrupted someplace else.