For the time being Linux PPC64 regression tests of Jikes RVM CVS head will
be started manually. I plan on doing this at least once a week.
The need for this arises from the fact that for some reason the host IBM Java SDK crashes during the build process if regression tests start as a cron job:
(built utility_kludge.jar) \
/opt/IBMJava2-ppc-141/bin/java \
-Xbootclasspath/p:/opt/regression/powerpc... -Xbootclasspath/a:/opt/regression/powerpc64-... \
-Xmx400M \
-classpath /opt/regression/powerpc64-unkn... \
-Djava.security.policy=rvm.security \
\
BootImageWriter\
-classpath /opt/regression/powerpc64-unkn... \
-n /opt/regression/powerpc64-unknown-linux-gnu/nightShadow/images/production/RVM.primordials \
-o /opt/regression/powerpc64-unknown-linux-gnu/nightShadow/images/production/RVM.image \
\
-demographics -m /opt/regression/powerpc64-unknown-linux-gnu/nightShadow/images/production/RVM.map \
-ia 0x31000000JVMXM008: Error occured while initialising System ClassException in thread "main" Could not create the Java virtual machine.
make: *** [/opt/regression/powerpc64-unknown-linux-gnu/nightShadow/images/production/RVM.image] Error 1
make: Leaving directory `/opt/regression/powerpc64-unknown-linux-gnu/nightShadow/rvm/src/tools/bootImageWriter'
jbuild.linkImage: Exiting unexpectedly with status 2.
jbuild.linkImage: Cleaning up...
jbuild.linkImage: ...cleaned up.
69 s
jbuild: Trouble while running "./jbuild.linkImage -demographics"
(exit status 2); aborting execution
I’ve tried changing environmental variables of the cron job so they look exactly like those of a regular account, but it didn’t help.
The problem may have something to do with the 64-bit version of PPC Linux we have. It is one of the first pre-releases of 64-bit Yellow Dog Linux, which has a bunch of custom workarounds and kludges, including a custom-patched kernel 2.6.1
The long term solution I’d like to try is installing the latest Fedora Core 4, which does have 64-bit PPC kernel, and running the regression tests on top of that.