On 3/11/18, 7:48 PM, David Holmes wrote:
Hi Sherman,
Thanks for working on this and adding the new functionality to
OutputAnalyzer, but note that:
+ private static final String jvmwarningmsg =
+ "Java HotSpot\\(TM\\) 64-Bit Server VM warning:.*";
is Oracle JDK specific and 64-bit and server VM specific! Other tests
use:
Pattern.compile(".*VM warning.*")
to exclude all VM warnings. Sorry that wasn't clear from previous
discussions.
Hi David,
I was trying to be conservative to only filter out the "vm warnings"
particularly for this
test failure. The webrev has been updated as suggested to filter out any
line with
"VM warning".
Verified with the warning msg turned on as
http://cr.openjdk.java.net/~sherman/8196748/webrev.hs/src/hotspot/share/runtime/arguments.cpp.sdiff.html
and ran the jtreg with -XX:+FastTLABRefill.
That will not hit all cases as it requires the -XX:+FastTLABRefill to
be passed through to all JVMs launched by tests. The original problem
occurred with unconditional warnings that did not depend on a
particular flag being specified on the command-line. All of those
cases have now been fixed however so just reenabling the message
doesn't achieve anything.
The reason I did not try to turn "PrintSafepointStatisticsCount" back on
(to reproduce) is that it was
not included in the changeset published in JDK-8196739 as
http://hg.openjdk.java.net/jdk/jdk/rev/74be5b4ed152
It appears the reason I failed to reproduce LeadingGarbage.java is the
test in my repo does not forward
the vm options (I'm not sure why the test forwards those vm options in
your repo). I have updated the test
case to use the new test.lib (it was using the "old" jdk.testlibrary),
and forward the vmoptions (this one
should have nothing to do with old libs or new libs though)
Now the failure is reproducible and has been fixed with new method in
OutputAnalyzer.java.
webrev: http://cr.openjdk.java.net/~sherman/8196748/webrev
(arguments.cpp will not be put back, just to help reproduce)
Thanks,
Sherman
