RFR (S): 7900943: Change '7900913: new @requires tag' breaks JTreg in '-gui' mode

Volker Simonis volker.simonis at gmail.com
Fri Jun 6 17:14:38 UTC 2014


could sombody please review and sponsor the following small change
which fixes a problem of jtreg in '-gui' mode which was introduced by
change '7900913: new @requires tag':


The details are in the bug reported and repeated here for your convenience:

Change 7900913 introduced the new 'RegressionContext' class which
calls RegressionParameters.getTestVMJavaOptions() in its constructor.
This getTestVMJavaOptions() subsequently calls
getTestJavaOptions()/getTestVMOptions() which in turn may initilize
the testJavaOpts/testVMOpts to new empty ArrayLists's if there exist
no corresponding options.

If JTreg runs in '-gui' mode, it calls
RegressionParameters.load()/save() before executing tests.
Unfortunately, saving an empty ArrayList, results in saving an empty
option string (e.g. "regtest.testJavaOpts="). If such an option is
loaded back, it will result in ArrayList with an empty string element
(NOTICE that this is different from the empty ArrayList which was

The ArrayList with the empty string element is finally converted into
a Java or VM option without a parameter (e.g. "-J ") which breaks the
test execution (i.e. javac will complain about an empty '-J' option).

The problem can easily be reproduced by running the jdk regression
test in GUI mode in following way:

jtreg -dir:/OpenJDK/jdk9-dev/jdk/test -verbose:summary
-exclude:/OpenJDK/jdk9-dev/jdk/test/ProblemList.txt -conc:2 -Xmx512m
-a -ignore:quiet -timeoutFactor:5 -agentvm -gui
-w:/tmp/JTwork_jdk9_a -r:/tmp/JTreport_jdk9_a :jdk_stable

If you now choose the test 'com/sun/jdi/BadHandshakeTest.java' for
execution, it will fail in the compile step, because of an bogus '-J'
option. This can be verified by looking at the "rerun" string in the
test's "Test Run Message" tab.

The problem can be easily fixed by not loading/saving "empty options"
in RegressionParameters.load()/save0().

Thank you and best regards,

More information about the code-tools-dev mailing list