RFR(S) 8211336 [TESTBUG] appcds tests with incorrect usage of -XX:+UseStringDeduplication (original) (raw)

Ioi Lam ioi.lam at oracle.com
Tue Oct 30 02:23:42 UTC 2018

On 10/29/18 6:23 PM, David Holmes wrote:

Hi Ioi,

On 30/10/2018 7:32 AM, Ioi Lam wrote: http://cr.openjdk.java.net/~iklam/jdk12/8211336-cds-test-fix-for-string-dedup.v01/

https://bugs.openjdk.java.net/browse/JDK-8211336 Please review this small test fix: These are "shell" tests that need to pass certain options (such as -XX:+UseStringDeduplication) to child JVM processes that are launched via jdk.test.lib.process.ProcessTools. The tests were using this: * @run main/othervm -XX:+UseStringDeduplication ExerciseGC However, options passed this way are not passed by ProcessTools. The fix is to change to the following * @run driver ExerciseGC -XX:+UseStringDeduplication and let the main test program explicitly pass args[] (indirectly) to ProcessTools. That looks really weird to me. I would not expect to pass things this way. If the test is only intending to launch other VMs then the logic to add -XX:+UseStringDeduplication should be an explicit part of the test, directly in the main test logic, not incidentally passed through via the mechanism you suggest.

The 11 tests each originally had only one @run flag, like

       * @run driver ExerciseGC

Later, it was suggested (JDK-8185531) to add more flag combinations, such as -XX:+UseStringDeduplication and -XX:-CompactStrings, to increase test coverage. It's pretty common to do this by adding a new @run tag with different flags, as done in JDK-8185531. The only problem is JDK-8185531 doesn't work because jtreg doesn't pass the additional flags to the child processes. Hence the fix in this RFR.

This way it's easy to add other VM flag combinations to the tests. You don't need to modify the test sources identically 11 times just to add a new flag combination to these 11 tests.

Thanks

David -----

(I also cleaned up some unnecessary @module tags)

Test in progress with hs-tier1/2. Thanks - Ioi

More information about the hotspot-runtime-dev mailing list