A few comments about the test:
1. "-mx10m" should be replaced with "-Xmx10M"
2. assertGC() can be implemented more reliably:
List<int[]> l = new LinkedList<>();
while (true) {
try {
l.add(new int[10000]);
} catch (OutOfMemoryError e) {
break;
}
}
After such a loop, GC is guaranteed to have collected all the dead objects.
Thanks,
Artem
On 12/25/2012 11:34 AM, Petr Pchelko wrote:
Hello.
Thank you for your feedback.
Here is the new version of the fix:
http://cr.openjdk.java.net/~art/pchelko/7079254/webrev.01/
Changes from the previous one:
1. Added a test.
2. Moved creadLightweightDispatcherOnRemove into a synchronized section.
Best, Petr.
On Dec 24, 2012, at 4:03 PM, Artem Ananiev wrote:
Hi, Petr,
the fix looks fine. Please, move clearLightweightDispatcherOnRemove() into the
synchronized section, as lightweight dispatcher is used under tree lock.
As Sergey suggested, it would be fine to have a regression test for this change.
Thanks,
Artem
On 12/21/2012 4:50 PM, Petr Pchelko wrote:
Hello.
Could you please review the fix for the issue
7079254 Toolkit eventListener leaks memory
http://bugs.sun.com/bugdatabase/view_bug.do?bug_id=7079254
The webrev is available at:
http://cr.openjdk.java.net/~art/pchelko/7079254/webrev/
The LightweightDispatcher stored strong references to mouseEventTarget and
targetLastEntered which were not cleaned up when the component is removed,
which lead to a memory leak. The isCleaned field is added to be able to consume
events which would be dispatched to a removed component if we did not clean up
a reference to it.
The fix testet on Windows and Mac on toy apps and automatic regression tests
related to event dispatching, Component, Container, dnd and Mouse
Best, Petr.
