On Tue, 8 Sep 2020 12:43:19 GMT, Johan Vos <j...@openjdk.org> wrote: >> This PR removes an old fix (RT-13965/JDK-8113318), which was applied in 2011 >> to avoid a memory leak in >> `VirtualFlow::sheetChildren`, after new items were constantly added. >> With the current VirtualFlow implementation, there are in place the >> necessary methods that take care of cleaning or >> adding new cells to the sheetChildren list, and such memory leak doesn't >> exist, the size remains constant when new >> items are added or removed (see included unity test). The call to >> `sheetchildren.clear()`, on the contrary, has a big >> impact in performance and CPU usage, when new items are constantly added, as >> has been reported in JDK-8185886/#108. > > This looks ok to me. It would be ideal to have the test failing before > JDK-8113318 was applied, as that would show the > issue for which JDK-8113318 was created is now fixed. But I do understand it > is very hard to apply this test to a code > snapshot from 2011.
As there is no automated test to make sure we are not re-introducing the leak, I have verified by running Ensemble8 KeyEvent sample mentioned in JDK-8113318 and typing more than 800 keys. There is no performance degradation observed. The change and test looks OK. ------------- PR: https://git.openjdk.java.net/jfx/pull/298
