-
Bug
-
Resolution: Not an Issue
-
P3
-
fx2.1
There is 69% (-1,168 objects) footprint regression in NodeMemory.ComboBox-rendering-on-64m observed in
fx2.2-29 (b03) promoted build comparing to fx2.2-23 (b02).
Regression analysis shows that regression appeared somewhere between fx2.2-control-scrum-33 build
and fx2.2-control-scrum-43 build (no more builds are left in hudson).
Steps to run the benchmark and see the issue:
> cd JFX_WS/jfx/tests/performance/NodeMemory
> ant
> java -client -XX:+UseSerialGC -Xms64m -Xmx64m -cp "JFX_HOME/rt/lib/jfxrt.jar;./dist/NodeMemory.jar"
nodememory.ComboBoxTest -tr 120 -wt 0 -render true -allocCnt 10
I ran the test with "-allocCnt 1 -allocMax 100 -sleep true" to create only 100 ComboBox objects with both
both fx2.2-29 and fx2.2-23, then created heap dump files and compared them. Heap comparison shows that now
we at least one create com.sun.javafx.scene.control.skin.VirtualFlow per each ComboBox objects.
As result we have other more objects created which leads to footprint increase.
The full heap histogram difference is attached.
Need to investigate whether this is done intentionally and what can be improved.
fx2.2-29 (b03) promoted build comparing to fx2.2-23 (b02).
Regression analysis shows that regression appeared somewhere between fx2.2-control-scrum-33 build
and fx2.2-control-scrum-43 build (no more builds are left in hudson).
Steps to run the benchmark and see the issue:
> cd JFX_WS/jfx/tests/performance/NodeMemory
> ant
> java -client -XX:+UseSerialGC -Xms64m -Xmx64m -cp "JFX_HOME/rt/lib/jfxrt.jar;./dist/NodeMemory.jar"
nodememory.ComboBoxTest -tr 120 -wt 0 -render true -allocCnt 10
I ran the test with "-allocCnt 1 -allocMax 100 -sleep true" to create only 100 ComboBox objects with both
both fx2.2-29 and fx2.2-23, then created heap dump files and compared them. Heap comparison shows that now
we at least one create com.sun.javafx.scene.control.skin.VirtualFlow per each ComboBox objects.
As result we have other more objects created which leads to footprint increase.
The full heap histogram difference is attached.
Need to investigate whether this is done intentionally and what can be improved.