eclipse-capella
Performance regression around RotativeImageEditPart in Capella 6.0
Because of a regression detected in a T4C perfomance test, I had launch a Yourkit session to compare T4C 5.2.0 and T4C 6.0.0. The regression concerns the opening time of a "big" diagram. After analysis, it seems that the cause is some changes made on RotativeImageEditPart.
In T4C 5.2.0, the method org.eclipse.sirius.ui.business.internal.dialect.DialectUIManagerImpl.openEditor(Session, DRepresentation, IProgressMonitor) takes 66% of org.polarsys.capella.core.sirius.ui.runnable.OpenRepresentationsRunnable.run(IProgressMonitor).
In T4C 6.0.0, the same method takes 73%. In Yourkit, we can observe that 7% are for RotativeImageEditPart. These "elements" do not appear in 5.2.0. So we can deduce that the perfomance problem is caused by the changes about RotativeImageEditPart in Capella/Kitalpha.
For information, there is 1264 border nodes in the concerned diagram ("BigPhysicalPAB3"). I have not added the use case projects because there are big. I can send it (in private) if you want (projects compatible with 5.2.0 and 6.0.0).
Hi @lredor, you face large performance regression?
With a capella 6.0 v6.0.0.S20220321 or +, (with fix https://github.com/eclipse/kitalpha/pull/721 ), shall be longer but not blocker as before.
From an average of 12s62 to 17s (Capella 6.0.0.202205051516). With the fix and your measures, there is still a ratio of 1.5.