Community
Participate
Working Groups
We appear to have a significant decrease in performance for opening the Ant editor...which is surprising considering the limited changes made to the editor in the 3.2 timespan. I need to figure out what is to blame.
This would appear to be a duplicate of bug 126901
I have found a small improvement to add in 3.2 which removed one additional refresh of the outline view. Changes to AntEditorContentOutlinePage
So with no changes to the Ant editor we now have an even larger regression looking at http://download.eclipse.org/eclipse/downloads/drops/I20060314-0800/performance/eclipseperfwin2_R3.2/org.eclipse.ant.tests.ui.editor.performance.OpenAntEditorTest.testOpenAntEditor2().html#Elapsed%20Process Do either of you guys have an idea on what has changed in the after I20060307 that could cause this slowdown?
cc dp as well.
Mmh - the various "open text editor" tests are still red but improved. What exactly are you testing? - open directly after startup - open after doing some warm up? - reopen one that has been opened before? The diff looks so huge that it should be quite easy to detect this regression. Did you try that?
I tried to find a difference on a WindowsXP machine by comparing I20060307-1315 with I20060314-0800 but so far the only big difference was: 46.41% - 319627 ms - org.eclipse.core.internal.jobs.WorkerPool.sleep() which I cannot explain but shouldn't do much harm. Time and invocation numbers looked pretty identical when looking at the two builds. Side Note: Starting with I20060322-1335 the JRE on the releng WinXP and Linux machines changed from 1.4.2_08-b03 to 1.4.2_10-b03.
I spend some more time on this and I profiled N20060411-0010 compared to Eclipse 3.2M3 for org.eclipse.ant.tests.ui.editor.performance.OpenAntEditorTest#testOpenAntEditorNoFolding(). Running on my machine (WinXP, JDK 1.4.2_08, 3ghz, 2.5 GB) I see little regression: 3 to 5%. Why the huge difference from the releng perf tests? For releng: org.eclipse.ant.tests.ui.editor.performance.OpenAntEditorTest#testOpenAntEditorNoFolding() (Win XP Sun 1.4.2_08 (3 GHz 2 GB)) is showing a 75% regression?? http://fullmoon.rtp.raleigh.ibm.com/downloads/drops/N20060411-0010/performance/eclipseperfwin2_R3.2/org.eclipse.ant.tests.ui.editor.performance.OpenAntEditorTest.testOpenAntEditorNoFolding().html#Elapsed%20Process Dani, I do not understand your comment on the switch to 1.4.2_10...this is for running the perf tests? The pages are all still claiming 1.4.2_08
>Dani, I do not understand your comment on the switch to 1.4.2_10...this is for >running the perf tests? The pages are all still claiming 1.4.2_08 Yes, saw that one as well. It looks like the new version is only used for the correctness tests.
no better understanding and the results are still bad. Opportunity for improvement in 3.3
nothing more planned here.