Community
Participate
Working Groups
Since JDT/Core v_756, we observed a regression around 2% for Batch Compiler performance tests. Not all tests are concerned, testCompileJDTCoreProjectJavadoc and testCompileJDTCoreProjectAllWarnings seem to be flat.
Seems to be an impact of fix for bug 185768. Doc comment support is now activated when annotations are processed but as this option is set by default, the doc comment support is now always activated for the batch compiler hence the performance regression in our tests...
Created attachment 68369 [details] Proposed patch I post estimated perfs numbers when running local tests will be finished
+1. Process annotation should never be set for compliance < 1.6.
Sorry. I withdraw my +1. I checked too quickly. If the option -proc:none has been set, we should not reenable the annotation processing. So we should add a check to see if the option is already disabled. I'll provide a new patch.
Created attachment 68381 [details] New patch
Frédéric, please review.
(In reply to comment #6) > Frédéric, please review. > +1
The comment should reference both bug 185768 and this bug. Otherwise +1.
I got following numbers while running local tests: Proj | jdt.core | jdt.core | jdt.core | jdt.core | swt sett. | No_Warn | Default | Javadoc | All_Warns | Default --------|----------|----------|----------|-----------|-------- v_755 | 53315 | 51834 | 54310 | 63771 | 23933 patch | 53412 | 51840 | 54246 | 63850 | 23978 These numbers show that we are back to M7 numbers with this patch.
I got a mid-air collision which seemed to have flushed review flags! I reset them...
Olivier, Jerome, can you set again the review flag, please? Thanks
I'll release for tonight's build.
Released for 3.3RC2.
Verified for 3.3 RC2 with performance tests results: http://fullmoon.ottawa.ibm.com/downloads/drops/I20070524-0010/performance/org.eclipse.jdt.core.php?