Community
Participate
Working Groups
Verifying performances for 3.6M7 candidates builds, I observed a big regression on formatter tests FullSourceWorkspaceFormatterTests#testFormatDefaultBigFile(): around -30%! Note that there's a similar regression on testFormatDefault() but less visible.
99.99% chance that this regression is a consequence of the fix for bug 59891... I see if something can be done for RC1...
Created attachment 166863 [details] Proposed patch This patch adds to performance improvements: 1) It disable the fix for bug 59891 in the case when there are a string concatenation inside the message send. In the peculiar case of the formatter perf test showing the regression, there are tons of this with complex string concatenation and the formatter can raise until 280 AlignmentException before getting a formatted string... Relaxing the previous fix for that does not have any impact on the massive formatter tests which confirms that the performance was really a corner case! 2) It tries to reduce the instanciation of StringBuffer to avoid unnecessary time and garbaged memory consuming. Point 1) reduce the regression of around 75% and 2) around 25%, which means that with this patch we get back to M6 performance levels for our peculiar perf test case.
Olivier, please review, thanks
+1
(In reply to comment #2) > Created an attachment (id=166863) [details] > Proposed patch > Released for 3.6RC1 in HEAD stream
There is still 8% performance degradation but it is much better than 30%. Hence verifying. Verified for 3.6RC1 using build I20100513-150
Verified.
*** Bug 230188 has been marked as a duplicate of this bug. ***
