Community
Participate
Working Groups
After having run JDT/Core perf tests on our local Linux test box, I noticed a thin regression on FullSourceWorkspaceFormatterTests.testFormatDefault() test. Here are the numbers for this test: M1 0.36% M2 0.28% v_815 0.37% v_816 0.40% v_817 0.09% v_818 -0.21% v_819 0.05% v_820 0.17% v_821 -0.00% M3 -1.79% v_823 -1.80% v_824 -2.88% v_825 -2.36% These numbers are the average of 20 runs (except for v_825 - only 8 - as tests are currently running for this version) and each run has a standard error around 1%. So, it seems that a change between v_821 & v_822 made the formatter a little bit smaller and should be investigated... Note that, depending on the test box, this test lasts between 1 second and 1.5 second. So, this regression should not be noticeable by users. Honestly, it was impossible to see it on Platform/Releng perf results as it was run only once per build... That's why I opened this bug with a 'minor' priority'.
Looking at difference between those two JDT/Core versions, it looks like the fix for bug 203304 may be responsible of this regression...
Not a surprise. This change is required to fix some other performance issues, but it penalizes a little the case where there is only one region.
This regression was fixed while activating the new comment formatter in 3.4M7...
Looking at 3.4M7 performance resutls, there's now a gain between +1.8% and +5.0%