Community
Participate
Working Groups
In 3.3 M6, the performance test "Reconcile editor change" is between -2.3% and +0.1%). We should investigate why it is getting slower, and fix reconcile so that it gets faster.
Created attachment 63020 [details] SourceTypeConverter improvement Avoids creating unnecessary strings by reusing the PackageFragment's names to create the import reference.
Created attachment 63021 [details] DefaultCommentMapper improvement Use a stack of siblings (instead of a HashMap). This gains 2% of the reconcile.
Created attachment 63055 [details] DefaultCommentMapper improvement This patch has now 2 improvements: - uses a stack of siblings instead of a HashMap - passes in the line range of the parent node so that line numbers are computed faster
Here are the results I got on our local perf tests (12 runs): Test Reconcile Reconcile Big CU Reconcile+SearchAllTypes v_747 -0.17% 0.44% 1.95% patch 5.24% 0.59% 7.09% I guess the result for Big CU test is not impacted because the CU built in this test has no comment...
Created attachment 63172 [details] DefaultCommentMapper improvement Same improvement as with the previous patch, except that getLineNumber(int, int[], int, int) is moved to Util, and all calls have been changed to use the Util#getLineNumber(...).
Created attachment 63183 [details] DefaultCommentMapper improvement Partial fix of DefaultASTVisitor (to handle MarkerAnnotations)
SourceTypeConverter and DefaultCommentMapper improvements released for 3.3M7 in HEAD.
*** Bug 179690 has been marked as a duplicate of this bug. ***
Verified for 3.3 M7 using build I20070427-0010