Since integrated Parsing API, there has been a regression in up-to-date check.
See current data at http://wiki.netbeans.org/JavaScanDesc#section-JavaScanDesc-UpToDateCheck
Details about 6.5 implementation:
I'm looking into this.
I can confirm that touching class files is not a problem. They aren't touched at all during up-to-date check for roots without change (the same behavior as in
6.5 even without md5). Tomas was right.
Automated test confirmation - RepositoryUpdater.testFileTouch(). To ensure, I has finished also manual testing.
This should now be fixed. Please see issue #162706.
*** This issue has been marked as a duplicate of 162706 ***
I can confirm that it helps, but there are is still a regression in other areas. Reopening.
Created attachment 81988 [details]
Profiler snapshot showing many calls to store() method
There are about 918 calls of org.netbeans.modules.parsing.spi.indexing.support.IndexingSupport$MyAccessor.store(),
cosuming about 15% of scanning time. There is no change on source roots, expecting there is no need to flush the same
indexes to disk every time. Vita works on it.
See profiling snapshot above for further details.
After applying the following two changes my profiler shows approx 20-30% improvement in the up-to-date check times.
Could you please retest? Thanks
I don't think we can get much better then this in the indexers part. The most of the time is now spent in computing
classpaths dependencies (approx 75%) and so any further optimization should probably be targeted there.
I just found a problem in PathRecognizers which was contributing to slow dependencies resolution. Fixing it slightly
improved the speed (approx by 25%). This gave me an idea that we could try to make it even faster (especially when
running the full IDE with all the ergonimized features turned on). It's unlikely to have a big effect when only java
cluster is on though.
Created attachment 82177 [details]
Verify store() is not called when no change
Data are much better than in beta, see details at:
Tested on custom build 090515. Considering the issue as fixed and verified although the data are not still perfect. Separated issue will be filled if there is
some space for immediate improvements. Thanks.
Integrated into 'main-golden', will be available in build *200905151401* on http://bits.netbeans.org/dev/nightly/ (upload may still be in progress)
User: Vita Stejskal <firstname.lastname@example.org>
Log: #164009: do not flush lucene indices when there were no changes