On 30 April 2012 20:13, Piotr Piastucki <[email protected]> wrote: > Hi, > > The attached patch introduces a new matcher class that provides additional > optimizations more suitable for character-based matching. Some code > refactoring is also included to make easier to extend preprocessing phase in > MyersSequenceMatcher class. The optimization assumes that the minimum number > of matching characters is 3. I ran some simple tests and > InlineMyersSequenceMatcher seems to be up to 8 times faster when the files > are almost completely different.
I just wanted to note that this *is* something I'm looking at. The changes look good, but I'm trying to see whether I can unravel the InlineMyersSequenceMatcher preprocessing to be somewhat more understandable. Also, the attached patch has a bug: InlineMyersSequenceMatcher is derived from difflib.SequenceMatcher instead of MyersSequenceMatcher, which is what was obviously intended. Also, I do see significant performance gains from using the inline optimisations, so if it doesn't affect the quality of matches too much, I definitely want to get it in. cheers, Kai _______________________________________________ meld-list mailing list [email protected] http://mail.gnome.org/mailman/listinfo/meld-list
