On Fri, Apr 01, 2022 at 06:08:14PM +0200, Johan Corveleyn wrote: > I suppose the "modified for better performance" refers to some > optimisations done by Morten Kloster, who then later submitted the > patch adding this comment in r1128862. His optimisations were more > related to the LCS algorithm (further reducing the problem space, and > providing early exits in certain cases or some such). > > The core LCS algorithm in libsvn_diff/lcs.c was from way before my > time, and according to 'svn blame' was mainly written by Sander > Striker. I don't understand it fully either :-), but I always assumed > it was basically some clever implementation of Myers' algorithm.
Yes, that is my understanding as well. I believe this performance problem has always existed. I have already tried reverting optimizations made over time, and it did not make any difference with the problematic test data I have.