On Wed, Jun 2, 2021 at 4:05 PM Daniel Shahaf <d...@daniel.shahaf.name> wrote: > > Stefan Sperling wrote on Wed, 02 Jun 2021 12:20 +00:00: > > The test suite is passing, which implies that trivial diffs aren't affected > > by this change. I expect that most, if not all, diffs which people actually > > want to read will remain unaffected. But I cannot tell with 100% certainty. > > Not with 100% certainty, but you could run «svn diff -c $i ^/» on the > last few thousand revisions of /repos/asf, diff the resulting diffs, and > review the non-empty interdiffs.
Another thing I could try is running my old "blame huge XML file with lots of revisions", which I used to do when I worked on diff optimisations back in 2011 (prefix/suffix-scanning and some other things -- also some other people back then did some other optimisation contributions which I double checked this way). I'm guessing my huge XML file is not huge enough (or our diffs not pathological enough) for this limit to have much impact, because after the 2011 optimisations I can't remember running into longer times than say 5 seconds or so per diff. Or otherwise, I can at least give some feedback about a good "effort limit" with a real-world case where we still consider the actual diff output somewhat important (for assigning blame :-)). I'll try to give it a go in a couple of days (but I'll have to dust off my local build first). BTW: for whoever is interested, some discussion about this on IRC just now: https://colabti.org/irclogger/irclogger_log/svn-dev?date=2021-06-02 Feel free to join on #svn-dev on irc.libera.chat ... -- Johan
