[18:25:27] MaxSem, thanks for investigating further. :) [18:25:28] What are the implications of those perf test results? (Idk if "2 times" is [acceptable/bad/terrible]...?) [18:25:28] I'd like to keep pushing on the idea of using (or at least taking a longer look at) Extension:WikEdDiff - at least as a alternative/optional diffengine for powerusers (related to Jon's dropdown idea from the hackathon https://phabricator.wikimedia.org/T121469 ) [18:25:28] Basically... I hope we can use any good ideas/code in that extension (either by using it directly or via porting the features), and perhaps in the long-term to encourage him to contribute to the official diff engine(s) by giving his efforts the recognition they deserve... Because that feature-list looks impressive (?) https://www.mediawiki.org/wiki/Extension:WikEdDiff#Features even if the current version isn't directly usable. [18:25:56] so: [18:26:13] average diffs are fast enough [18:26:46] its like 1-2% of the slowest ones that are ridiculously slower [18:27:22] coincidentally, the more complex is the diff, the more you need it to be tetter to read well [18:27:37] s/tetter/better/ [18:28:40] also, diffing in php is 1-2 orders of magnitude slower than in C++ which we're using in production [18:29:33] does that mean more stress on the servers, or just a slower pageload time for the user? [18:29:39] from that comes the conclusion that I don't see us ever using WikEdDiff unless it's ported to C++ [18:29:57] stress on servers is not an issue [18:30:15] diffs that take an hour is an issue [18:30:21] ah! [18:30:34] (hour is not an overexaggeration!) [18:32:18] and porting 4000 lines of PHP to C++... [18:34:26] so, IIUC, WikEdDiff is probably stuck as a gadget forever (unless it gets ported to C++), and the most practical path would be to encourage him to take a look at contributing to the current C++ diffengines? [18:35:00] an insane solution would be to try it as a Node service [18:35:57] the issue is that even in a separate service, you still want it to perform well [18:37:08] so my approach to this whole "our diffs suck" issue is pragmatic: instead of 3 places to change stuff, make it 2 [18:37:52] then ramp up improvements that don't require a rewrite yet improve diffs [18:38:16] nod. (thanks again for all the details. :) [18:39:17] ifI were working on it full time I would have been able to do more and try more radical solutions, of course [18:40:30] * quiddity clones MaxSem. [18:43:17] how many votes do we have, meanwhile? [18:44:24] mhm, 102 votes with people still slightly preferring wikidiff3