[01:16:18] the fix is ready for review now [02:13:45] great work, Krinkle [12:42:08] https://gerrit.wikimedia.org/r/c/mediawiki/core/+/1148315 is now live, looks like it's a ~30ms improvement for p75 and p95 latencies [13:17:57] mszabo: yep! huge drop at https://grafana-rw.wikimedia.org/d/QLtC93rMz/backend-pageview-timing?orgId=1&from=now-24h&to=now&timezone=utc&var-platform=$__all [13:18:16] 70ms drop for mobile pageview p75 [13:20:27] Krinkle: nice! I'm still surprised over the larger impact on mobile because the transformation is the same for both desktop and mobile [13:22:27] ^ this. i do not understand this. i'm happy about it, mind you, but i do not understand it. [13:23:55] one reason i could see, and i hope it's not the case, is if SOMEHOW we run that transformation twice on mobile [13:27:27] could it be that this transform is the difference between zero and non-zero calls to a thing that's lazily computed? [13:27:55] something that we don't trigger on canonical/desktop but do on mobile/minerva [13:34:11] hmm as far as I see, deduplicateStyles is part of the default pipeline, and it shouldn't be overridden in this case [15:46:35] I love that every time I run the installer I can find numerous bugs [19:34:34] ok I'll bite: which bugs did you find this time? Reedy [19:34:45] filed in phab :P [19:36:23] had to hunt around a bit to find it [19:36:45] that's...special [19:37:13] I guess, how do you get CI to check that sort of thing though