Nexus 7 2013, Android 5.1.1, todays Nightly. All of the below are regressions in this Nightly. I tested on this URL: http://www.pianoworld.com/forum/ubbthreads.php/topics/1215363/15.html Issue observed: There appears to be a lag between the touch input when scrolling, and the page actually scrolling. I have no good way to quantify it, but it feels like an obvious increase compared to before and makes the browser feel sluggish. It's possible there is no actual lag but the scrolling dynamics and inertia are completely off as compared to before and compared to previous versions.

(In reply to Gian-Carlo Pascutto [:gcp] from comment #0) > are completely off as compared to before and compared to previous versions. That should've said: as compared to other Android apps and compared to previous versions.

I have bug 1229462 for the physics changes, so let's keep this bug for the touch lag.

This might be a result of a recent change we made to the APZ code that wasn't in the Java PZC. Specifically, we suppress touch move events until they are greater than a specific threshold, so that pages that listen for touchmove events and use that to cancel scrolling don't also end up cancelling clicks (bug 1141127, bug 1174532). You can help verify this is the issue by reducing the value of the apz.touch_start_tolerance from the default of 0.222222 to 0.1 (which is what we use on B2G) or even lower if you like and see if that helps. Also related is the apz.touch_move_tolerance pref which is currently sitting at 0.0 on Fennec but is 0.03 on B2G. We should probably just set Fennec to the B2G values for both of these prefs.

We should probably blind test this in Orlando :-)

I think I am noticing the lag. It seems ok after scrolling starts, but the initial frames seem to lag behind. I adjusted the pref kats mentions in comment #3, but didn't seem to help (maybe I didn't restart though). I also wonder if the vsync observer crap is still in play here, since I think we drop two frames before compositing when we have no observer.

Randall, can you see if lower values help here. I've been running with .6 on nightly for a while and it seems to be good. We should have this set as low as possible.

(In reply to James Willcox (:snorp) (jwillcox@mozilla.com) from comment #6) > Randall, can you see if lower values help here. I've been running with .6 on > nightly for a while and it seems to be good. We should have this set as low > as possible. Which pref do you have set to .6? apz.touch_start_tolerance? .6 seem high?

I think he means .06. That was also the suggested value in bug 1230709. I'm fine with making that change.

Stealing, I can do this now.

rs=snorp based on above comments. It's a trivial change but I've broken stuff a lot lately so I'm doing a try push first: https://treeherder.mozilla.org/#/jobs?repo=try&revision=8e2f13828170

I did mean .06. Thanks. We should do a measurement to see if this could be made lower, though.