Tested using Epiphany 3.36.1 with WebKitGTK 2.28.2 and WebKitGTK 2.28.0 on Fedora 32. Using Nouveau graphics with default settings done by Fedora. With WEBKIT_FORCE_COMPOSITING_MODE=1, the scrolling behaviour is fast but abrupt and does not have kinetic scroll. With WEBKIT_DISABLE_COMPOSITING_MODE=1, the scrolling behaviour is slower but has kinetic scroll. On default behaviour without setting any of the above environmnent variables, scrolling is stuttery, lacks kinetic scroll and upon the use of Web Inspector, reveals high CPU usage with most of it being spent on Styles Invalidated and Styles Recalculated (an example website for this was macrumors.com). Websites where the above two things aren't happening (eg. phoronix.com) have smoother scrolling.
In other words: "Kinetic scrolling doesn't work in AC mode"
WEBKIT_FORCE_COMPOSITING_MODE=1 means async scrolling is used. Kinetic scrolling as not implemented for async scrolling in 2.28, but it landed in trunk (r259112) se we can probably backport it.
(In reply to Carlos Garcia Campos from comment #2) > WEBKIT_FORCE_COMPOSITING_MODE=1 means async scrolling is used. Kinetic > scrolling as not implemented for async scrolling in 2.28, but it landed in > trunk (r259112) se we can probably backport it. Added to backports list. This was a serious regression, it should not have happened in stable. :/
Well, it only affects people using the force AC mode, so I don't think it's so serious. I agree with backporting it, of course.
(In reply to Alexander Mikhaylenko from comment #1) > In other words: "Kinetic scrolling doesn't work in AC mode" No, this is not true. It doesn't work when AC mode is forced. When ondemand is used, AC mode doesn't use async scrolling.
Um... OK, that's an odd choice. I thought https://gitlab.gnome.org/GNOME/epiphany/-/issues/1210 was a duplicate of this issue, but it must not be. We'll probably report a new bug then.
(In reply to Michael Catanzaro from comment #6) > I thought https://gitlab.gnome.org/GNOME/epiphany/-/issues/1210 was a > duplicate of this issue, but it must not be. We'll probably report a new bug > then. This is not related to AC mode after all, sorry for the noise.
Looks like it's been fixed in https://bugs.webkit.org/show_bug.cgi?id=210382?(other than the regressions, but probably better to track them there)