RESOLVED FIXED 271698
[Skia][GLib] Misplaced tiles when scrolling drakenhh.com
https://bugs.webkit.org/show_bug.cgi?id=271698
Summary [Skia][GLib] Misplaced tiles when scrolling drakenhh.com
Adrian Perez
Reported 2024-03-26 02:07:47 PDT
Created attachment 470603 [details] Screenshot showing misplaced tiles Steps to reproduce: 1. Have a WebKitGTK build with USE_SKIA=ON (tested with 276677@main) 2. run-minibrowser --gtk https://drakenhh.com 3. Scroll around the page. Typically scrolling all the way to the bottom and then backing up will trigger the issue, which is most noticeable over big images. 4. Hovering elements that cause repaints/recomposition (like buttons) will “fix” the misplaced tiles, but if one keeps scrolling the issue resurfaces.
Attachments
Screenshot showing misplaced tiles (2.57 MB, image/png)
2024-03-26 02:07 PDT, Adrian Perez
no flags
Adrian Perez
Comment 1 2024-03-26 02:10:45 PDT
Talking with Carlos Garcia he mentioned that he's seen similar artifacts in https://app.element.io (the flagship Matrix chat client) but harder to reproduce.
Adrian Perez
Comment 2 2024-03-26 02:24:37 PDT
Checked to affect WPE as well, summary updated accordingly.
Adrian Perez
Comment 3 2024-03-26 02:32:31 PDT
(In reply to Adrian Perez from comment #2) > Checked to affect WPE as well, summary updated accordingly. Also: the issue is somehow more evident passing --use-wpe-platform-api to run-minibrowser.
Adrian Perez
Comment 4 2024-03-26 02:54:00 PDT
(In reply to Adrian Perez from comment #3) > (In reply to Adrian Perez from comment #2) > > Checked to affect WPE as well, summary updated accordingly. > > Also: the issue is somehow more evident passing --use-wpe-platform-api > to run-minibrowser. One more data point: setting WEBKIT_SKIA_ENABLE_CPU_RENDERING=1 seems to sidestep the issue. This is true regardless of the number of threads used for painting chosen by setting WEBKIT_SKIA_CPU_PAINTING_THREADS.
Adrian Perez
Comment 5 2024-09-26 14:37:26 PDT
This has been working fine for a while, both on main and the 2.46.0 release. I went as far as checking that this has been working already at 282416@main, which was the branch point for 2.46 =]
Note You need to log in before you can comment on or make changes to this bug.