https://bugs.freedesktop.org/show_bug.cgi?id=73444
--- Comment #1 from Michel Dänzer michel@daenzer.net --- (In reply to comment #1)
12:30 < pq> so if super+s, or zoom, or having it in a window makes the problem go away, then the fundamental difference is likely that when weston composites, everything goes well, but when weston tries to scan out the client image directly, something breaks.
Yes, it sounds like when weston tries to scan out the client surfaces directly, you're intermittently seeing the wrong surface being scanned out, or the right surface at the wrong time.