In the event of a compositor crash or error these changes will reconnect to the new compositor and re-fetch seat information, outputs and then re-send any windows, active pointer constraints and alike. From an X11 POV it's as though we got a flurry of events at once, and then the X11 compositor got replaced.
See wayland-devel: https://lists.freedesktop.org/archives/wayland-devel/2021-August/041959.html
This both has a few code quirks to iron out and discuss plus it uses unreleased dependencies. Posting as Ofourdan wanted to share some comments.
You maybe don't need mesa if you force shm mode.
They need some touchup before making MRs.
Rootless mode: Rootless mode is challenging to test, as obviously you can't use -wm fd or -listen fd, so it requires compositor changes delegating the launching elsewhere. These are not shipped anywhere.
Rootfull mode is easier to test xwayland changes without any compiling.
kwin_wayland_wrapper (any stock release within the last 9 months)
WAYLAND_DISPLAY=wayland-1 Xwayland :2
killall -9 kwin_wayland
Code wise I have some slight issues with the lifespan of Device objects outliving the Seat and the cursor sits in a weird middle ground across these that I'm not happy with. Rest is relatively trivial I think.