when for example the download-widgets window causes a deadlock in plasmashell (see bug #337077) and the user tries to close the apparently unresponsive window via the window decoration kwin will detect that the application doesn't actually quit (what with it being in a deadlock) and offers to kill it. for plasmashell this however has the side effect that the user is then left without a discoverable way to manually recover (logout/shutdown) and gets to sit in front of a black background, crying (most likely). (relevant invalid kwin bug #337163) Reproducible: Always Expected Results: Plasmashell should restart itself when killed by kwin. In fact perhaps one should think about always forcing a restart of the shell when killed by means other than an actual sig quit?
Created attachment 87608 [details] screenshot
perhaps the behavior is correct, but in case it's plasmashell (and only plasmashell) it should restart it right after haviung killed it?
Please note the kwin bug report. They don't want to support client specific workarounds which seems reasonable to me. Generally speaking I can totally see this as a plasma specific problem, much like pulseaudio, plasma probably should restart just about always unless instructed not do so or having already done so a number of times in the immediate past (i.e. to prevent getting stuck in a restart loop). Simplest approach to this probably would be a kded module that watches plasmashell and klaunches it when it disappears. (FTR: the way pulseaudio achieves respawns is by having libpulse talk to the daemon and if there is none it will try to launch one, not sure that is a viable scheme for plasma though)
Harald is this behavior still present in current versions?
Yep. - open wallpaper config - killall -STOP plasmashell # to simulate a deadlock - click close on wallpaper config - kwin offers to kill plasmashell and leave you stranded
Confirmed on KDE Neon User Edition. I'm not even prompted to kill plasmashell, it just kills everything and I'm left with my konsole.
The real problem here seems to be that Plasma does not automatically restart itself when this happens, as it does in other cases.