when/if plasmashell gets itself into a bit of a sticky situation kwin will detect this and offer to kill the shell. however neither plasma nor kwin do anything to recover the shell again. this effectively leaves the user on a black background (unless there was a window open anyway) without any discoverable way of recovering things manually (read: logout/shutdown) which is incredibly rude and incredibly bad from a UX perspective. kwin should autorestart whatever it killed or plasmashell should autorecover when killed. Reproducible: Always
This is outside the scope of KWin's killer. The killer is only shown if you click the close button. If one uses the ctrl+alt+esc key sequence it's not the killer at all. Offering a restart for an window which is supposed to be closed doesn't make any sense at all. KWin doesn't detect a hanging application, only a not closing application that is quite a difference.
kwin will detect a pending close request for plasma-desktop in-process diaglogs. this one's not simple. a) kwin cannot restart the hanging client. this does usually not make any sense and we could *at best* (on linux, with mounted /proc) restore the commandline, but not the state (you loose 10 kwrite docs and get one empty window in return) b) plasma-desktop (or anything) cannot restart itself for a sigkill (let's assume it would be technically possible: the user could never again get rid of a plasma-desktop that's life-locked in some plasmoid) -> plasma-desktop would probably require a watchdog kded module it had to deactivate it on sigterm the dog needs to ask before recovering plasma-desktop at least i don't see another way.
plasma bug report is bug #337164 I didn't actually realize that the kwin interaction only kicks in when one tries to close a window, makes more sense that way.