https://bugs.kde.org/show_bug.cgi?id=363753
--- Comment #44 from RJVB <rjvber...@gmail.com> --- Maybe that has to do with the kind of signals we're connecting to but I don't understand how this could happen. I don't see how QtCurve could be unloaded without calling the Style dtor; dlclose'ing it seems certainly something that shouldn't be happening in normal applications while they are still running. I've never seen that kind of crash happen with the attached patch but I would guess it must be something else that we should be able to address more properly than with the current cleanup callbacks. Clearly Breeze and Oxygen aren't affected so it must be possible. There's always the possibility of storing the plugin instance in each Style instance, and maintaining a list of Style instances in each plugin instance. If the Style dtor tells its plugin instance to remove it the plugin dtor can know if there are still "open" Style instances and delete those first. I think that's cleaner (and easier to debug!) than the current approach with a closure (lambda). Have you ever looked into this phenomenon that multiple Style instances are created and tried to figure out which one is actually used in the end? If that's always the last one we could disconnect all existing Style instances from DBus whenever a new Style instance is created. But we really should try without all that complexity first. There must be a reason Breeze and Oxygen don't need it. In the worst case we get (highly) sporadic crashes rather than regular/systematic crashes like we get now, which is a form of progress. Come to think of it: both styles only connect to DBus signals that are carry their own signature and which they send out themselves. Maybe that's where we need to rethink matters. QtCurve connects to more signals than both styles together. -- You are receiving this mail because: You are watching all bug changes.
