On Friday 08 May 2015 12:49:45 Simon Hausmann wrote: > During the investigation - we also talked about this on IRC - I noticed > that plasma creates a lot of QQmlEngine instances - it seems one per > applet. In the light of each instance coming with its own garbage > collection heap and its own QML type loader thread, this seems rather > heavy. I think that if you can change Plasma to share one QQmlEngine for > everything and isolate your objects in different QQmlContext instances > instead, then you should be able to save _a lot_ of memory and also get > much better performance. I'm mentioning performance because this very bug
I gave a try at sharing an engine at least for the desktop/panel widgets: I've managed to make something load, for instance loading the systemtray widget in plasmoidviewer seems to work and to be functional .. at the highly unscientific memory consumption estimate of top, it says memory usage is exactly half :O so, if this is a figure that will have some foundation in reality, this is a road *definitely* to be taken But now i've hit a blocker that i'm not sure makes it possible to change it in a compatible way: It relies on QQmlAbstractUrlInterceptor to change some accessed paths in a way that's specific of every different plasma widget. so with an engine per applet it was easy: every engine had its own QQmlAbstractUrlInterceptor instance.. now that there would be a single global interceptor instance, i can't know from within the interceptor what is the applet that is trying to access a file (or even from what QQmlContext that access was done) probably could have known that using QQmlAbstractUrlInterceptor was asking for troubles, but... any idea? :) -- Marco Martin _______________________________________________ Plasma-devel mailing list Plasma-devel@kde.org https://mail.kde.org/mailman/listinfo/plasma-devel
