> > On Fri, Feb 10, 2017 at 11:17:44AM -0500, Frediano Ziglio wrote: > > > > > > On Fri, Feb 10, 2017 at 10:36:10AM -0500, Frediano Ziglio wrote: > > > > So are you saying you prefer an heavy unsafe signal instead of a > > > > function call just for statistics that's disabled by default? > > > > > > My desktop is built on such signals and works just fine, thank you ;) > > > > These are patches for spice-SERVER, not spice-DESKTOP, server > > code require more performance and safety, I agree with Gnome > > to use such code for desktop and I'd use too. > > A crash on server is a DoS or a security issue for instance. > > Once you tested your code, signals/callbacks are not really a safety > issue, you'd catch fairly quickly memory corruptions related to that... > > > About performances GStreamer that use GLib for some stuff recommends > > to not use signals as too slow but allows to register callbacks. > > GStreamer has very strict requirement on performance indeed, for some > stuff they have their own GstMiniObject rather than using GObject as > object instantiation was a significant performance hit for them. > In the specific case of stat gathering when sending data on the network, > you probably have a point that we should be careful that this is not too > costly (and g_file_copy_async uses callbacks in a vaguely similar > situation). > However, in the general case, when some of notification is needed, I'd > still use the generic infrastructure which is provided to us by GObject > rather than using callbacks everywhere. > > Christophe > I think we went a bit OT a bit too generic and a bit too grumbling. Apology from my side. On the specific case I think we agree that is a bit heavy a signal for statistics code that's compiled out by default. Frediano _______________________________________________ Spice-devel mailing list Spice-devel@xxxxxxxxxxxxxxxxxxxxx https://lists.freedesktop.org/mailman/listinfo/spice-devel