From looking at it, it seems that it doesn't - that it is running the app in the server and updating the screen in HTML5 canvas. It would be cool if it compiles the code to Javascript though... so that you get complete client side platform independent apps out of the current entire Gnome repository!
However, a project like GCC could certainly do that. If someone really wanted that to happen, they could add a Javascript code generator to GCC and pair that with GTK's HTML5 backend to produce something like what you want.
This would be much more interesting if the GTK components map to real HTML elements.
This is where we're eventually headed, the mozilla project "chromeless" is an interesting experiment down the path. IIRC It gets rid of XUL for the application chrome and uses html/js for the entire program gui.
This is really hard to do wrt skinning. The only thing you really want is text selection to work properly for cut-and-paste. A good question to ask is what does the Canvas backend for GTK for cut-and-paste now?
I've been looking at Cappuccino since it does support stuff like tables with > 10.000 entries and 'complex' controls.
Now with GTK+ HTML5 support this might also be an avenue to explore.
And yes, there are absolutely cases where it would be easier to create a GTK application than an HTML application. You don't need to worry about choosing a widget library, for one, and placement is very straightforward.
Basically, toplevel windows (i.e. GtkWindow) act as proxies for remote browsers.
I really want to start playing with this, is there any reason why it wouldn't work with the python GTK bindings?
