On Wed, 2004-07-21 at 16:12, Christian Neumair wrote: > Is anybody of you already working on a XCB/XCL [1] GTK+ backend? If not, > I'd like to volunteer for adding such a backend. Although I'm not yet > experienced with that kind of low-level stuff, I hope to gain enough > experience to get good results quickly. To my knowledge, nobody is working on this. So, if you want to start, great! A few things I should mention though: - Xlib is implicitly part of the GDK API/ABI. I don't think we could ship a libgdk-x11-2.so.0 that didn't link to an Xlib binary compatible with the current Xlib. Since producing such a Xlib is part of the goals of the Xcb project, this is not a fundamental barrier, but it does limit the advantages you can get in footprint. - If you want to look at the advantages you can get with a compatible backend, you'd probably want to start off by rewriting gdk/x11/gdkasync.c to use Xcb. Then do some protocol traces (using xmon or whatever) and see what remaining excess roundtrips could be improved. (Though I think we got the main ones already) - Writing an "incompatible" backend - an Xlib-free gdk-xcb-2.so.0 - is probably a better test of Xcb, though I'm not sure it's immediately useful for applications. There are some other interesting things you could fool around within a GDK backend that broke hybrid GDK/Xlib application compatibility ... e.g., getting rid of server-side subwindows altogether. But it's probably best to concentrate on one thing at a time. Regards, Owen
Attachment:
signature.asc
Description: This is a digitally signed message part