Jasper, I looked through your xdg-shell-rewrite branch and, other than some things that need better docs/rewording, I think it looks OK. In particular, I think we're probably OK removing set_unstable_version in 1.6. Things that need attention (some of this may be a repeat of what Pekka said): - The wording in the xdg_surface description about destruction is contradictory: Either it gets destroyed by the destroy request or by destroying the surface. Pick one (preferably the former). - Incorporate the discussion on maximized. I think we're pretty settled on how that should work now. - In set_parent, you say the parent is raised when the child is raised. I think you want that the other way around. Also, a little more description in general would be good. In particular, a note about destruction order (which I think you want) is better than vague wording about mappedness. Also, what are the symantics of reparenting? Is this tied to the commit in any way? - move: who is responsible for the cursor? Does the surface get a leave event and then the compositor set it? - resize: same question about cursors. Also, how does the client know the event has been ignored? Does it just not get a configure? Is this going to look to look terrible if the client and server are using different cursor themes? TBH, I don't know how solvable that last one is. - The wording on configure and ack_configure needs work. I think you need to be more explicit that when the client receives the configure it needs to repaint with the adjusted state and send an back along with the commit. Also be explicit that the ack_configure is double-buffered like other surface state. - Geometry: First, I think configure needs a see-also pointing to the set_geometry event. Also, this needs to be explicitly stated to be double-buffered with commit. I think popups look OK but I don't have much experience there. --Jason Ekstrand