Fix initial position of X11 override-redirects

[djpohly]

Presumably this fixes #228. The position of the scene node for XWayland override-redirect windows was never being set.

[emersion]

• When an Xwayland view is created, shouldn't we read back xwayland_view->xwayland_surface->{x,y} if it's unmanaged?
• When set_geometry is triggered, we update view->{x,y} but we don't update the scene node position?

[djpohly]

Good points, both. Updated.

Edit: Further updated to fix a segfault when a window is mapped, unmapped, moved, and mapped again. There's a little bit of null-checking clunkiness here, because set_geometry can happen while a window is unmapped, but the scene node in cage only exists while the window is mapped. An alternative would be for each view to have an empty "container" node that always exists, then add/remove the surface tree as a child of that container on map/unmap.

[Supreeeme]

Is there anything blocking this from getting merged? It fixes positioning issues with Steam's dropdown menus.

[emersion]

Hm, should we add a if (!xwayland_view_should_manage(view)) here?

[djpohly]

Then it wouldn't cover the case where override_redirect is set between creating and mapping the window, e.g.:

Window w = XCreateSimpleWindow(dpy, root, 100, 100, 50, 50, 3, 0xffff00, 0xff00ff);

XSetWindowAttributes attrs = { .override_redirect = True };
XChangeWindowAttributes(dpy, w, CWOverrideRedirect, &attrs);

// Should appear at position (100, 100)
XMapWindow(dpy, w);
XFlush(dpy);

[emersion]

Hm… What about override redirect changing after the window is unmapped? Sounds like we should probably set up a set_override_redirect listener to handle all cases correctly?

I'd be more comfortable only copying the X11 position when the window is O-R.