[cairo] Road to 1.10

[Alexander Larsson]

On Mon, 2009-02-16 at 10:25 -0500, Kristian Høgsberg wrote:
> On Mon, Feb 16, 2009 at 4:42 AM, Chris Wilson <chris at chris-wilson.co.uk> wrote:
> > On Mon, 2009-02-16 at 10:11 +0100, Alexander Larsson wrote:
> >> This is more of a feature request, since I've no time to implement this
> >> myself as things stand. But, after Gtk+ 2.16 is released (should be
> >> shortly now) we will try to merge the client side windows branch. The
> >> client side windows branch of gtk virtualizes all drawing to GdkWindows
> >> such that we can emulate the effects of subwindow clipping on the client
> >> side.
> >
> > I agree that a form of device clipping is necessary to safely expose a
> > sub-surface to a direct rendering client, in particular I've been
> > thinking about this with regards to wayland.
> >
> > So at the moment my thought is basically to add a clip extents to a
> > surface which may be smaller than its true extents. The biggest
> > difficultly lies in applying source clipping. Not too difficult for
> > cairo-drm where we can set up a sampler to a subset of a surface, but
> > for xlib support, I fear we may need to make a copy of the sub-surface
> > for the source.
> >
> > Has anyone else had any thoughts about sub-surfaces?
> 
> In the context of wayland, I've been thinking that we can just always
> allocate a temporary surfaces for the clipped rendering.  We typically
> need to do that for double buffering anyway, but in the case where
> we're emulating a sub-window during the off-screen rendering, we'd
> need to allocate an extra surface.  Having a device clip would remove
> the need for that surface, but I don't know that that's something that
> is worth adding API for.  How often is this needed after all, and how
> big are these sub-windows?  Isn't it mostly gtk+ menus that still use
> child windows?  And of course this could be done for cairo xlib too by
> just allocating a pixmap for the child window.

Everything that scrolls use subwindows, plus GtkEventBox. Plus a bunch
of other widgets.

We already allocate a pixmap when rendering double buffered, true, but
we only allocate one such pixmap per toplevel when exposing, so we still
need to clip in that pixmap when drawing multiple subwindows in that
toplevel.