Top-most windows

[Keith Packard]

On Wed, 2006-01-18 at 14:45 -0800, Deron Johnson wrote:

> I'm not convinced that the Nvidia DDX is the one that is busted. There
> are no specific semantics in the DDX porting layer that specify when
> DID painting should and should not occur. The Nvidia DDX is doing
> nothing wrong when it is painting DIDs in the ValidateTree and
> HandleExposures functions.

Sure, but PaintWindowBackground happens to be the 'obvious' place as it
is always presented with a minimal region needed to adjust the DID area.
  
> Could you remind me what EXA is and how it will fix this problem? I
> haven't been following the EXA stuff very closely. Is EXA going into the
> proprietary ATI drivers, or just the DRI drivers?

The CompositeWrapper code was written because XAA (and other drivers) 
didn't use the macros and functions provided by the DIX layer to access
the correct pixmap for each window. Instead, they assume that all
windows live in the screen pixmap. This breaks Composite which places
windows in separate pixmaps. To fix this, CompositeWrapper presents the
underlying DDX with the pixmaps themselves, rendering commands are
suitably adjusted to offset and clip to the window location within the
pixmap.

It's a horrible kludge, but needed by XAA-based and the closed source
drivers.

EXA, on the other hand, works with Composite without needing the
CompositeWrapper layer. This eliminates all kinds of nasty
window-painting related issues. As CompositeWrapper hides windows from
the driver, PaintWindowBackground is not visible to the driver any
longer, replaced with the usual painting operations involving pixmaps
and GCs.

> If one had access to the DDX code one could modify the following DDX
> functions to not paint DIDs for a composite-redirected window:
> ValidateTree, HandleExposures, PaintWindowBackground, PaintWindowBorder.

Actually, I think you need deal only with CopyWindow and
PaintWindowBackground/Border, and in those cases, the only check needed
would be to paint DIDs only when the window pixmap was the same as the
screen pixmap. No Composite-knowledgable code should be needed at all.
 
-- 
keith.packard at intel.com
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: This is a digitally signed message part
Url : http://lists.x.org/archives/xorg-arch/attachments/20060118/02e17c2e/attachment-0001.pgp