Originally Posted by elanthis
The paper is all about how you do 2D on top of 3D. That's what XGL is all about, too - rendering all of the 2D X drawing commands using the 3D hardware.
XGL manages to make those effecets possible by forcing all 3D to be done indirectly; that is, the clients no longer directly talk to the hardware, but instead talk to the X server that talks to the hardware. That's necessary to make a composite manager that uses 3D work.
The work done by IBM, X.org, and Red Hat on the accelerated indirect rendering branch of X.org allowed the current X.org server to support both direct and indirect hardware accelerated rendering. And it does it with much less impact on the X server code, and doesn't require building an entire new set of drivers from scratch like Xegl will.
Allowing direct rendering isn't necessarily a good thing. Do you think it was a great feature that older operating systems like MacOS would allow you to write directly to memory rather than through a protected memory manager?
The X server code for XGL is already written, so it's not like this is a huge problem. Finally, at least on the ATI side, the OpenGL portion of the linux drivers is based off the same code in the windows drivers. I assume nVidias drivers are built similarly. Going forward, it is the opengl portion of the drivers that really matter, and it's not like they would be rewriting the Xegl drivers from scratch.