So I have looked at the code you referenced.  Specifically the code
you made this claim (among others) about:

<perry> one of the nice things about Simon's way is it deals with turbochannel, vesa, etc. cleanly, and will deal with the next bus cleanly too.

You made multiple factual errors:

* There is no new kernel API in that code.

* It does not support other busses at all.

* It doesn't work for anything other than i386.  (The macppc and alpha
  support were both incomplete as posted.)

Furthermore, it assumes that the server is only talking to one PCI
device, which breaks Xinerama/multihead.

In short, this is very much not a step forward, and I'm not going to
spend any more time looking at it.


Meanwhile, with a few small changes we will have XF86 4 working on
macppc, alpha, and perhaps cats.