> Andreas, do you do something unusual to the projection matrix? This is
> known to mess up fog because eye coordinates get wrong.

I don't *think* so since linear fog seems to work fine on my machine (which
obviously is my preferred test platform) but I could never do anything
reasonable with either exp or exp2 fog. Is there a chance that GL likes a
different projection matrix better?! (eventually one that has a -1 instead
of a 1 in m34) The D3D documentation says that a "w-friendly" projection
matrix (which is required for accurate fog) should have 1 in m34 and
currently, that's what we're using. Might be worthwhile trying to find it
out.

Cheers,
  - Andreas