[Mesa3d-users] More depth buffer troubles...

[Ross G. M. <rg...@cs...>]

Hey All,

I'm getting some bad pixels when I try to do some rendering and I need a 
little help sorting out what's wrong.  Here's a pseudocode representation
of what I'm trying to do:

        <Initial Mesa setup (lighting, model-view and projection matrices,
        etc...)>

        <clear the buffers>

        glDisable( GL_LIGHTING);
        glColorMask( GL_FALSE, GL_FALSE, GL_FALSE, GL_FALSE);
        glDepthFunc( GL_LESS);
        glDepthMask( GL_TRUE);

        <Draw lot's of polygons>

        glEnable( GL_LIGHTING);
        glColorMask (GL_TRUE, GL_TRUE, GL_TRUE, GL_TRUE);
        glDepthFunc( GL_EQUAL);
        glDepthMask( GL_FALSE);

        <Draw those same polygons again>

        glFlush();


Is there some reason why this shouldn't work?  (I know, I know - it would 
be a lot more efficient to update the depth and color buffers in one 
pass.  For various and sundry reasons - having to do with the fact that 
I'm eventually going to be doing this rendering on a Beowulf cluster - I 
can't.)

What I'm getting back is an image with several bad pixels. (They're black 
when they should be colored.)  There's not very many - maybe two dozen out 
of 250000 - but it's enough to make the image look bad.

I did some further checking and compared the depth buffer from that 
algorithm (obtained by doing a glReadPixels with unsigned short as the 
data type) with one obtained after drawing the same scene with a 
single-pass algorithm.  What I found was that the bad pixels had values 
that were 1 less than their corresponding value in the depth buffer from 
the single-pass algorithm.

The off-by-1 thing makes me think this is some kind of roundoff error.  
What really puzzles me is why enabling/disabling lighting would affect 
the values that are written into the depth buffer.

Anyone have any thoughts?

Thanks,
-Ross