Michael Phipps wrote on Mon, 09 Jun 2003 19:01:31 -0400: > I didn't consider this. Given the 80 bit nature of the FPU and the (fairly) > small numbers that we are using, it wouldn't seem like it could ever be a > more than 1 pixel error. I did see differences between the two algorithms, > but I really had/have no real way to prove error. That sort of rounding problem shows up as gaps between polygons in 3D systems, since the edges are usually placed using a line drawing algorithm. You want the line to be exactly the same, no matter what the order of the endpoints is. It's a bit trickier since the vertices are floating point numbers until the very last transformation step. Takes a lot of care to get it "pixel perfect". - Alex