Le 15/01/2012 22:20, clipka a écrit :
> The root cause has been identified by now; it only shows when a ray
> happens to hit the image map extremely close to the image border (about
> a trillionth (1e-12) of a pixel off in this case), and is due to
> precision issues apparently not considered by the original code author.

I guess it's yet another case of computer ending with mathematically
impossible result, like computing the dot product of two unit vectors
and being aware that the result might indeed be bigger than 1, before
trying to compute the arccos of the result to get the value of the
angle. It should not happen... and yet it does in a very few cases.

Good detective work then, dear Clipka!

Post a reply to this message