> On Tue, 06 Nov 2001 10:39:40 -0500, "Greg M. Johnson"
> <"gregj56590[:-0]"@aol.com> wrote:
>
> > This bug-carry-over-from 3.1g could be fixed by changing:
> > PIXEL_COLOR = exp(-d/D) * OBJECT_COLOR + (1-exp(-d/D)) * FOG_COLOR
> > to
> > PIXEL_COLOR = exp(-d/(D+GGG)) * OBJECT_COLOR + (1-exp(-d/(D+GGG))) * FOG_COLOR
> > Where GGG= an infitesimal, just bigger than the smallest number pov can handle
>
> Balancing between floating point accuracy and division by zero is rather not
> very best thing. "The smallest number pov can handle" depends on platform.
> Morover why this addition should be done for all values (even large integers)?
> It is rather bug in parser which allows even "distance -1".

Oh, how about
GGG=1.5*max(smallest_number_for_platform [n])    for all n's?

Post a reply to this message