Tek wrote:

> Speaking as a programmer this sounds like rather shoddy coding! Okay I'm
> sure values near the limits of floating point numbers would cause
> problems, but the limit is around 1e38, so I'd expect it could get much
> bigger than 1e7 without any problems. Ah well, I'll just use a global
> scale in all my big scenes...

Hmm... It might have something to do with the nature of floating point 
numbers - that a number's accuracy decreases as the magnitude increases 
("the dot is shifted", so to say). Thus, a unit-radius sphere a billion 
units away could be prone to artifacts - but what's the point of using that 
small objects that far away anyway?-)


> BTW, does anyone know if there's a similar lower accuracy limit? i.e. the
> shortest distance below which two surfaces will be judged to be in the
> same place?

There seem to be several different constants to be used with different 
calculations. Particularly, each object type has its own "depth_tolerance" 
value for ignoring intersection points too close to the ray origin. For a 
sphere it is 1e-6, but for a blob as much as 1e-2.

Post a reply to this message