I had a thought. And yes, this may slow the engine, but for some 
applications it may be a lot more accurate. Things we know, or should 
know:

1. Direction of the ray.
2. The 'object' that a point the ray hits belongs to.
3. Where in 3D space that 'should' place the surfaces in relation to each 
other.

What is needed is to identify which object when hit is 'behind' the ray 
at the point of intersection. If you know that you are leaving object X, 
but the point you hit, due to precision issues, is on Y, then obviously 
those calculations need to be delayed in cases where you will also hit X 
at roughly the same exact point. It would seem to me that the odds of two 
objects being so close together that they don't coincide, but numerical 
precision makes them seem to, is much less often than intentional direct 
contact between surfaces. If you know that they should occupy the same 
apparent space and that one must mathematically appear prior to the 
other, there should be some way to figure out which is which and act 
accordingly, right? Even if this double checking could only be turned on 
manually in certain CSG where you 'know' such a check is needed it would 
help.

Then again, I don't have a clue about the internal mechanics, so maybe I 
am missing something obvious.. It is definitely not something I could 
code and test for myself.. :( And yeah, I know how you get around it, I 
just think a better way has to be possible.

-- 
void main () {

    call functional_code()
  else
    call crash_windows();
}

Post a reply to this message