[grt-talk] Re: CVS commits

[Nikodemus Siivola]

On Sun, 13 Apr 2003, Simon Adameit wrote:

I put this on the list as well. (This is about strange artefact with the
new plane code where all the planes are represented by ground-plane that
Simon is working on).

> I just started working on them and ran into a strange problem:
> When I render classic-demo with this plane code the checker pattern gets
> inverted and there are random dots in a circular pattern.

That's a wierd one!

It's caused by the discontinuiety in the checker pattern at ground (or any
unit level). Before it was possible to counter this by moving the plane of
ground-level slightly, but the new design results in this being handled in
the transformation. The transformations end up being done to the sampled
point as well for determining it's color:  so samples are always actually
taken at ground plane -- at the singularity.  :(

The easy fix is to shift the checker-pattern by +grt-epsilon+ in each
direction. I've done that, along with a couple of other minor cleanups in
the CVS. Simon: can you please verify that it solves the problem and then
commit the new plane.

The corrrect thing would be to not apply the "pre" transform to pigment
samples, nor to anything except intersection tests: we have to rewire the
architecture to hold separate matrices for intersection test (which include
both the base-transform and user applied transforms) and just the user
applied transforms for texturing stuff.

Alternatively we could take the stance that procedural texture-space is
always oriented according to each object (which it is now).

Cheers,

  -- Nikodemus