Hi Japanese programmer guys!

Gustav Taxen is using a circular texture to mask dual-paraboloid maps.
His approach, however, inevitably leads to a certain artifact.

In a +Z paraboloid map, infinite homogeneous texture coordinates such as
(s, t, q) = (1, 0, 0) or (-2, 1, 0) actually represent the identical -Z direction.
Therefore, given that the three reflection vectors at the vertices of a triangle
are near to (0, 0, -1), each of the corresponding non-homogeneous texture
coordinates gets extremely large. As a result, they can surround the circle of
the mask in some circumstances.

Since the rasterizer of today's GPUs always interpolates texture coordinates
linearly, values interpolated between such coordinates can fall into the circle,
which means that the mask has a small "hole" in the -Z direction.

Such an artifact can be observed in Kano's OpenGL sample, which I guess
justifies the fact that he is using the same method as that of Gustav.

What is important to note here is that, should you choose the view vector
as the -Z axis of the environment map space, the hole would be invisible.
However, it might contradict view-independency of dual-paraboloid mapping.

What do you think of that?