RE: [Algorithms] fast triangle-segment test *with precomputation*

[<Chr...@Pl...>]

Charles Bloom wrote:
>Urgh; of course you're right; boxes that are outside but
>near the corners are classified wrong.  I think you could
>probably detect that case and handle it from there; eg.
>when the box straddles two edge planes, look at the vertex
>those planes share, and then test whether the box is away
>or towards the center of the triangle relative to that vertex
>(or something).

I kinda seems that way, doesn't it, but it's not quite that
easy I'm afraid.

Consider all the relative orientations that the triangle and
the box can be in, and I think you'll find that what you'll end
up with is going to be equivalent to doing all the required
separating axis tests.

Let's see... 3 for the edges/face normals of the box, 1 triangle
face normal, 3 triangle edges, and 9 for combination of edges of
both, so 16 axes.

Christer Ericson
SCEA, Santa Monica