I've been talking to Joshua by email but I think it would be better to discuss here so that others may participate.
I've implemented support for cylindrical projection, but I have some questions regarding spherical:
The picture is nice but it's missing the location of point 3, which I think should be defined in the standard. Without it it's unclear if one should do (p3-p1)x(p2-p1) or (p2-p1)x(p3-p1) to calculate the horizontal plane normal. If the plane normal isn't defined then some implementations could "flip" the textures and draw them upside down.
That led me to think, it might be better to define the sphere similar to how the cylinder is defined: p1 = center, p2 = front and p3 = top. This way part authors could have an ellipsoid to define the projection and different scale values for u and v. Even if the different scales are not useful, having the up direction defined would be better IMO.
Also, I think the spec should state that planar projected textures are clamped (it's implied given that it says uvs go from 0 to 1). Maybe adding an extra parameter to allow uvs to wrap would be useful so that we can have tiled textures, but I can't think of any parts right now that could use it besides a few checkerboard patterns.
I've implemented support for cylindrical projection, but I have some questions regarding spherical:
The picture is nice but it's missing the location of point 3, which I think should be defined in the standard. Without it it's unclear if one should do (p3-p1)x(p2-p1) or (p2-p1)x(p3-p1) to calculate the horizontal plane normal. If the plane normal isn't defined then some implementations could "flip" the textures and draw them upside down.
That led me to think, it might be better to define the sphere similar to how the cylinder is defined: p1 = center, p2 = front and p3 = top. This way part authors could have an ellipsoid to define the projection and different scale values for u and v. Even if the different scales are not useful, having the up direction defined would be better IMO.
Also, I think the spec should state that planar projected textures are clamped (it's implied given that it says uvs go from 0 to 1). Maybe adding an extra parameter to allow uvs to wrap would be useful so that we can have tiled textures, but I can't think of any parts right now that could use it besides a few checkerboard patterns.