The four attributes a "surface" graph may read — exactly R-19's four fixed
vertex locations (0 position, 1 normal, 2 uv, 3 colour). §53's remaining
standard attributes (tangent, secondary uv, joints/weights, instance
transform) are unnameable until the packets that add them land — the same
closed-union staging, applied to the attribute set.
In the "screen" domain only "uv" is nameable, and it means the pass's
own normalized coordinate — 0..1 across the destination surface. This is
a recorded deviation from RFC 0001 §5, which rejected every attribute in
the screen domain: without a coordinate the domain cannot express even a
copy, a full-screen pass does have a uv (the one varying its triangle
already interpolates), and "the surface parameterization of the thing being
shaded" is what "uv" means in both domains. "position", "normal" and
"color" stay rejected there — there is no mesh.
The four attributes a
"surface"graph may read — exactly R-19's four fixed vertex locations (0 position, 1 normal, 2 uv, 3 colour). §53's remaining standard attributes (tangent, secondary uv, joints/weights, instance transform) are unnameable until the packets that add them land — the same closed-union staging, applied to the attribute set.In the
"screen"domain only"uv"is nameable, and it means the pass's own normalized coordinate —0..1across the destination surface. This is a recorded deviation from RFC 0001 §5, which rejected every attribute in the screen domain: without a coordinate the domain cannot express even a copy, a full-screen pass does have a uv (the one varying its triangle already interpolates), and "the surface parameterization of the thing being shaded" is what"uv"means in both domains."position","normal"and"color"stay rejected there — there is no mesh.