I'm noticing that OSL does not seem to use the surface curvature when
calculating the differentials of reflected or refracted rays. This is
especially obvious when rendering a magnifying glass over a finely detailed
image texture with mip maps, the texture lookups in refracted rays have
much larger ds and dt than they should. Are there any plans of extending
the BRDFs to take dNdx and dNdy into account?