This approach should do a better job of picking points
in cases where points are coincident for points lines
and polygons. In that case all three points would be picked
if all three primitives were visible. It relies on the normal
coincident topology support for determining what is visible
and using larger points for lines and surfaces.