I also stumbled about something else. Reading this section of the FAQ…
NURBS: Standard CAD that is based on NURBS has several drawbacks which make the modeling process tedious and time-consuming. The consequences of the limitations of NURBS are non-interpolating control points, non-local refinement, non-intuitive and challenging handling of smoothness, junctions or curvature.
Subdivision and polygons: Classical subdivision has the drawback that the control-net does not interpolate the limit surface and uv-mapping or solid modeling is more difficult than with NURBS, whereas polygon meshes do interpolate a surface but do not allow for smooth modeling.
… I got the impression that this is rather a list of arguments one may find in a Math-book describing geometry representation methods – just looking at properties of single surfaces.
This description however has quite little to do with what I would perceive limits of these paradigms in actual software implementations . Somewhat irritating to read – from people who announce nothing less but a revolution…
This statement on Nurbs…
Non-interpolating control points, non-local refinement, non-intuitive and challenging handling of smoothness, junctions or curvature.
…may apply to a single surface.
But one rarely ever works with single surfaces in Nurbs, anywhere. Whether it is Rhino, SolidWorks or Creo, an architectural model or a car rim – one always works with networks of surfaces. Of all popular geometry representation types Nurbs, without a question gives the greatest freedom for precise local refinement: Users may add or remove geometry at every spot of the model. The outcome is a polysurface.
A great actual disadvantage of Nurbs in software implementation is that one may not deform existing networks of surfaces without compromizing already established boundaries. Deforming a neutral face into a laughing face caused kinks and cracks and tons of extra work – it’s the reason why character modelers without exception moved on to Subdivision Surfaces which makes the same task a childs play. Nurbs tools in Maya or 3DSMax are still there, but they haven’t been updated since the Nineties.
A second large problem of Nurbs (not just inside Rhino) is that complex shape transitions generally need a lot of planning and knowledge about the way Surfacing works. It’s hard to try certain shapes real quick – just to get an impression. Beginners may sit for hours and likely still fail to create something like a good looking y-branch.
About Subdivision and Polygons you say…
Classical subdivision has the drawback that the control-net does not interpolate the limit surface and uv-mapping or solid modeling is more difficult than with NURBS, whereas polygon meshes do interpolate a surface but do not allow for smooth modeling.
This odd UV-statement must have been quoted and requoted a thousand times…
Yeah, there’s inherent UV’s in Nurbs while one first needs to assign UV’s to mesh geometry. Again in software implementation there is no drawback at all – quite in contrast. Nurbs automatic per surface UV’s usually have nothing at all to do with the desired material distribution, one often wants to get rid of them.
UV assigment to low poly mesh cages on the other hand is a highly elegant concept – every free SubD Modeler (> software implementation) beats the UV handling of the most expensive Nurbs programs money can buy.
The most critical disadvantage of SubD in software implementation is what you attribute to Nurbs: Local refinement, such as cutting hole while maintaining the shape or connecting separate bodies to one part may be a major issue example here. But every control cage which was built Catmull Clark approximated inside a mesh package does convert to Nurbs exactly as well or as bad as a control cage which previously got represented via limit surface approach.