thank for this interesting link. This is a great concept and I also think it will improve Rhino if implemented.
However we shouldn’t mix up two different things. Controllability is important, but not only in sense of a local refinement. Hiding would be great to dismiss unimportant cps, but the problem here is indeed that its local. That is what I actually like to prevent, by trying to reapproximate a complex shape into a single-span or lowcount-span shape.
One of the big advantages in staying single span is the fact that you get a global modification, which allows you to fully control the overall curvature. Its nearly impossible to get nice curvature if you only change parts of your shape. So if you choose such approach, you still need something to smooth out . My assumption is that most designers prefer shape quality instead of less deviation. Engineers might judge different. I see approximation as a compromise, with the positive aspect of data reduction. Sure global modification is limited and not suited for every task, sometime there even is no solution. However:
I would always prefer to create shapes as simple as possible, but as precise as needed.
Rebuilding does exactly match the first part of this statement, but ineffectively fulfil the second statement. So people always manually relocate cps after rebuild, which could be better assisted by a “better guess”. This is what I’m aiming for with approximation, and as proven, other platforms already can do, so its nothing impossible.