Still on a phone - otherwise I would test it myself.

So which is faster? Putting a critical section around this or making an extra on the stack copy of each surface multiple times per boolean operation?

Obsuously this may depend on many things, CPU, compiler, RAM but if the difference is drastic on some "typycal" system it may help to choose.

A different question is which is cleaner...

@ruevs You had some model that showed significant performance change with OpenMP. You might try both methods with that. Just put a "omp crtical" with { } around most of this function from before WeightControlPoints() to after the last UnWeightControlPoints(). compare that to this PR. I'm in favor of this approach because it fixes the underlying problem, and also modifying a surface and putting it back just seems like a bad idea, and this doesn't pollute the code with an extra #pragma

If you want to do some tests you might also try using OMP in SShell::CopyCurvesSplitAgainst() I think the last 2 lines need to be inside a critical (or not the last one). I tried this once before and the problem with surfaces distorting was even worse but that will be fixed with this PR.

In the mean time I'm probably going to merge this one.

Yes, that model is attached here #613
With very small chord tolerance and large number of segments the difference was very noticeable. I will experiment when I get some time. Also have not forgotten about your triangulation changes.