The Delaunay2 code was failing on a platform using a virtual machine and Microsoft Visual Studio 2015. The problem is that the compiler is optimizing away some floating-point expressions that should not be due to rounding mode. I mentioned this problem in Section 3.4 of https://www.geometrictools.com/Books/RobustAndErrorFreeGeometricComputing/RAEFGC_BookCorrections.pdf. I removed the final uses of std::fesetround and std::fegetround, replacing them with SWInterval interval arithmetic support that works with the default FPU rounding mode (round to nearest ties to even). These modifications affect only the new ConstrainedDelaunay2 and Delaunay2 implementations. The older implementations do not use <cfenv> for floating-point environment changes, and they are currently tagged as deprecated but will exist until the full GTL distribution is available.