Fix for issue 773

[mborland]

No description provided.

[mborland]

@NAThompson do you concur with this handling of NaNs? I thought this would be much lighter weigh than throwing an exception.

[NAThompson]

@mborland : In computer graphics, nans always render black. So this is the correct behavior.

Why doesn't the std::isnan from std work?

[mborland]

Why doesn't the std::isnan from std work?

Isn't the behavior of boost::math::isnan more stable across compilers/optimizations? Passing a NaN to std::clamp should be UB since clamp requires the less than comparable concept. I was trying to avoid this with configurations that replace std::isnan with false.

[NAThompson]

I was trying to avoid this with configurations that replace std::isnan with false.

Oh yeah that's just awful. . .