There's a whole set of different floating-point values which are NaN, and this canonicalizes them into one particular NaN, as I understand it. I'm not even sur

There's a whole set of different floating-point values which are NaN, and this canonicalizes them into one particular NaN, as I understand it. I'm not even sure that covering type conversions was intended here.