I forgot to add that 0.0 is such a common floating point value, its erroneous use cannot be reliably distinguished from valid uses. Just finding a 0 is not good enough, but finding a NaN is.