If that were true, they wouldn't have defined the equality operators for floating point types. Some numbers are exactly representable and some numbers are not [1]. It's difficult, but you can in fact reason about the exact results of floating point calculations.
https://news.ycombinator.com/item?id=1847462