The example is wrong, because they used 1.0.

But in general x-x does not have to equal 0, that is true. I'm pretty sure Nan and infinity would yield not 0.0, but Nan instead.

And if you reach x with two different calculations, e.g. x1 = a - b - c and x2 = a - c - b it is certainly not guaranteed that x1 - x2 == 0.0