The point isn't just that you can't tell exactly when something is zero, although indeed you often can't. It's that if you have some calculation that needs to

The point isn't just that you can't tell exactly when something is zero, although indeed you often can't. It's that if you have some calculation that needs to be special-cased at zero (e.g., because you're dividing by something and it might explode) then it probably actually needs to be special-cased near zero (because near zero it might overflow or run into other numerical problems), and it's probably worth seeing if there's another way of organizing the calculation that doesn't misbehave for small values in the first place.