Question

In order to prevent DivideByZeroException in C#, people often write things like

double f(double x) {
  if (x != 0.0) return 1000.0/x;
  else return 0.0;
}

Given the fact that floating-point arithmetic always has imprecisions, I wonder whether it is guaranteed that this function never throws a DivideByZeroException.

Answer 1

The documentation says:

Dividing a floating-point value by zero will result in either positive infinity, negative infinity, or Not-a-Number (NaN) according to the rules of IEEE 754 arithmetic. Floating-point operations never throw an exception. For more information, see Single and Double.

So yes, “it is guaranteed that this function never throws a DivideByZeroException.” – even without any checking, but it may return positive infinity, negative infinity, or Not-a-Number (NaN) even if you check for 0.0, for example when you divide a rather large value by a really small value so that the result exceeds the range covered by double.