On Tue, Nov 24, 2020 at 09:01:27AM +1100, Steven D'Aprano wrote:
On Mon, Nov 23, 2020 at 04:26:20PM -0500, Wes Turner wrote:
Is there a different IEEE spec or CAS that distinguishes between 1/x and 2/x where x=0?
No. x/0.0 would either signal an error (in Python terms: raise an exception) or return a NAN. The specific NAN it might return is not set by the IEEE-754 standard.
Correction (as noted in my next post), it actually returns a signed infinity. It *ought to* return a NAN, unless we interpret the denominator as an underflowed non-zero quantity rather than precisely and exactly zero, which seems to be what the standard does. In any case, there's only a single pair of infinities, so you can't distinguish between two expressions which return infinities with the same sign. You can't distinguish between cases that "overflow to infinity" from "actual mathematical infinity" (if such a thing existed in the real numbers) either. So IEEE-754 should be considered as trying to make practical, pragmatic decisions rather than aiming for mathematical purity. -- Steve