Is it possible to get 0 by subtracting two unequal floating point numbers?

Question

Is it possible to get division by 0 (or infinity) in the following example?

public double calculation(double a, double          


        

Answer 1

In a floating-point implementation that conforms to IEEE-754, each floating-point type can hold numbers in two formats. One ("normalized") is used for most floating-point values, but the second-smallest number it can represent is only a tiny bit larger than the smallest, and so the difference between them is not representable in that same format. The other ("denormalized") format is used only for very small numbers that are not representable in the first format.

Circuitry to handle the denormalized floating-point format efficiently is expensive, and not all processors include it. Some processors offer a choice between either having operations on really small numbers be much slower than operations on other values, or having the processor simply regard numbers which are too small for normalized format as zero.

The Java specifications imply that implementations should support denormalized format, even on machines where doing so would make code run more slowly. On the other hand, it's possible that some implementations might offer options to allow code to run faster in exchange for slightly sloppy handling of values which would for most purposes be way too small to matter (in cases where values are too small to matter, it can be annoying having calculations with them take ten times as long as calculations that do matter, so in many practical situations flush-to-zero is more useful than slow-but-accurate arithmetic).