HP C A.06.05 Reference Manual

Expressions and Operators

Relational Operators (>, >=, <, ==, !=)

Chapter 5 129

if (func())

proceed

;

else

error handler

;

Dangers of Comparing Floating-Point Values

You may get unexpected results if you compare ﬂoating-point values for equality because

ﬂoating-point representations are inexact for some numbers. For example, the following

expression, though algebraically true, will evaluate to false on many computers:

(1.0/3.0 + 1.0/3.0 + 1.0/3.0) == 1.0

This evaluates to 0 (false) because the fraction 1.0/3.0 contains an inﬁnite number of decimal

places (3.33333…). The computer is only capable of holding a limited number of decimal

places, so it rounds each occurrence of 1/3. As a result, the left side of the expression does not

equal exactly 1.0.

This problem can occur in even more subtle ways. Consider the following code:

double divide(double num, double denom)

{

return num/denom;

}

int main(void)

{

double c, a = 1.0, b = 3.0;

c = a/b;

if (c != divide(a, b))

printf("Fuzzy doubles\n");

}

Surprisingly, the value stored in c may not equal the value returned by divide(). This

anomaly occurs due to the fact that some computers can represent more decimal places for

values stored in registers than for values stored in memory. Because the value returned by

divide() is never stored in memory, it may not be equal to the value c, which has been rounded

for memory storage.

NOTE To avoid bugs caused by inexact ﬂoating-point representations, you should

refrain from using strict equality comparisons with ﬂoating-point types.