V550. An odd precise comparison. It's probably better to use a comparison with defined precision: fabs(A - B) < Epsilon or fabs(A - B) > Epsilon

26.02.2012

The analyzer detected a potential error: the == or != operator is used to compare floating point numbers. Precise comparison might often cause an error.

Consider this sample:

double a = 0.5;
if (a == 0.5) //OK
  x++;

double b = sin(M_PI / 6.0);
if (b == 0.5) //ERROR
  x++;

The first comparison 'a == 0.5' is true. The second comparison 'b == 0.5' may be both true and false. The result of the 'b == 0.5' expression depends upon the processor, compiler's version and settings being used. For instance, the 'b' variable's value was 0.49999999999999994 when we used the Visual C++ 2010 compiler. A more correct version of this code looks this way:

double b = sin(M_PI / 6.0);
if (fabs(b - 0.5) < DBL_EPSILON)
  x++;

In this case, the comparison with error presented by DBL_EPSILON is true because the result of the sin() function lies within the range [-1, 1]. But if we handle values larger than several units, errors like FLT_EPSILON and DBL_EPSILON will be too small. And vice versa, if we handle values like 0.00001, these errors will be too big. Each time you must choose errors adequate to the range of possible values.

Question: how do I compare two double-variables then?

double a = ...;
double b = ...;
if (a == b) // how?
{
}

There is no single right answer. In most cases, you may compare two variables of the double type by writing the following code:

if (fabs(a-b) <= DBL_EPSILON * fmax(fabs(a), fabs(b)))
{
}

But be careful with this formula - it works only for numbers with the same sign. Besides, if you have a row with many calculations, there is an error constantly accumulating, which might cause the DBL_EPSILON constant to appear a too small value.

Well, can I perform precise comparison of floating point values?

Sometimes, yes. But rather rarely. You may perform such comparison if the values you are comparing are one and the same value in its sense.

Here is a sample where you may use precise comparison:

// -1 - value is not initialized.
float val = -1.0f;
if (Foo1())
  val = 123.0f;
if (val == -1.0f) //OK
{
}

In this case, the comparison with value "-1" is permissible because it is this very value which we used to initialize the variable before.

We cannot cover the topic of comparing float/double types within the scope of documentation, so please refer to additional sources given at the end of this article.

The analyzer can only point to potentially dangerous code fragments where comparison may result unexpectedly. But it is only the programmer who may understand whether these code fragments really contain errors. We cannot also give precise recommendations in the documentation since tasks where floating point types are used are too diverse.

The diagnostic message isn't generated if two identical expressions of 'float' or 'double' types are being compared. Such a comparison allows to identify the value as NaN. The example of code implementing the verification of this kind:

bool isnan(double X) { return X != X; }

References:

You can look at examples of errors from real projects which were detected by this diagnostic message.