V674. Expression contains a suspicious mix of integer and real types.

The analyzer has detected a potential error in an expression where integer and real data types are used together. Real types are data types such as float/double/long double.

Let's start with a simple case. A literal of the 'double' type is implicitly cast to an integer, which may indicate a software bug in the code.

int a = 1.1;

This fragment is meaningless. The variable should be most likely initialized with some other value.

The example shown above is an artificial one and therefore of no interest to us. Let's examine some real-life cases.

Example 1.

int16u object_layer_width;
int16u object_layer_height;
if (object_layer_width == 0 ||
    object_layer_height == 0 ||
    object_layer_width/object_layer_height < 0.1 ||
    object_layer_width/object_layer_height > 10)

An integer value is compared to the constant '0.1', and that's very strange. Assume the variables have the following values:

• object_layer_width = 20;
• object_layer_height = 100;

The programmer expects that division of these numbers will give '0.2'; it fits into the range [0.1..10].

But in fact the division result will be 0. Division is performed over integer data types, and though the result is extended to the type 'double' when compared to '0.1' a bit later, it is too late. To fix the code we need to perform an explicit type conversion beforehand:

if (object_layer_width == 0 ||
    object_layer_height == 0 ||
    (double)object_layer_width/object_layer_height < 0.1 ||
    (double)object_layer_width/object_layer_height > 10.0)

Example 2.

// be_aas_reach.c
ladderface1vertical =
  abs( DotProduct( plane1->normal, up ) ) < 0.1;

The argument of the abs() function is of the 'double' type. The code seems to execute correctly at first sight, and one may think it was just "silly" of the analyzer to attack this good code.

But let's examine the issue closer. Look how the function abs() is declared in header files.

int __cdecl abs( int _X);
#ifdef __cplusplus
extern "C++" {
  inline long __CRTDECL abs(__in long _X) { .... }
  inline double __CRTDECL abs(__in double _X) { .... }
  inline float __CRTDECL abs(__in float _X) { .... }
}
#endif

Yes, abs() functions are overloaded for different types in C++. But we are dealing with a C code (see the file: be_aas_reach.c).

It means that a 'float'-type expression will be implicitly cast to the 'int' type. The abs() function will also return a value of the 'int' type. Comparing a value of the 'int' type to '0.1' is meaningless. And this is what analyzer warns you about.

In C applications, you need another function to calculate the absolute value correctly:

double  __cdecl fabs(__in double _X);

The fixed code:

ladderface1vertical =
  fabs( DotProduct( plane1->normal, up ) ) < 0.1;

This diagnostic is classified as: