CWE Rule 681

This issue occurs when converting a floating point number to a smaller floating point data type. If the variable does not have enough memory to represent the original number, the conversion overflows.

The exact storage allocation for different floating point types depends on your processor. See Target processor type (-target).

Overflows can result in unpredictable values from computations. The result can be infinity or the maximum finite value depending on the rounding mode used in the implementation. If you use the result of an overflowing conversion in subsequent computations and do not account for the overflow, you can see unexpected results.

The fix depends on the root cause of the defect. Often the result details show a sequence of events that led to the defect. Use this event list to determine how the variable being converted acquires its current value You can implement the fix on any event in the sequence. If the result details do not show the event history, you can trace back using right-click options in the source code and see previous related events. See also Interpret Bug Finder Results in Polyspace Desktop User Interface or Interpret Bug Finder Results in Polyspace Access Web Interface (Polyspace Access).

You can fix the defect by:

In general, avoid conversions to smaller floating point types.

See examples of fixes below.

If you do not want to fix the issue, add comments to your result or code to avoid another review. See:

By default, a Bug Finder analysis does not recognize infinities and NaNs. Operations that results in infinities and NaNs might be flagged as defects. To handle infinities and NaN values in your code, use the option Consider non finite floats (-allow-non-finite-floats).

A default Bug Finder analysis might not raise this defect when the input values are unknown and only a subset of inputs cause an issue. To check for defects caused by specific system input values, run a stricter Bug Finder analysis. See Extend Bug Finder Checkers to Find Defects from Specific System Input Values.

float convert(void) {

    double diam = 1e100;
    return (float)diam; //Noncompliant
}

In the return statement, the variable diam of type double (64 bits) is converted to a variable of type float (32 bits). However, the value 1^100 requires more than 32 bits to be precisely represented.

This issue occurs when you cast an integer value to a floating-point type that cannot represent the original integer value.

For instance, the long int value 1234567890L is too large for a variable of type float.

If the floating-point type cannot represent the integer value, the behavior is undefined (see C11 standard, 6.3.1.4, paragraph 2). For instance, least significant bits of the variable value can be dropped leading to unexpected results.

Convert to a floating-point type that can represent the integer value.

For instance, if the float data type cannot represent the integer value, use the double data type instead.

When writing a function that converts an integer to floating point type, before the conversion, check if the integer value can be represented in the floating-point type. For instance, DBL_MANT_DIG * log2(FLT_RADIX) represents the number of base-2 digits in the type double. Before conversion to the type double, check if this number is greater than or equal to the precision of the integer that you are converting. To determine the precision of an integer num, use this code:

 size_t precision = 0;
 while (num != 0) {
    if (num % 2 == 1) {
      precision++;
    }
    num >>= 1;
 }

Some implementations provide a builtin function to determine the precision of an integer. For instance, GCC provides the function __builtin_popcount.

#include <stdio.h>

int main(void) {
  long int big = 1234567890L;
  float approx = big; //Noncompliant
  printf("%ld\n", (big - (long int)approx));
  return 0;
}

In this C code, the long int variable big is converted to float.

One possible correction is to convert to the double data type instead of float.

#include <stdio.h>

int main(void) {
  long int big = 1234567890L;
  double approx = big;
  printf("%ld\n", (big - (long int)approx));
  return 0;
}