False positive on comparison of IEEE854 floats: plus and minus zero

[Grigory Rechistov]

I have code that performs flushing of subnormal 32-bit floats to plus or minus zeroes.

static inline float flush(float x) {
        if (fabs(x) < FLT_MIN) {
                return (x >= 0.0) ? 0.0 : -0.0; // reports "Same value in both branches of ternary operator"
        } else return x;
}

Cppcheck reports "Same value in both branches of ternary operator." . It considers 0.0 and -0.0 to be equal.
But they are not. E.g. the following program compiled with GCC returns with 1, meaning the comparison return non-equality.

int main()
{
     float p0 = 0.0;
     float m0 = -0.0;
     return p0 == m0;
}

I am using Cppcheck 2.7 dev built from SHA 31ea13eb0cf396c6b7ff66948808999c74f60f36 at 4th of February.

Thank you!

[Grigory Rechistov]

This is how invoke the analysis:

$ cppcheck --enable=all --platform=native -D__GNUC__ -D__linux__ -D__x86_64__ pmzero.c 
Checking pmzero.c ...
Checking pmzero.c: __GNUC__=1;__linux__=1;__x86_64__=1...
pmzero.c:6:41: style: Same value in both branches of ternary operator. [duplicateValueTernary]
                return (x >= 0.0) ? 0.0 : -0.0; // reports "Same value in both branches of ternary operator"
                                        ^

The input file:

$ cat pmzero.c 
#include <math.h>
#include <float.h>

static inline float flush(float x) {
        if (fabs(x) < FLT_MIN) {
                return (x >= 0.0) ? 0.0 : -0.0; // reports "Same value in both branches of ternary operator"
        } else return x;
}

[Grigory Rechistov]

Also reproduced if I use Cppcheck built from the yesterday's code snapshot (SHA 8672e12a7a5173c0322b2f3341786e6b370afdde)

[Jens Yllman]

Hmmm, I wonder how well cppcheck handles other special values in IEEE854.