Question

Does anybody can explain to me why the execution of this program enters the infeasible if?
I know the existence of precisions problems with double and float, but I can’t figure out the problem in this example..

Thank you in advance!

NB: I compiled it under Ubuntu 10.10 with GCC 4.4.5

#include <stdio.h>

#define NUM_JOINTS 1

typedef struct {
    double t1;
    double t2;
    double t3;
} jointProfile;

jointProfile jp[NUM_JOINTS];

int main() {
    jp[0].t1 = 0.51639777949432230652604403076111339032649993896484375;
    jp[0].t2 = 0.00000000000000000000000000000000000000000000000000000;
    jp[0].t3 = 0.77459666924148340427791481488384306430816650390625000;

    double maxTime = (jp[0].t1 + jp[0].t2 + jp[0].t3);

    if ((jp[0].t1 + jp[0].t2 + jp[0].t3) < maxTime) {
        printf("Infeasible\n");
    }

    return 0;
}

Answer 1

Well, apparently you don’t know enough about the subtleties of floating point computations.

See e.g. http://download.oracle.com/docs/cd/E19957-01/806-3568/ncg_goldberg.html

If you’re doing this on a 32-bit Ubuntu, the problem is likely that maxtime is spilled to memory and thus loses bits.

One fix is to change the conditional to

if (fabs(jp[0].t1 + jp[0].t2 + jp[0].t3 - maxtime) &lt DBL_EPSILON) {
 ...
}