psVectorSort doesn't

[Paul Price]

Calling psVectorSort with a psVector of type F32 yields a vector with the
elements in the same order as the input:

Input vector:
0.186915 0.951040 0.545436 0.074626 0.229219 0.522971 0.092768
Sorted vector:
0.186915 0.951040 0.545436 0.074626 0.229219 0.522971 0.092768

This is because the comparison function is effectively:
int psCompareF32(float a, float b) {

return a - b;

}

Note that, for the above list of numbers (the absolute difference between any of
which is always less than unity), this comparison function will always return 0
(meaning that all the values are equal). The comparison function should be
something like:

int psCompareF32(float a, float b)
{

float diff = a - b;
return (diff > 0.0) ? +1 : ((diff < 0.0) ? -1 : 0);

}

Code used for test:

#include <stdio.h>
#include "pslib.h"

#define SIZE 7

int main(int argc, char *argv[])
{

psTraceSetLevel(".", 10);

psRandom *random = psRandomAlloc(PS_RANDOM_TAUS, 1);
psVector *vector = psVectorAlloc(SIZE, PS_TYPE_F32);

printf("Input vector:\n");
for (int i = 0; i < SIZE; i++) {

vector->data.F32[i] = psRandomUniform(random);
printf("%f ", vector->data.F32[i]);

}
printf("\n");

psStats *stats = psStatsAlloc(PS_STAT_SAMPLE_MEDIAN);
(void)psVectorStats(stats, vector, NULL, NULL, 0);
printf("Median: %f\n", stats->sampleMedian);

psVector *sorted = psVectorAlloc(SIZE, PS_TYPE_F32);
sorted = psVectorSort(sorted, vector);

printf("Sorted vector:\n");
for (int i = 0; i < SIZE; i++) {

printf("%f ", sorted->data.F32[i]);

}
printf("\n");

for (int i = 0; i < SIZE - 1; i++) {

printf ("%d ", psCompareF32(&sorted->data.F32[i], &sorted->data.F32[i+1]));

}
printf("\n");

}

[robert.desonia@…]

That sounds serious; I'll need to look into what Ross did on that one.

[robert.desonia@…]

I rewrote the F32 and F64 versions of the compare functions to something along
the line of your example. I used FLT_EPSILON / DBL_EPSILON instead of 0.0,
though, as if two floats/doubles are within epsilon of each other, they are to
be considered 'equal'.

I'll update Ross's tests for regression purposes as well. The fix will be in
the next release.

[Paul Price]

Thanks for the quick fix!

[Paul Price]

Bugs have been fixed... closing.