The last release of one of my modules works fine except in 5.6.2:
http://cpantesters.perl.org/show/Algorithm-
Combinatorics.html#Algorithm-Combinatorics-0.22
A tester reported the failure and I can reproduce it in my Mac with
5.6.2.
The test suite runs OK but for the last two tests of a single
subroutine (partitions_of_size_p), where Linux gives a segfault, and
in my Mac I get a bus error. This only happens in 5.6.2, it does not
happen in 5.8.x where the test suite passes.
I think it is unlikely that the function itself is buggy because the
tests verify result sets of several thousands of elements in 5.8.x,
so I wonder whether there's some XS gotcha behind this I am not aware
of. The code is pretty simple, I attach it below not to check what it
does, just to let you see whether you identify something that was
different in 5.6.2 and could be the culprit.
-- fxn
#define UPDATE(av, i, n) (SvIVX(AvARRAY(av)[i]) = n)
#define INCREMENT(av, i) (++SvIVX(AvARRAY(av)[i]))
#define GETIV(av, i) (SvIVX(AvARRAY(av)[i]))
#define GETAV(avptr) ((AV*) SvRV(avptr))
int __next_partition_of_size_p(SV* k_avptr, SV* M_avptr, int p)
{
AV* k = GETAV(k_avptr); /* follows notation in [3] */
AV* M = GETAV(M_avptr); /* follows notation in [3] */
int i, j;
IV mi, x;
I32 len_k, n_minus_p;
len_k = av_len(k);
for (i = len_k; i > 0; --i) {
if (GETIV(k, i) < p-1 && GETIV(k, i) <= GETIV(M, i-1)) {
INCREMENT(k, i);
if (GETIV(k, i) > GETIV(M, i))
UPDATE(M, i, GETIV(k, i));
n_minus_p = len_k + 1 - p;
mi = GETIV(M, i);
x = n_minus_p + mi;
for (j = i+1; j <= x; ++j) {
UPDATE(k, j, 0);
UPDATE(M, j, mi);
}
for (j = x+1; j <= len_k; ++j) {
UPDATE(k, j, j - n_minus_p);
UPDATE(M, j, j - n_minus_p);
}
return i;
}
}
return -1;
}
