Re: Accepted octave-statistics 1.6.1-1 (source all amd64) into unstable
* Sébastien Villemot <sebastien@debian.org> [2024-01-20 09:51]:
Thanks for updating octave-statistics.
Note that the package failed to build on several architectures:
https://buildd.debian.org/status/package.php?p=octave-statistics
It seems that the solution to some problem is not unique, and that on
some architectures it returns a different solution.
The FTBFS problem on arm64, ppc64el, riscv64, and r390x comes from this
failing BIST, in inst/pdist2.m, involving the Mahalanobis distance:
yy = [1 2 3;5 6 7;9 5 1];
d = [0, 8.4853, 18.0416; 2.4495, 10.0995, 19.4808; ...
2.4495, 10.6771, 19.7104; 2.4495, 10.6771, 20.4573];
i = [2, 2, 2; 1, 4, 4; 4, 1, 1; 3, 3, 3];
[D, I] = pdist2 (y, yy, "mahalanobis", "smallest", 4);
assert ({D, I}, {d, i}, 1e-4);
I have already patched inst/pdist2.m by transforming the two BISTs
previous to the above one into xtest. The BIST above succeeds on amd64
and did not catch my attention.
I know, transforming BISTs into xtest is an ugly solution, but it would
be a pity to not have version 1.6.1 because of these failing BISTs. At
any rate, I have already reported the issue upstream [*].
A simple solution to the problem would be to transform this third BIST
using the Mahalanobis distance into xtest, as the two other ones.
Best,
Rafael
[*] https://savannah.gnu.org/bugs/?65178
Reply to: