Serious bug in atlas, yet no action

Cyrus Hall hallc at
Tue Oct 6 15:41:11 BST 2009

Ciao all-

I've recently run into what appears to be a known and reported problem
in Atlas 3.2.1: the SSE2 optimized version is broken.  This version is
currently shipped with Jaunty, and is reported to still be broken in the
Karmic (see bug reports below).  There are numerous unassigned bug
reports in launchpad which appear related to the problem, some of which
are six months old:

I understand that it is unusual to e-mail a dev list to call for action
on a particular bug, but the failure mode here seems particularly
serious.  Programs that depend on the broken library compile perfectly
fine, seem to return valid numerical results, and show no signs of being
broken unless one compares against correct results.  I fear that many of
my colleagues who use Octave and other numerical tools that depend on
Atlas are less in careful in such ways, and it concerns me that bogus
results are probably, at this moment, being accepted as valid in
research labs around the world.

The way I came across the bug was by compiling the new release of
Octave, 3.2.3.  Both the eig and eigs functions (eigenvalue
decompositions, for full and sparse matrices respectively) returned
unexpected results in make check.  For the set of messages related to
the problem, see the following threads on the Octave list:

A fix for this bug seems critical to get into Karmic, and possibly into
older Ubuntu versions still supported.

Cyrus Hall

More information about the ubuntu-motu-science mailing list