Re: trivial or significant test failure?

[marco atzeri]

On 7/28/2012 4:19 PM, Ben Abbott wrote:
> Using atlas-3.10.0 on MacOS X, I'm seeing this failure that at first glance 
> looked trivial to me.  But after running the test several times I'm reliably 
> getting the same answer. Should I open a bug report or just increase the 
> tolerance?
>
> Ben
>
> > > > > > processing 
> > > > > > /Users/bpabbott/Development/mercurial/default/src/DLD-FUNCTIONS/eigs.cc
>    ***** testif HAVE_ARPACK
>   [v1,d1] = eigs (A, k, "si");
>   d1 = diag (d1);
>   for i=1:k
>     assert (max (abs ((A - d1(i)*eye (n))*v1(:,i))), 0, 1e-11);
>   endfor
> !!!!! test failed
> assert (max (abs ((A - d1 (i) * eye (n)) * v1 (:, i))),0,1e-11) expected
> 0
> but got
>   3.0465e-11
> maximum absolute error 3.04655e-11 exceeds tolerance 1e-11

the test pass on cygwin with lapack-3.4.1, atlas-3.9.85
and openblas-0.2.2

for all I have
octave:1> test /pub/hg/octave/src/DLD-FUNCTIONS/eigs.cc
PASSES 149 out of 149 tests

Regards
Marco