Re: One failing test with svds with developer sources

[Martin Helm]

Am Sonntag, 14. November 2010, 20:22:08 schrieb Kai Habel:
>   On 14.11.2010 19:39, Martin Helm wrote:
> > Hi all,
> > 
> > maybe someone has seen this also or can shed light on it. One of the
> > tests always fails for me when performing make check after building the
> > developer sources (it is the only test which fails at all), since it is
> > somehow arpack related I am not sure if it is a problem with my
> > configuration and would be interested to hear if I am the only one who
> > has that.
> > (linux, openSUSE 11.3 64bit)
> > libarpack2-2.1-10.1.x86_64
> > 
> >>>>>> processing
> >>>>>> /home/martinh/Projekte/octave/hg/octave/scripts/sparse/svds.m
> >    
> >    ***** testif HAVE_ARPACK
> >   
> >   [u2,s2,v2,flag] = svds(a,k,0,opts);
> >   s2 = diag(s2);
> >   assert(flag,!1);
> >   assert(s2, s(k:-1:1), 1e-10);
> > 
> > !!!!! test failed
> > assert (flag,!1) expected
> > 0
> > but got
> > 
> >   1
> > 
> > values do not match
> > 
> > Please tell me if you need more detailed info.
> > _______________________________________________
> > Help-octave mailing list
> > address@hidden
> > https://www-old.cae.wisc.edu/mailman/listinfo/help-octave
> 
> Hello Martin,
> 
> I remember having problems with arpack on OpenSuse myself. Since 11.3 I
> use arpack from
> 
> URL:
> http://download.opensuse.org/repositories/home%3a/ocefpaf/openSUSE_11.3/
> 
> and I don't have problems anymore.
> 
> Hope this helps,
> 
> Kai

That is where my arpack library comes from (ocepaf). In the past i compiled 
arpack myself. The interesting thing is that the numeric result from this test 
is ok. The difference between the calculated and the expected value is about 
1e-13 (the test itself asks only for < 1e-10). It just irritates me that the 
flag variable indicates that  convergence failed (which seems not to be the 
case really).
I think I will ignore that for the moment and debug through the svds.m during 
the next week to see where it comes from.
Thanks for the feedback.