Octave 3.2.0 test failures under Mac OS X

[Marius Schamschula]

Thomas,

ARPACK seems to require some voodoo.

I had built ARPACK using the MacPorts patches, but w/o the parpack  
extensions (I didn't want to add MPI to the mix) using g95 (my  
default fortran compiler). I just revisited the MacPorts trak and  
noticed that the ARPACK Portfile was last modified 12 months ago.

I  rebuilt with your arguments for LAPACKLIB and BLASLIB, i.e. vecLib  
vs. Accelerate using the base package. Same set of errors.

Then I tried the Debian package that you mentioned. I didn't get a  
library built at all until I tweaked the Makefiles in SRC and UTIL  
(needed to remove a few # marks).

After rebuilding octave, eigs.cc passed all tests.

Thanks!

On Jun 8, 2009, at 1:47 PM, Thomas Treichl wrote:

> Marius Schamschula schrieb:
>> Hi all,
>> I built octave 3.2.0 with the new dependencies i get:
>> ...
>>   src/DLD-FUNCTIONS/eigs.cc .............................. PASS    
>> 12/148  FAIL 136
>>   src/DLD-FUNCTIONS/fft.cc ............................... PASS    
>> 18/19   FAIL 1
>> ...   src/data.cc ............................................  
>> PASS  506/509  FAIL 3
>> ...
>>   scripts/sparse/svds.m .................................. PASS     
>> 0/3    FAIL 3
>> ...
>> Summary:
>>   PASS   5586
>>   FAIL    143
>> I can send fntests.log if needed.
>> Marius
>
> Hi Marius,
>
> with the introduction of the Arpack dependency for Octave 3.1.5x  
> and following I had the same problems as you. Here is a little  
> description about what my problems have been on OSX in detail and  
> maybe this also then helps for you. Definitely I can say that  
> Arpack is a beast, but once compiled correctly it works perfectly ;-)
>
> I used fort77/f2c before and I downloaded the original Arpack  
> *.tar.gz from their website. This combination does not work for me  
> at all until today. Next I have chosen to switch to g95 instead of  
> fort77/f2c and things became much better but still not absolutely  
> good (some of the eigs.cc errors disappeared but some still where  
> present and the svds.m error, too). Finally I switched to Debian's  
> Arpack source package found here
>
> http://ftp.de.debian.org/debian/pool/main/a/arpack/arpack_2.1 
> +parpack96.dfsg.orig.tar.gz
>
> and voila, everything worked perfectly (together with g95). I also  
> have to say that I compile Arpack against vecLib's Lapack, like this
>
>    ARPACKLIB  = $(home)/libarpack.a
>   -LAPACKLIB =
>   -BLASLIB =
>   +LAPACKLIB = -Wl,-framework -Wl,vecLib
>   +BLASLIB = -Wl,-framework -Wl,vecLib
>
> So what I want to say in short ;-) try to choose Debian's Arpack  
> source package and make sure once again that your Fortran compiler  
> is well suited to compile Arpack correctly. Finding out this took  
> me, I remember, 5..6 evenings or even more, so I really hope you  
> can save some time ;o)
>
> I see the three src/data.cc errors, too. This is a problem of our g+ 
> + compiler that produces another result than on Linux and other  
> systems, no idea how to fix this, see
>
> http://www-old.cae.wisc.edu/pipermail/bug-octave/2008-August/ 
> 006584.html
>
> I cannot see your last error in fft.cc. But I've got another one in  
> test_diag_perm.m.
>
> Good luck,
>
>   Thomas
>
> --
> >>>>> processing test_diag_perm
>   ***** test
>  n = 7;
>  A = sprand (n, n, .5);
>  scalefact = rand (1, n-2) + I () * rand(1, n-2);
>  Dc = diag (scalefact, n-2, n);
>  assert (full (A / Dc), full(A) / Dc)
> !!!!! test failed
> assert (full (A / Dc),full (A) / Dc) expected
>  Columns 1 through 3:
>
>    0.01185 - 0.06596i   0.00000 + 0.00000i   0.24809 - 0.36095i
>    0.00000 + 0.00000i   0.00000 + 0.00000i   0.00000 + 0.00000i
>    0.00000 + 0.00000i   0.00000 + 0.00000i   0.62913 - 0.91533i
>    0.12441 - 0.69223i   0.00000 + 0.00000i   0.00000 + 0.00000i
>    0.15257 - 0.84895i   0.43735 - 0.09750i   0.12501 - 0.18188i
>    0.00000 + 0.00000i   0.00000 + 0.00000i   0.00000 + 0.00000i
>    0.08113 - 0.45146i   0.00000 + 0.00000i   0.00000 + 0.00000i
>
>  Columns 4 and 5:
>
>    0.18947 - 0.31809i   0.42914 - 0.62782i
>    0.00000 + 0.00000i   0.00000 + 0.00000i
>    0.00000 + 0.00000i   0.00000 + 0.00000i
>    0.10383 - 0.17431i   0.00000 + 0.00000i
>    0.12150 - 0.20397i   0.63806 - 0.93346i
>    0.00000 + 0.00000i   0.00000 + 0.00000i
>    0.00000 + 0.00000i   0.12584 - 0.18411i
> but got
>  Columns 1 through 3:
>
>    0.01185 - 0.06596i   0.00000 + 0.00000i   0.24809 - 0.36095i
>    0.00000 + 0.00000i   0.00000 + 0.00000i   0.00000 + 0.00000i
>    0.00000 + 0.00000i   0.00000 + 0.00000i   0.62913 - 0.91533i
>    0.12441 - 0.69223i   0.00000 + 0.00000i   0.00000 + 0.00000i
>    0.15257 - 0.84895i   0.43735 - 0.09750i   0.12501 - 0.18188i
>    0.00000 + 0.00000i   0.00000 + 0.00000i   0.00000 + 0.00000i
>    0.08113 - 0.45146i   0.00000 + 0.00000i   0.00000 + 0.00000i
>
>  Columns 4 and 5:
>
>    0.18947 - 0.31809i   0.42914 - 0.62782i
>    0.00000 + 0.00000i   0.00000 + 0.00000i
>    0.00000 + 0.00000i   0.00000 + 0.00000i
>    0.10383 - 0.17431i   0.00000 + 0.00000i
>    0.12150 - 0.20397i   0.63806 - 0.93346i
>    0.00000 + 0.00000i   0.00000 + 0.00000i
>    0.00000 + 0.00000i   0.12584 - 0.18411i
> values do not match

Marius
--
Marius Schamschula
Webmaster

The Huntsville Macintosh Users Group
www.hmug.org

webmaster at hmug dot org
marius at schamschula dot com



-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://www-old.cae.wisc.edu/pipermail/bug-octave/attachments/20090608/b612e822/attachment.html