[SciPy-user] installation problems

[Robert Kern]

manouchk wrote:
> Le Mardi 14 Mars 2006 14:12, Hanno Klemm a écrit :
> 
>>Philippe,
>>
>>I had a similar problem. When you are installing on a Red Hat
>>distribution, it is highly likely that your BLAS or another numerical
>>library is incomplete. That seemed to be the problem on the Red Hat
>>distribution I have been using.
>>
>>Probably you then have to compile the numerical libraries yourself
>>(that's at least what I did).
> 
> On mandriva 2005LE there is a similar problem when I run scipy.test(level=1)
> 
> it gives a lo of line with the problem "undefined symbol: srotmg_" like this 
> one :
> 
> import signal -> 
> failed: /usr/lib/python2.4/site-packages/scipy/linalg/fblas.so: undefined 
> symbol: srotmg_
> 
> and then there is a warning :
> 
> WARNING: clapack module is empty
> 
> If I refer to old mail of scipy-user mailing list :
> http://www.scipy.net/pipermail/scipy-user/2002-June/000447.html
> 
> the solution was already given there :
> 
> "Yes, srotmg_ is missing in the BLAS libraries that are included in LAPACK.
> You have to download http://netlib2.cs.utk.edu/blas/blas.tgz and rebuild
> BLAS to fix it."
> 
> Question 1:
> only blas.tgz is needed?

Yes.

> Question 2:
> is it better to use the cblas.tgz in order to have coherent *.c and *.f files?

I'm not sure what this means. I would imagine that it's only worth using CBLAS
if it's optimized like ATLAS. Don't bother.

> Question 3:
> the solution is to replace all *.f and .c files and then build?

You can actually just unpack the blas.tgz somewhere and set the environment
variable BLAS_SRC to that directory. The "Linear Algebra libraries" section of
http://old.scipy.org/documentation/buildscipy.txt tells you how (it still
applies, although it is an old document).

> Question 4:
> what is the license of blas.tgz and cblas.tgz?

According to Debian, which is quite picky about such matters, blas.tgz at least
is in the public domain.

> The problem is still there in mandriva 2005! It is a pity that the problem was 
> not resolved so that the distribution chip the complete version. 
> 
> I hope it is not to much out of the subject but I'd like to know if mandriva 
> is now a exception and other distribution ship the "complete" version?

Redhat seems to have that problem, too. Debian and its progeny do not. Go Ubuntu!

-- 
Robert Kern
robert.kern at gmail.com

"I have come to believe that the whole world is an enigma, a harmless enigma
 that is made terrible by our own mad attempt to interpret it as though it had
 an underlying truth."
  -- Umberto Eco