[Numpy-discussion] *ALPHA* release of packaged numpy and scipy (RPM) for openSUSE, Fedora Core 5 and 6

David Cournapeau david@ar.media.kyoto-u.ac...
Mon Apr 9 09:34:38 CDT 2007


   This is an annoucement for packaged binaries of numpy and scipy (eg 
rpm and deb, for easy installation).


   I have an initial version of numpy and scipy RPM for openSUSE and 
Fedora core ready for testing:
       * both numpy and scipy are available, WITH blas and lapack 
support dependencies (99% of the work was making shared libraries for 
blas/lapack work).
       * supported distributions: openSUSE 10.2, Fedora Core 5 and 6 are 
available (x86 only).

   ****** BIG FAT WARNING *******

   I am new to RPM packaging (and do not use any rpm based 
distribution), and a mistake may have the potential to break the system; 
also, because the packages do not use (yet) ATLAS, they are not really 
efficient. Finally, I did not attempt to follow the packaging policy of 
the various distributions (AFAIK, rpm does not make that easy). I can 
only say that the packages were installed on test machines (fedora core 
6 and open suse 10.2 on x86) successfully, and ran all scipy tests 
successfully. I *strongly* recommend that for now, only experienced 
users try those packages, and for testing purpose only.

   * Where ?

   You can find the repositories for the supported distributions here 
(please do not ask me how to add the repository to your system; if you 
do not know how to do it, you should not install those packages anyway):


   Packages' name are python-numpy and python-scipy.

   * Technical details:

   Those packages are done using the open suse build system: 

   "The openSUSE Build Service is open and complete distribution 
development platform that provides infrastructure for a development of 
the future openSUSE based distributions. It provides software developers 
with a tool to compile, release and publish their software for the broad 
user audience, including creation of their own Linux distribution based 
on openSUSE for various hardware architectures."

   The big point of this service compared to doing the packaging by hand 
is that everytime you modify one of the dependency, all packages are 
rebuilt for all distribution automatically. The system also provides a 
build farm, so that I don't have to build the rpm on each supported 
distribution. Note that the code to do the packaging is 100 % compatible 
with normal rpm tools, so that the building itself can be done the usual 
way (eg if for some reason the opensuse build system is not available 
anymore, the work is not lost).

   * TODO:

   I did those packages more for a proof of concept than for final 
releases. If other people find this solution useful, I hope that people 
more knowledgeable than me for each supported distribution will help 
polishing the packages:
       * For now, only minimal dependencies were added (netlib BLAS and 
LAPACK). Ideally, we would put fftw, umfpack and atlas.
       * Make the packages follow each distribution policy
       * Make x86_64 available (I know nothing about 64 bits convention, 
where to put libraries, and I do not have access to any 64 bits machines).
       * add Debian/Ubuntu packages (they will be much easier to 
package, because debian actually packages the BLAS and LAPACK correctly, 
contrary to fedora and suse which are kind of broken in that respect), 
and other useful packages such as matplotlib, ipython and pytables.



More information about the Numpy-discussion mailing list