build from source on RHEL6
Hello SciPy people, I just spent a few days trying to figure out the build process for SciPy on our new Linux desktops. It was an arduous journey, and I though I should share: http://idolinux.blogspot.com/2011/02/atlas-numpy-scipy-build-on-rhel-6.html It seems to be working fine now. Tips and pointers are much appreciated. Cheers, -- Gavin W. Burris Senior Systems Programmer Information Security and Unix Systems School of Arts and Sciences University of Pennsylvania
On Tue, Feb 22, 2011 at 4:06 AM, Gavin W. Burris <bug@sas.upenn.edu> wrote:
Hello SciPy people,
I just spent a few days trying to figure out the build process for SciPy on our new Linux desktops. It was an arduous journey, and I though I should share: http://idolinux.blogspot.com/2011/02/atlas-numpy-scipy-build-on-rhel-6.html
It seems to be working fine now. Tips and pointers are much appreciated.
Thanks for sharing Gavin. It looks like you followed the explanation at http://scipy.org/Installing_SciPy/Linux pretty closely. Can you think of any changes to that description that would have made your life easier? Or did you look at some other doc? Cheers, Ralf
Hi Ralf, I think it would be very hard for anyone but a professional programmer to build scipy with the current directions. It was a very time consuming. Since ATLAS takes so long to compile, I could only try a new build method about once a day. The directions on scipy.org reference older versions and are confusing. They should be updated for the current versions, including prerequisites. The directions should prominently note how to configure a site.cfg file prior to building numpy, which is a prerequisite to building scipy. Without the numpy site.cfg step, I was left searching and rebuilding many, many times. I tried different options to the BLAS and LAPACK environment variables. I even tried to build reference versions of BLAS. I was desperate. The numpy build with a proper site.cfg was what finally made things work for me. Cheers. On 02/22/2011 05:22 AM, Ralf Gommers wrote:
On Tue, Feb 22, 2011 at 4:06 AM, Gavin W. Burris <bug@sas.upenn.edu> wrote:
Hello SciPy people,
I just spent a few days trying to figure out the build process for SciPy on our new Linux desktops. It was an arduous journey, and I though I should share: http://idolinux.blogspot.com/2011/02/atlas-numpy-scipy-build-on-rhel-6.html
It seems to be working fine now. Tips and pointers are much appreciated.
Thanks for sharing Gavin. It looks like you followed the explanation at http://scipy.org/Installing_SciPy/Linux pretty closely. Can you think of any changes to that description that would have made your life easier? Or did you look at some other doc?
Cheers, Ralf _______________________________________________ SciPy-User mailing list SciPy-User@scipy.org http://mail.scipy.org/mailman/listinfo/scipy-user
-- Gavin W. Burris Senior Systems Programmer Information Security and Unix Systems School of Arts and Sciences University of Pennsylvania
On 02/23/2011 07:18 AM, Gavin W. Burris wrote:
Hi Ralf,
I think it would be very hard for anyone but a professional programmer to build scipy with the current directions.
We are aware of the issue. It is unfortunately quite difficult to fix, because the issue is complex. Building numpy/scipy itself is not so hard nowadays on linux because most distributions have only one fortran compiler (gfortran), instead of g77 vs gfortran which are incompatible between each other. Blas/lapack/atlas are often badly packaged, poorly tested by distributions (to be fair, they are hard to build and test). There are tons of inconsistencies between them: the library names, locations and conventions are different between distributions and even between versions of a same distribution. That's why documenting it is so hard - it takes time, and has to be done for many combinations of softwares. On top of that, the build infrastructure around python is poorly thought out for complex packages like numpy - the pletora of options you saw are a direct consequence of those issues. cheers, David
participants (3)
-
David
-
Gavin W. Burris
-
Ralf Gommers