[pymvpa] Install problem on Scientific Linux

[Loris Bennett]

"Loris Bennett" <loris.bennett at fu-berlin.de>
writes:

> Hi,
>
> I am have tried versions 0.4.6, 0.4.7 and the git snapshot and getting
> the following error when trying to install on Scientific Linux 6.1 with
> some package from SL 6.2:
>
> ,------------------------------------------------------
> | # ./setup.py build
> | Traceback (most recent call last):
> |   File "./setup.py", line 12, in <module>
> |     from numpy.distutils.core import setup, Extension
> | ImportError: No module named distutils.core
> `------------------------------------------------------
>
> I can fix this with the following patch:
>
> ,----------------------------------------------------
> | 12c12,13
> | < from numpy.distutils.core import setup, Extension
> | ---
> | > import numpy
> | > from distutils.core import setup, Extension
> `----------------------------------------------------
>
> I then get the error:
>
> ,----------------------------------------------------------------------------------------------
> | mvpa2/clfs/libsvmc/svmc_wrap.c:2705:31: error: numpy/arrayobject.h: No such file or directory
> `----------------------------------------------------------------------------------------------
>
> which I can fix with this patch:
>
> ,---------------------------------------------------
> | 21c22
> | < libsvmc_include_dirs = []
> | ---
> | > libsvmc_include_dirs = ['.',numpy.get_include()]
> `---------------------------------------------------
>
> The next error is:
>
> ,-----------------------------------------------------------------------------------------------------
> | mvpa2/clfs/libsvmc/svmc_wrap.c:2718: error: expected specifier-qualifier-list before ‘svm_parameter’
> `-----------------------------------------------------------------------------------------------------
>
> which is due to my libsvn not being in an expected place and can be
> fixed by adding the appropriate options.  However, I get a further error:
>
> ,------------------------------------------------------------------------------------------------
> | # python setup.py build_ext --with-libsvm -I/usr/include/libsvm
> | running build_ext
> | building 'mvpa2.clfs.libsvmc._svmc' extension
> | swigging mvpa2/clfs/libsvmc/svmc.i to mvpa2/clfs/libsvmc/svmc_wrap.c
> | swig -python -I. -I/usr/lib64/python2.6/site-packages/numpy/core/include
> | -I/usr/include/numpy -I/usr/include/libsvm-3.0/libsvm
> | -I/usr/include/libsvm-2.0/libsvm -I/usr/include/libsvm
> | -I/usr/local/include/libsvm -I/usr/local/include/libsvm-2.0/libsvm
> | -I/usr/local/include -o mvpa2/clfs/libsvmc/svmc_wrap.c
> | mvpa2/clfs/libsvmc/svmc.i
> | gcc -pthread -fno-strict-aliasing -O2 -g -pipe -Wall
> | -Wp,-D_FORTIFY_SOURCE=2 -fexceptions -fstack-protector
> | --param=ssp-buffer-size=4 -m64 -mtune=generic -D_GNU_SOURCE -fPIC
> | -fwrapv -DNDEBUG -O2 -g -pipe -Wall -Wp,-D_FORTIFY_SOURCE=2 -fexceptions
> | -fstack-protector --param=ssp-buffer-size=4 -m64 -mtune=generic
> | -D_GNU_SOURCE -fPIC -fwrapv -fPIC
> | -I. -I/usr/lib64/python2.6/site-packages/numpy/core/include
> | -I/usr/include/numpy -I/usr/include/libsvm-3.0/libsvm
> | -I/usr/include/libsvm-2.0/libsvm -I/usr/include/libsvm
> | -I/usr/local/include/libsvm -I/usr/local/include/libsvm-2.0/libsvm
> | -I/usr/local/include -I/usr/include/libsvm -I/usr/include/python2.6 -c
> | mvpa2/clfs/libsvmc/svmc_wrap.c -o
> | build/temp.linux-x86_64-2.6/mvpa2/clfs/libsvmc/svmc_wrap.o
> | mvpa2/clfs/libsvmc/svmc_wrap.c:3210: error: conflicting types for ‘svm_node_array_set’
> | mvpa2/clfs/libsvmc/svmc_wrap.c:3204: note: previous definition of ‘svm_node_array_set’ was here
> `------------------------------------------------------------------------------------------------
>
> This seems to be because 'gcc' is used to compile, rather than 'g++',
> and the file name ends in '.c', so the compiler expects C rather than
> C++.  If I copy the last call and replace 'gcc' with 'g++', it compiles.
>
> I assume these difficulties do not normally occur and that my set-up is
> somehow broken.  Does anyone have an idea what the problem might be?
>
> Cheers
>
> Loris

I found the answer myself.  On Scientific Linux (SL) and presumably
other RedHat-based systems, the 'numpy' package does not contain certain
submodules, such as 'distutils', which are present in the Debian 'numpy'
package.  On SL I had to install 'numpy-f2py', which I assumed just had
something to do with converting Fortran to Python.  Apparently not.

Cheers

Loris