[SciPy-user] Install scipy on ubnutu 8.10

Ondrej Certik ondrej@certik...
Mon Oct 6 20:40:12 CDT 2008


On Mon, Oct 6, 2008 at 2:28 PM, David Cournapeau
<david@ar.media.kyoto-u.ac.jp> wrote:
> Xavier Gnata wrote:
>> Hi,
>>
>> I have installed an ubuntu 8.10 within a kvm image.
>> The goal is to write an as simple as possible procedure to compile
>> scipy on this distribution.
>> I would like to be able to compile scipy only after having installed
>> some packages. No tricks.
>> Good news, g77 is not needed anymore :)
>
> It was already the case for 8.04, but it was a bit confusing: both g77
> and gfortran libraries were available. Is it better with 8.10 ?
>
>>
>> I have one issue:
>> scipy/sparse/linalg/dsolve/umfpack/umfpack.i:192: Error: Unable to
>> find 'umfpack.h'
>> I have installed libsuitesparse-dev providing /usr/include/suitesparse/umfpack.h
>> Of course, I can solve this with a symlink but it is very ugly.
>> It looks like there is something to fix in scipy auto-detection
>> because this "usr/include/suitesparse/umfpack.h" should be detected
>> and used.
>
> That's really the debian packagers' fault. Why do they think it is a
> good idea to change the header path of the library is beyond me; it
> breaks every single package which depends on it. That's stupid.
>
> We could get around it; but I though umfpack was being deprecated in
> scipy  (that is, we would do a scikit from it, but scipy would not
> depend on it anymore).

The way forward is to get involved with Debian packaging and get this fixed.

I did that for the scipy package and fixed that by applying a simple
patch to scipy. As to the default umfpack location, I also wondered
just like you, but it's useful to ask on the Debian list itself and
ask the people who do the packaging. :) So you can read the rationale
here (read the whole thread):

http://lists.alioth.debian.org/pipermail/pkg-scicomp-devel/2008-September/003133.html

I.e. citing:

"
we had a discussion with the author of suitesparse, the author of the
suitesparse interface in Octave and the person in charge of the fedora
octave/suitesparse package and we agreed in using /usr/include/suitesparse
as the place of the headers.
"

So if you (or anyone) have opinions on this, please join our teams in Debian:

http://wiki.debian.org/Teams/DebianScientificComputingTeam
http://wiki.debian.org/Teams/PythonModulesTeam

and let's get things fixed/discussed/moving.


As to building the official scipy/numpy packages in Debian or Ubuntu,
if something isn't working, it's my fault, as I did the uploads for
the last couple (a lot) revisions. Please report a bug in that case.



On Tue, Oct 7, 2008 at 12:59 AM, Xavier Gnata <xavier.gnata@gmail.com> wrote:
>>
>> That's really the debian packagers' fault. Why do they think it is a
>> good idea to change the header path of the library is beyond me; it
>> breaks every single package which depends on it. That's stupid.
>>
>>
> ok. I'm going to write a bug report. w&s.

Before you do, please read the thread I posted above.

Thanks,
Ondrej


More information about the SciPy-user mailing list