Gentoo Archives: gentoo-science

From: fbissey@××××××××××××.nz
To: gentoo-science@l.g.o
Subject: [gentoo-science] [sage-on-gentoo] python 3.2 + numpy-1.5.1 + scipy = hurt
Date: Mon, 20 Feb 2012 01:02:41
Message-Id: 20120220132107.2sgkccsog880g4c4@webmail.slingshot.co.nz
Ok folks so python 3.2 has hit stable and a few problem that were plaguing
people using ~arch are now hitting us squarely in the face.
Namely you cannot emerge scipy (0.9/0.10) with python 3.2 and numpy-.1.5.1...

Using numpy-.16 is kind of functional... Coercion of sage numerical type to
available numpy numerical type is broken - in a strange way.
Numpy functions coded in C are not broken: numpy.arange for example.
Numpy functions coded in python are broken: numpy.linspace for example.

For a long time I didn't quite grasp the problem in
http://trac.sagemath.org/sage_trac/ticket/11334

But now with these details I think there is something either broken or
undocumented in numpy-1.6.x.

Moving sage to numpy 1.6 will break doctest and will require you to do explicit
casting if you want to do numerical stuff with numpy from sage.

So I propose that we move sage to numpy 1.6 with a warning to users about the
problems encountered.

The other solution is to tell people not use python 3.2 for scipy. This could be
done by adding a file in /etc/portage/env/sci-libs/ called scipy with just the
line:
USE_PYTHON="2.7"

I am using the later right now. Other suggestions welcome.

Francois