[SciPy-User] Completing source install of numpy/scipy

[Miha Marolt]

I was trying to do something similar, not because of speed though, but because
packages in Debian stable are outdated. Here is what i found out:

You could make .deb packages of your customly build scipy and numpy. It is very
easy to do so with the checkinstall program (i use it on Debian, but i guess it
should work on Ubuntu too). All you have to do is that you install scipy/numpy
with the following command

su root
checkinstall python setup.py install --prefix=$HOME/local

This will automaticaly make .deb package and install it. But be aware that this
can break packages from repository that depend on numpy and scipy. I personally
build and installed numpy 1.5.1 like this, but had to uninstall it, because some
other packages that i installed from repository (mayavi2, glade,..) depended on
numpy < 1.5!


So i think it is generally a bad idea to mess with package management system
like this. What is the alternative? 

1. When writing scripts, you could change sys.path at the begining of your
script. That way only the script would use numpy and scipy in $HOME/local, while
all other programs would use numpy and scipy from repositories (in /usr/lib/..).

2. For interactive work with ipython for example, you have to go to directory
with your scipy/numpy and launch it from there. When you import scipy, your
custom version will be loaded. This doesn't work with "ipython -pylab"
unfortunately.

I only recentrly started building my own versions, so my solutions are probably
not that good. Any advice is very welcome.