Le samedi 15 avril 2017 à 04:40 +0000, Zbigniew Jędrzejewski-Szmek a écrit :
On Fri, Apr 14, 2017 at 08:15:54PM +0200, Jeandet Alexis wrote:
Yep, my point is more for new users or non-geek users who just want it
to work (make it default or automatic...). My point is more how would
we avoid classic pitfalls on Feora, on Linux?
I have been there, done that, and there *are* situations in which anaconda seems
like the easiest solution for newcomers. In my case I needed to provide
a "scientific python" environment with a bunch of different modules which
were not packaged for Fedora, or were packaged in versions that were slightly
too old. Installing anaconda and then 'conda install a b c' — with a carefully
tailored list of a, b, c — worked great. It certainly was less work than
making rpm packages for all the missing stuff, and much faster then installing
from source using pip.
In my experience the only working solution was pip in virtualenv, for example in my lab we use a lot Spacepy package which isn't in anaconda repos.
Then our install workflow is pretty easy:
sudo dnf install python3-virtualenv
mkdir ~/P3Venv
virtualenv-3.X --system-site-packages ~/P3Venv
source ~/P3Venv/bin/activate
pip install --upgarde jupyter matplotlib pandas spacepy numpy ...
Then we add few aliases on our bashrc to get quick access to jupyter-notebook and activate venv.
The main pitfall here is that we need gcc and some devel packages, here "dnf provides" is a friend.
So yeah, conda is a pretty nice solution for some environments, and I think
Fedora would be better served by providing conda in the standard repos and
allowing people to run with it, and fixing any rough edges, rather then
pretending it doesn't exist.
and I think orion would be amenable to making this into a real package.
Zbyszek
_______________________________________________
