Hi Phil and Simon,
Thanks both for the input and advice (not citing it all here).
First the details. Simon wrote:
How did you installed poetry? Via Guix or something else?
Which python packages do you have in your profile?
Is Guix System or Guix on foreign distro?
I'm using Guix System, and installed poetry via Guix. I have the
following python packages installed: python, poetry,
python-jupyterlab (from guix-science), and python-black. And
guix-jupyter.
(I am aware that python-pandas is packaged for guix, but some
packages I will need down the road might not be; this project
is
also in collaboration with non guix users, so I am looking for
a
workflow which will work for them too.)
What are these packages?
I don't know yet, but I suspect the situation might crop up. Phil
pointed to the PyPi importer, and I haven't used it yet, but it
does seem to make things simpler if I need to add a package to
Guix.
Phil wrote:
ImportError: libstdc++.so.6: cannot open shared object file: No
such
file or directory
If it's only missing a reference to the underlying C++ library
you could
install this library using Guix, or if you're using a foreign
distro point to
the non-Guix version under /usr/lib which, at a guess, Poetry
may have
picked up - but this is going to cause you more trouble down the
line
than repackaging in Guix in my experience.
Yes, my experience with trying this for some Julia packages
(GLMakie) convinced me not to try LD_LIBRARY_PATH hacks any more.
Now the broader view. Phil:
Guix offers pretty much everything that virtualenvs offers, but
it does
things a little differently - you'll need to look at
environments,
profiles, manifests, and use of a guix.scm to recreate (IMHO) a
better
experience in the end.
I just set up a simple guix.scm for my current project, which
indeed does the job, and of course provides all the great
guarantees of Guix! Given the number of python packages currently,
and the existence of the PyPi importer, I'm pretty convinced it
covers my personal needs.
Now my wish is to find ways to collaborate with others on this
kind of project, who unfortunately often use macOS. I was
exploring the possibility of Guix on Darwin, but the best way to
do that looks like running Guix System inside Docker:
http://logs.guix.gnu.org/guix/2021-03-11.log#044004
https://lists.gnu.org/archive/html/guix-devel/2017-10/msg00111.html
So it seems that to collaborate with non-guix and non-GNU/Linux
users, the options are 1) maintain a second environment management
configuration (e.g. using poetry) to provide a stop-gap, or 2)
convince colleagues to run Guix in Docker. The situation for
Windows might be a bit better using the Linux Subsystem.
Thanks again all,
Sébastien
On 10 Mar 2021 at 19:02, Phil <p...@beadling.co.uk> wrote:
Hi Sébastien,
Sébastien Lerique writes:
Dear all,
(I am aware that python-pandas is packaged for guix, but some
packages
I will need down the road might not be; this project is also in
collaboration with non guix users, so I am looking for a
workflow
which will work for them too.)
Under the covers Poetry is just using regular Python virtualenvs
I think.
These don't work well with Guix for a number of reasons - but in
my
previous attempts to use them I found that Guix's use of
PYTHONPATH
meant that virtualenvs where never isolated from packages
installed by
Guix.
Guix offers pretty much everything that virtualenvs offers, but
it does
things a little differently - you'll need to look at
environments,
profiles, manifests, and use of a guix.scm to recreate (IMHO) a
better
experience in the end.
When I started using Guix I went to some lengths to harmonize
virtualenvs and Guix - and others have made more substantial
efforts
than I did - but you end up creating as many problems as you
fix:
https://lists.gnu.org/archive/html/help-guix/2020-11/msg00237.html
If you search the mail archive a couple of other people have
asked
similar questions to do with poetry, pyenv, pip, etc - they're
all
interesting reading and easy to grep with those keywords:
https://lists.gnu.org/archive/html/help-guix/
The advice I was given at the time was don't mix package
managers, and
in hindsight I'm glad repackaged what I needed under Guix.
I've packaged quite a few python libraries now, and most are
pretty
trivial to do - if they are in PyPi you can just use this:
guix import pypi <package_name>
If you need a later version than Guix provides (eg pandas), my
advice is
to use the ability to "inherit" the current definition in Guix
and tweak
the version. For Pandas I did this for a more recent version I
needed
and all I had to do was disable 2 unit tests (which failed for a
non-Guix reason).
Apologies - not exactly the answer you want (I have a stab at
that
below), but it's worth considering moving to Guix packaging
exclusively
to keep your life simple.
File
"/home/sl/.cache/pypoetry/virtualenvs/poetry-test-vOWX4_Vr-py3.8/lib/python3.8/site-packages/pandas/core/window/ewm.py",
line 9, in <module>
import pandas._libs.window.aggregations as
window_aggregations
ImportError: libstdc++.so.6: cannot open shared object file: No
such
file or directory
If it's only missing a reference to the underlying C++ library
you could
install this library using Guix, or if you're using a foreign
distro point to
the non-Guix version under /usr/lib which, at a guess, Poetry
may have
picked up - but this is going to cause you more trouble down the
line
than repackaging in Guix in my experience.
