.. _sec-packaging-python:

==================
 Packaging python
==================

Motivation and plan
===================

We talk a lot about packaging, and there is much to be said.  [FIXME:
outline here the issues surrounding system-native packaging, versus
handing in a single archive, versus containerized approaches, ...]

Plan: to explore the myriad ways in which one can package python, and
try to make sense of it.  The ultimate goal is to find a clean way to
make an rpm (or dpkg) out of a well-structured python package.


Prerequisites
=============

An RPM-based (and later debian-based) system, or maybe VM or container.

Concepts and the big problem
============================

https://xkcd.com/1987/

The challenge is to deliver python software that might include
C/C++/FORTRAN shared libraries, and to deliver it robustly.

This is made very difficult by the fact that Python has had many
different packaging and delivery approaches over time.

Compare it to the old days of C, then the progression of ruby, golang,
rust, ...


Listing all the python packaging approaches
===========================================

As with the distinction of dpkg and apt, and of rpm and yum/dnf, there
is probably a similar distinction between a python package and a
channel of packages with coordinated dependencies -- maybe wheel
vs. pip?

As I start writing (2021-10-26) I do not yet know enough to
distinguish these, especially because I think that wheels are
superseded by new approaches (distutils -> setup tools maybe? or wheels
-> eggs?)

Poetry
------

Apparently after setup you just do two things to make a package:

.. code-block:: console

   poetry build -f wheel
   poetry publish

But before that you have to enter the poetry gestalt, starting at:

https://python-poetry.org/docs/basic-usage/

So I did this on a CentOS7 machine with python3 installed from the
rh-python36 software collection.

I did the following:

.. code-block:: console

   curl -sSL https://raw.githubusercontent.com/python-poetry/poetry/master/get-poetry.py | python -
   poetry --version   # returns 1.1.11 at this time, 2021-10-26
   mkdir -p ~/repo/
   cd ~/repo/
   poetry new try-poetry
   cd try-poetry
   git init .
   git add `find . -type f`
   git commit -a -m "initial files created by potry new"

Now you can look at `pyproject.toml` to see the project metadata.

Eventually I run in to the fact that if I do:

.. code-block:: console

   poetry add cartopy@0.19.0.post1
   poetry add matplotlib
   poetry run try_poetry/try_poetry.py

it works, but it required that I do an "apt install" of the projection
library.

So I set this aside for now, since it looks like I will really need
conda stuff.


miniconda inside an RPM
-----------------------

I'd like to use conda, since it handles the building of C/FORTRAN
libraries.

Links:

https://stackoverflow.com/questions/44280661/reallocate-full-conda-installation

https://github.com/pelson/conda-rpms

try things like:

::

   curl https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh > Miniconda.sh
   bash Miniconda.sh -b -p $HOME/conda
   
   conda create -y --prefix /tmp/try_conda_rpm
   # conda init bash
   conda activate /tmp/try_conda_rpm
   conda install -y cartopy --prefix /tmp/try_conda_rpm


Digging deep into some of them
==============================



Additional resources
====================

Here are various links that I started with, to then come up with the
recipes I tried.

https://news.ycombinator.com/item?id=26733423