Open Discussion: Use python dependencies management

[arneboockmeyer]

I would recommend to use a python based dependencies management system. Therefore I would create a setup.py with details about the project and all dependencies (like aexpr or pylxd).

See: https://packaging.python.org/tutorials/packaging-projects/
And: https://python-packaging.readthedocs.io/en/latest/dependencies.html
Example: https://github.com/active-expressions/active-expressions-static-python/blob/master/setup.py

Advantages:

• Use common way to manage python dependencies
• Let python manage his dependencies itself - and not docker
• Offers the option to have a pip installation of the testbed

[lpirl]

We should definitively provide a ``setup.py``. Everything beyond ``pip install <repo-url>`` is too much burden for users.

[arneboockmeyer]

👍 @lpirl I totally agree. And it would be really really nice, if we are able to provide the functionality of the testbed just with pip install <repo-url>

[arneboockmeyer]

Then the installation inside of the container could only be pip install <repo-url>. This would keep the dependencies at one place.

[mgjm]

If we manage to build an easily installable package for ns-3, then our cohydra framework can depend on this ns-3 package. But the current situation is that the workflow to get the ns-3 python package is quite tedious and requires manual recompilation. Therefore I think it is out of the scope for this project to build such an easily installable python package for ns-3.
But all other dependencies can be managed with e.g. pip.

But then it would still not be possible to pip install cohydra 🤔

[lpirl]

It is true that cohydra has dependencies beyond setup.py and that we probably won't focus on streamlining the on-boarding process.
From my understanding, this issue is about the Python packages which cohydra requires and which can be installed from the CheeseShop right away.

[arneboockmeyer]

@mgjm Yes, we will not build a ns-3 pip package or something like that. But we also do not install docker, lxd or sumo on the machines. But we should cover as much as possible at a single point. The setup.py is the intended way from python and we should follow that. I opened this issue because I tried to run the lxd branch and it crashed the first time because of a missing dependency. We should fix that.

[lpirl]

For installing cohydra with all its dependencies, we /could/ provide – for example – an Ansible role but that's another discussion (just to put the idea into heads here).