Pymatgen requiring Python 3.6?

[ajjackson]

This build is blowing up because Pymatgen is using f-strings which are only available in very recent Python versions. Should/can we pin Pymatgen to a compatible version in setup.py?

[ajjackson]

Issue opened on Pymatgen tracker: materialsproject/pymatgen#1522

[ajjackson]

It looks like there is a way to set different dependencies in the sumo setup.py depending on the local Python version. https://www.python.org/dev/peps/pep-0508/#environment-markers

As much as I would like to use f-strings, it seems a bit early to drop Python 3.5 for sumo if we can avoid it: many of our users are running sumo on cluster login nodes which have a limited selection of Python modules.

[ajjackson]

Pymatgen team has confirmed their move to 3.6 (Issue) and put the requirement in setup.py.

We have two options:

• Make Python 3.6 a hard requirement for Sumo as well (as I said in that Issue I think this is ok for our users but it's a bit tight for comfort.)
• Try and do some setup.py magic to pin 3.5 users to an older version of Pymatgen. This could be unsustainable in the medium term as quite a few bugs/feature-requests in sumo have been addressed by making changes to Pymatgen...

Maybe do a numbered 3.5-compatible release of Sumo that pins to an appropriate Pymatgen version, then move further Sumo development to 3.6-only? That way people on 3.5 have a known sumo version number they can safely install.

[utf]

I also think Python 3.6 is a little close for comfort.

I like the idea of a numbered 3.5 compatible release.

[ajjackson]

Closing as the immediate issue has been fixed. We still need to plan a "final 3.5 release" before things diverge too far.