Pip check failing due to pyiron_base vs mendeleev

[liamhuber]

Our pip checks fail:

mendeleev 0.12.1 has requirement SQLAlchemy<2.0.0,>=1.4.0, but you have sqlalchemy 2.0.1.

Because:

• pyiron_base/.ci_support/environment.yml: - sqlalchemy =2.0.3
• mendeleev/pyproject.toml: SQLAlchemy = "^1.4.0" (carat means don't allow changes to the leftmost non-zero digit)

This is also causing pip-check failures in all our downstream repos that depend on atomistics. Mybinder envs are still compiling so this is not an earth-shattering problem to be having, but it would be good to fix. I'm not going to get to it this afternoon, I just wanted to document exactly what's happening.

Unfortunately, we seem to be up-to-date with the latest version of mendeleev, so we'll either need to convince them to bump their dep, or pin pyiron_base's dep lower.

[jan-janssen]

Yes, the short answer is there is an issue with pandas requiring sqlalchemy=1.4.6 but there were no restrictions, so we currently primarily hope the update pandas soon #949

[liamhuber]

So this is actually becoming a bit of an issue for me. For packages that depend on atomistics, if I don't pin the atomistics version, conda pushes me back down to atomistics 0.2.34 (and corresponding base). If I do pin the atomistics version, MyBinder complains that the environment is unresolvable and I can't open any notebooks on binder.

I know it's more or less out of our control, I just wanted to clarify my earlier comment that "Mybinder envs are still compiling"... I guess I was previously just getting atomistics downgraded and it was a non-issue...

[liamhuber]

Pip check works after Jan's downgrade of the SQLAlchemy dependency. A long term solution still waits on the upstream libraries, but we're not getting test failure any more, which is good enough for me.