search

luigibonati / mlcolvar

A unified framework for machine learning collective variables for enhanced sampling simulations
MIT License
91 stars 24 forks source link

Bugfixes v1.1.1 #144

Closed luigibonati closed 2 months ago

luigibonati commented 2 months ago

In order to fix the dependency issues between our current PyPI version (mlcolvar=1.1.0) and the latest pytorch one (torch=2.3) which were reported in (#143) I would do a patch version v1.1.1 that:

I ran the tests locally with a fresh environment, and with these two modifications, they work fine. I was wondering whether we should also keep a stable (corresponding to the PyPI version) branch on which to run the CI in order to immediately detect dependency issues like this. @andrrizzi, any thoughts on this?

andrrizzi commented 2 months ago

I was wondering whether we should also keep a stable (corresponding to the PyPI version) branch on which to run the CI in order to immediately detect dependency issues like this.

That sounds like a good idea to me. The disadvantage would be that we'll have to apply bugfixes twice (in main and stable) when we find them.

An alternative could be add a separate nightly CI workflow to test the pypi installation, which shouldn't require maintaining a stable branch, and avoids running the entire test suite of the stable version every time.

luigibonati commented 2 months ago

I agree with your suggestion, I am working on it in #146 to add a new CI workflow! thank you