Test new python versions

[mikemhenry]

Testing python 3.9 and 3.10, will be useful to see where our dependencies are failing, openeye will probably be one of them

[codecov]

Codecov Report

Merging #880 (92c1648) into master (2a3614b) will increase coverage by 0.01%.
The diff coverage is 100.00%.

[ijpulidos]

@mikemhenry This looks much better now for testing, thanks for taking the time to improve this, it will help a lot.

@jchodera Do we want to closely keep up with OpenMM nightly? Such that if nightly tests fail we cannot merge.

[mikemhenry]

I was thinking that we don't make passing nightly required to merge, but in order to merge with nightly failing we need to make an issue in the openmm or perses tracker. I think it depends why nightly fails as well, if it is an API change that we can handle in a backwards compatible way, then we should

[ijpulidos]

@mikemhenry we should try running this on GPU CI as well, to check for possible NaNs.

[ijpulidos]

If these changes work, we should include some kind of Fixes #879 in the PR, I suppose.

[mikemhenry]

Right now the only failures are on nightly,

self = {'implicitSolvent': None, 'switchDistance': None, 'flexibleConstraints': False, 'drudeMass': Quantity(value=0.4, unit=dalton)}
fn = <bound method ForceField.createSystem of <openmm.app.forcefield.ForceField object at 0x7f81b9783e50>>

    def checkArgs(self, fn):
        """Throw an exception if any argument was never used."""
        parameters = signature(fn).parameters
        for key in self:
            if key not in self.accessed:
                if key not in parameters or self[key] != parameters[key].default:
>                   raise ValueError(f"The argument '{key}' was specified to {fn.__name__}() but was never used.")
E                   ValueError: The argument 'implicitSolvent' was specified to createSystem() but was never used.

It looks like this PR introduced this feature, openmm/openmm#3242
So we need to clean up our tests + be aware that in the future we need to make sure we are passing in args we actually use

[mikemhenry]

What isn't clear to me is why implicitSolvent isn't used in these two tests?
https://github.com/choderalab/perses/blob/master/perses/tests/test_geometry_engine.py#L1199
https://github.com/choderalab/perses/blob/master/perses/tests/test_geometry_engine.py#L1892

[mikemhenry]

@jchodera @ijpulidos
I think I can patch one of these failures but the other looks like I will need to change openmmforcefields

Should we just make issues about these failures but merge in the changes? I feel like our other PRs will benefit from these improvements and the failures are only on nightly

[ijpulidos]

This looks great! I'd only agree on raising separate issues with regards to the nightly tests that are failing as you already suggested. Thanks, LGTM!