Update for upcoming OpenFF Toolkit API breaks

[mattwthompson]

Hi OpenMM developers,

I'm seeing some test failures in our example that uses openmmforcefields with our big biopolymer feature branch, probably due to Atom.element being removed in our upcoming API break. This change should fix that and be backwards-compatible.

[mattwthompson]

If you don't want to shoulder the complexity of testing against an unreleased version, you can grab it with a similar command locally (pip install git+https://github.com/openforcefield/openff-toolkit.git@topology-biopolymer-refactor) and have a look around for yourself.

[mattwthompson]

And for the record: I'm not suggesting this change be merged in; I want to show in the context of this PR that it should pass with the new API. I will revert this before merging if approved.

[jchodera]

Thanks for this! We should test against both the current and preview versions, ideally. But to do that, we can't just eliminate testing the current version---I'll have to rework the test matrix.

[jchodera]

Oh, that's an unexpected change.

@mattwthompson, can you point to some information about the motivation for this? Is the idea that Atom objects don't have to be physical elements anymore?

[jchodera]

And if symbol supports non-physical elements or made-up symbols now, is there an API for telling whether the element is a real element or not?

[mikemhenry]

This is the PR where the change happened: openforcefield/openff-toolkit#1182
It comes from switching to a different package.

[mikemhenry]

@jchodera this is the only modification needed to get openmmforcefields working with the changes in the openff-toolkit/openff-element and given they are done in a backwards compatible way, I am happy with them. If you are happy with it, I can make a new PR add this changes + tests the latest version of the openff-toolkit.

[jchodera]

Thanks for this, @mattwthompson!

I'll need to finish up #182 first, but then will tackle integrating this PR.

[richardjgowers]

It would be great if this could get merged, this is breaking stuff for openfe currently

[jchodera]

Thanks!

[mattwthompson]

Thanks!

[mattwthompson]

Unfortunately our tests are still failing; I think L966 in this loop needs a similar change:

openmmforcefields/openmmforcefields/generators/template_generators.py

Lines 960 to 967 in a59b144

	# Generate atom types
	atom_types = etree.SubElement(root, "AtomTypes")
	for particle_index, particle in enumerate(molecule.particles):
	# Create a new atom type for each atom in the molecule
	paricle_indices = [particle_index]
	atom_type = etree.SubElement(atom_types, "Type", name=particle.typename,
	element=particle.element.symbol, mass=as_attrib(particle.element.mass))
	atom_type.set('class', particle.typename) # 'class' is a reserved Python keyword, so use alternative API

if I had to guess, I didn't notice this previous because I was grepping around for atom.element.symbol or something similar - and at first glance the traceback looks like particle is an OpenMM object of sorts, not an OpenFF Atom object.