Fix nightly and conda CI (#1965)

* fix gbsa test by relaxing tolerance

* see if gbsa tests pass using OpenMM 8.1

* skip external examples in conda tests due to parmed incompatibility

* tighten tolerance again

* see if tests pass on older macoses

* force use of reference platform

* re-enable test matrix

* skip bad bond test

* update nagl failure test mol to avoid success due to new lookup table

* Remove parmed and downpin nagl in conda tests

* update releasenotes

* revert slow test changes

.github/workflows/conda.yml

@@ -134,4 +134,5 @@ jobs:
 
       - name: Run example notebooks
         run: |
-          python -m pytest -r fE -v -x --tb=short --nbval-lax --ignore=examples/deprecated examples
+          # External examples are temporarily skipped due to parmed incompatibility https://github.com/openforcefield/openff-toolkit/issues/1532
+          python -m pytest -r fE -v -x --tb=short --nbval-lax --ignore=examples/deprecated --ignore=examples/external examples

devtools/conda-envs/conda.yaml

@@ -8,4 +8,6 @@ dependencies:
   - pytest=7.4
   - pytest-rerunfailures
   - nbval
-  - parmed=3
+  - openff-nagl=0.4
+  # https://github.com/openforcefield/openff-toolkit/issues/1532
+  #- parmed=3

devtools/conda-envs/conda_oe.yaml

@@ -10,4 +10,6 @@ dependencies:
   - pytest=7.4
   - pytest-rerunfailures
   - nbval
-  - parmed=3
+  # https://github.com/openforcefield/openff-toolkit/issues/1532
+  # - parmed=3
+  - openff-nagl=0.4

docs/releasehistory.md

@@ -19,6 +19,12 @@ Releases follow the `major.minor.micro` scheme recommended by [PEP440](https://w
 
 ### Improved documentation and warnings
 
+### Tests updated
+
+- [PR #1965](https://github.com/openforcefield/openff-toolkit/pull/1965): Fixes slight difference in GBSA energy reported in [Issue #1964](https://github.com/openforcefield/openff-toolkit/issues/1964) by forcing the use of the Reference platform in OpenMM for both calculations.   
+- [PR #1965](https://github.com/openforcefield/openff-toolkit/pull/1965): Disables testing of parmed-dependent notebook in `external` examples, since ParmEd 3 is no longer compatible with our upstreams and ParmEd 4 raises an error for our use cases.     
+- [PR #1965](https://github.com/openforcefield/openff-toolkit/pull/1965): Uses a different molecule in test_nagl's test_unsupported_molecule_bond to avoid having molecule handled successfully by nagl's new lookup table. 
+
 ## 0.16.6
 
 ### Behavior changes

openff/toolkit/_tests/test_nagl.py

@@ -140,7 +140,7 @@ def test_unsupported_molecule_element(self):
             )
 
     def test_unsupported_molecule_bond(self):
-        mol = Molecule.from_smiles("C=[Cl+1]")
+        mol = Molecule.from_smiles("CCC=[Cl+]")
         err = re.escape("Molecule contains forbidden SMARTS pattern [#17:1]#,:,=[*:2]")
         with pytest.raises(ValueError, match=err):
             mol.assign_partial_charges(

openff/toolkit/_tests/utils.py

@@ -658,8 +658,9 @@ def compare_system_energies(
 
     # Create Contexts and compare the energies.
     integrator = openmm.VerletIntegrator(1.0 * openmm_unit.femtoseconds)
-    context1 = openmm.Context(system1, integrator)
-    context2 = openmm.Context(system2, copy.deepcopy(integrator))
+    from openmm import Platform
+    context1 = openmm.Context(system1, integrator, Platform.getPlatformByName("Reference"))
+    context2 = openmm.Context(system2, copy.deepcopy(integrator), Platform.getPlatformByName("Reference"))
 
     def map_energies_by_force_type(potential_energy1, potential_energy2):
         """Convert dictionary force_group -> energy to force_type -> energy."""