Rework serialization and model dumping for v2 models#393
Rework serialization and model dumping for v2 models#393loriab merged 15 commits intoMolSSI:next2026from
Conversation
Codecov Report❌ Patch coverage is 🚀 New features to boost your workflow:
|
qcelemental/tests/test_molecule.py
Outdated
| if "v2" in request.node.name: | ||
| assert isinstance(water_dimer_minima.model_dump(mode="json")["geometry"], list) | ||
| else: | ||
| assert isinstance(water_dimer_minima.dict(encoding="json")["geometry"], list) |
There was a problem hiding this comment.
might want to add some of these changes to the MIGRATION.rst guide
qcelemental/tests/test_molecule.py
Outdated
| with warnings.catch_warnings(): | ||
| warnings.simplefilter("ignore") | ||
| mol_dict = water_dimer_minima.dict() | ||
| mol_dict = water_dimer_minima.model_dump() |
There was a problem hiding this comment.
depending on final disposition of def dict, may want to add this back with a comment to leave alone for testing dict() operational.
| if "v2" in request.node.name: | ||
| assert obj.model_dump(exclude_unset=True).keys() == {"scf_one_electron_energy"} | ||
| else: | ||
| assert obj.dict().keys() == {"scf_one_electron_energy"} |
There was a problem hiding this comment.
revisit when AtomicProperties None's decided
| assert obj.return_gradient.shape == (4, 3) | ||
| assert obj.scf_total_hessian.shape == (12, 12) | ||
| assert obj.model_dump().keys() == {"calcinfo_natom", "return_gradient", "scf_total_hessian"} | ||
| assert obj.model_dump(exclude_unset=True).keys() == {"calcinfo_natom", "return_gradient", "scf_total_hessian"} |
There was a problem hiding this comment.
these three also revisit after AtomicProperties and WavefunctionProperties
| if "v2" in anskey: | ||
| instance = model(**instance.model_dump()) | ||
| else: | ||
| instance = model(**instance.dict()) |
| # UNCOMMENT IF NEEDED FOR UPGRADE REDO?? | ||
| def dict(self, **kwargs): | ||
| warnings.warn("The `dict` method is deprecated; use `model_dump` instead.", DeprecationWarning) | ||
| return self.model_dump(**kwargs) |
There was a problem hiding this comment.
revisit after def dict() decided
| if key not in self.model_fields_set: | ||
| continue | ||
| # Handle "by_alias" is always true | ||
| alias = self.__class__.model_fields[key].alias |
There was a problem hiding this comment.
hooray, I'm glad this custom logic is gone
|
Notice of bennybp#1 |
2 participants
Description
This reworks how qcelemental v2 models are serialized/dumped to bring them more inline with standard pydantic behavior.
Previously, models had custom options that forced unset fields to be excluded. This was generally ok, but also counterintuitive when some applications did want those default fields for compatibility/provenance reasons. The led to needing custom serializers.
Models had some (buggy) logic for handling serialization to/from aliased fields. This is now handled by pydantic v2 natively. (see model_config options
serialize_by_aliasandvalidate_by_alias).Because of the above, models were not tolerant of
Nonebeing passed in (MoleculeandWavefunctionProperties), even though it was the default value. For example, constructing a Molecule withfragments=Nonewould result in an error.So this PR removes the customization, makes the models tolerant of
None, and updates some of the tests.The somewhat breaking change: by default now, calling
.model_dump()will dump all fields, even unset andNonefields. This can be controlled withexclude_unset=Trueandexclude_none=True(see pydantic docs). This is also true for things likemodel_dump_json, and pydantic functions liketo_jsonable_python).The
dict()function is removed to also bring our terminology in line with pydantic.Changelog description
Status