Notes from LLNL development work

[jvita]

I will be using this issue to track comments/bugs that I encounter while working with ColabFit for the LLNL project. I will update this periodically as I go.

• [ ] database.insert_property_definition should have an overwrite=True option. I kept making mistakes in my property definition, then having to manually delete the old definition. Knowing how to manually delete a document may be difficult if you are unfamiliar with MongoDB.
• [ ] get_cleaned_property_instances() seems to assume that property definition names have the format <type>-<name> (link to corresponding line in code). While this lets things like potential-energy, atomic-forces, and cauchy-stress to load the fields energy, forces, and stress, it should not be a mandatory naming convention.

[jvita]

• [ ] it may be desirable for delete_dataset to optionally also delete the linked COs/DOs/PIs/... (only allowed if no other DS points to them)

[jvita]

• [ ] property insertion fails if "_metadata" is not provided. See this line.

[jvita]

• [ ] it would be helpful if property_map supported nested key extraction. For example, when storing computed descriptors (or similar properties), in our workflow it was common to store various hyperparameters under atoms.info['_metadata'], then to want to extract the values during property insertion like property_map = {'hyperparam1': {'field': '_metadata.hyperparam1', 'units': None}}