RFC: Record Controller

[lnielsen]

Co-writing on Codimd

The goal of this RFC is to clearly separate the data flow for Invenio into several layers: a data access layer, a business logic layer and view layer.

The data flow in Invenio has been shown to suffer from several issues that overall causes confusion as well as possibly produce hard-to-main code due to lack of very clear separation of responsibilities. Examples of issues with the data flow:

• Serializers can fetch records from both the database and Elasticsearch but these representations are not identical causing serializers to having to first transform them to a standard format.
• Marshmallow schemas in loaders have dependencies on the Flask request context making them impossible to use outside of a request.
• The $schema attribute has not clear place to be injected - sometimes it is added to the marshmallow loader other times not.
• Indexing receivers receive all signals instead of only for records they actually can handle.
• Indexing is based on the $schema attribute, but sometimes this needs further customization.
• Deposit needs to preserve some fields in the JSON that should not be overwritten by the user.
• A record class can be used in several modules (example Invenio-Records-REST and Invenio-GitHub), but does not provide proper business-level validation, just structural validation via the JSONSchema.
• The $schema attribute is supposed to be used for versioning, but it hides the feature.