Add DatasetFieldBase model

[sanders41]

Closes

Code Changes

• Add a DatasetFieldBase modes.
• Change DatasetField to inherit from DatasetFieldBase.
• Import DatasetFieldBase in __init__.py.

Steps to Confirm

• Run the test suite
• In the fidesops repository, modify FidesopsDatasetField to the following and run the test suite.

class FidesopsDatasetField(FideslangDatasetFieldBase):
    """Extends fideslang DatasetField model with additional Fidesops annotations"""

    fidesops_meta: Optional[FidesopsMeta]
    fields: Optional[List["FidesopsDatasetField"]] = []

    @validator("data_categories")
    def valid_data_categories(
        cls, v: Optional[List[FidesOpsKey]]
    ) -> Optional[List[FidesOpsKey]]:
        """Validate that all annotated data categories exist in the taxonomy"""
        return _valid_data_categories(v)

    @validator("fidesops_meta")
    def valid_meta(cls, meta_values: Optional[FidesopsMeta]) -> Optional[FidesopsMeta]:
        """Validate upfront that the return_all_elements flag can only be specified on array fields"""
        if not meta_values:
            return meta_values

        is_array: bool = bool(
            meta_values.data_type and meta_values.data_type.endswith("[]")
        )
        if not is_array and meta_values.return_all_elements is not None:
            raise ValueError(
                "The 'return_all_elements' attribute can only be specified on array fields."
            )
        return meta_values

    @validator("fields")
    def validate_object_fields(
        cls,
        fields: Optional[List["FidesopsDatasetField"]],
        values: Dict[str, Any],
    ) -> Optional[List["FidesopsDatasetField"]]:
        """Two validation checks for object fields:
        - If there are sub-fields specified, type should be either empty or 'object'
        - Additionally object fields cannot have data_categories.
        """
        declared_data_type = None

        if values.get("fidesops_meta"):
            declared_data_type = values["fidesops_meta"].data_type

        if fields and declared_data_type:
            data_type, _ = parse_data_type_string(declared_data_type)
            if data_type != "object":
                raise InvalidDataTypeValidationError(
                    f"The data type {data_type} is not compatible with specified sub-fields."
                )

        if (fields or declared_data_type == "object") and values.get("data_categories"):
            raise ValueError(
                "Object fields cannot have specified data_categories. Specify category on sub-field instead"
            )

        return fields

Pre-Merge Checklist

• All CI Pipelines Succeeded
• Documentation Updated
• Issue Requirements are Met
• Relevant Follow-Up Issues Created
• Update CHANGELOG.md

Description Of Changes

This relates to the issue discussed here where fidesops_meta can't be added to the fideslang DatasetField. I tried several different options, and this was the only one that successfully passed the fidesops test suite.

Other things tried:

1. Added anfidesops_meta field to DatasetField with a FidesopsMeta type. This didn't work because doing this would require fideslang to import fidesops.

I tried multiple iterations of the following ideas and they all either failed the fidesops test suite, or caused Pydantic to error.

1. Added fidesops_meta to DatasetField making the field that requires fidesops to be imported a generic in order to avoid the import.
2. Added fidesops_meta to DatasetField making the whole field generic and controlling things from fidesops.
3. Added fidesops_meta to DatasetField making the whole field generic and controlling things from fidesops. In addition to this, made the fields type in DatasetField Optional[List[Generic[T]]].
4. Added fidesops_meta to DatasetField making the whole field generic and controlling things from fidesops. In addition to this, made the fields type in DatasetField Optional[List[Generic[T]]]. Then only added validators to the inheriting class with no extra fields.
5. Made the fields type in DatasetField Optional[List[Generic[T]]], and only had fidesops_meta in the inheriting model.

[seanpreston]

This is a nice solution in the end, and respectful of ethyca/fidesops#454 (comment).