fields: PersistentIdentifier not (de)serializing as expected

[slint]

The .schemas.fields.PersistentIdentifier marshmallow field, is assuming a wrong method signature when overriding Field._deserialize and Field._serialize:

# wrong
...
def _deserialize(self, value, attr, context):
    # "context" is actually the original "data" passed to "Schema(...).load"
    # and not the "Schema(context=context).load(...)" (see docs link above)
    pid = context.get('pid')
    return pid.get('pid_value') if pid else missing
...

# correct
def _deserialize(self, value, attr, data):
    pid = self.context.get('pid')
    return pid.get('pid_value') if pid else missing

When addressing this, keep in mind that:

• When doing a POST /api/records, there will be no request.view_args['pid_value'] and thus it won't be available in the loader schema's context, since the route doesn't have the PID yet, opposed to e.g. PUT /api/records/<pid(recid):pid_value>. The PID will be minted though and passed to the serializer and thus will be available in the context of Field._serialize because of this guy.
• When serializing search results (e.g. from a GET /api/records/ endpoint) the context['pid'] is not a invenio_pidstore.models.PersistentIdentifier, but actually a invenio_pidstore.fetchers.FetchedPID. It has pid.pid_value and pid.pid_type, but none of methods/attributes that the SQLAlchemy model does.
• Marshmallow has a pretty weird behaviour when it comes to loading/dumping fields that were not there originally. See a bunch of completely confused people trying to understand what went so terribly wrong in their lives... The closest thing to a solution is either using @post_load or this guy's custom Generated field. This is not optimal of course because in either way it requires educating developers on how to use marshmallow properly. The gist of it is:
  • default=<value or callable> is for providing a value when serializing
  • missing=<value or callable> is for providing a value when deserializing
  • Neither default or missing have access to context or data in any way...
  • Two solutions:
    • Modify our PersistentIdentifier field to some sort of Generator magic (seems a bit cleaner for the end-user)
    • Define some sort of @post_load mixin

[ppanero]

Fix in https://github.com/inveniosoftware/invenio-records-rest/pull/234