Load Nested field with only=str gives error when sent flat data structure instead of dictionary

When using a schema with the only attribute set to a string, the load method returns UnmarshalResult without errors nor data.

class MySchema(Schema):
    class Meta:
        fields = ('pk',)

    pk = fields.String()

MySchema(only='pk').load('an_id1234')
MySchema(only='pk').load({'pk': 'an_id1234'})

Both return the same UnmarshalResult UnmarshalResult(data={}, errors={})

Preferably the first should work, but the second could be used with a pre load method on the schema class.

Is there anything I have done wrong or is this a bug?

Note: I use the same schema for dumping data an thus use only='pk' and not only=('pk',) for getting the correct dump output

[erlingbo]

Okei. I’ll try to explain. I’ll want a dump that gives me a file (json or yaml) and then be able to load the file again later.

I want the relations to be flat, just using the primary key of the object it links to. This feature is documented here as only='pk'

When I try to load the file using the same schema I get an error UnmarshalResult(data={'pk': 123}, errors={'child': {u'_schema': [u'Invalid input type.']}})

If I use only=('pk',) the dumped relations will not be flat and include an object with the pk as attribute. E.g. {'pk': 123, 'child': {'pk': u'456'}} vs what I want: ``{‘pk’: 123, ‘child’: u’456’}`

I need to integrate with other teams, and the files need to be as human readable as possible, thus I want the flat structure. When this is a documented feature of the Nested field, I also expect it to be able to load the corresponding data.

[sloria]

This is the same issue addressed in #316. Further discussion should happen there.