RFC: Remove attribute parameter from fields

[sloria]

With data_key to designate the serialized key and the field name to designate the deserialized attribute, I can't think of a use case for attribute. Are there cases where the field name would need to be different from the deserialized attribute?

h/t @taion

cc @lafrech @deckar01

[lafrech]

There could be cases where the attribute name is not a valid Python variable name.

Currently, get_value first tries to access by index, then by attribute, so I assume something like object['my-attribute'] is expected, but you couldn't write

class MySchema:
    my-attribute = fields.Int()

I don't have a real-life use case for that. Nor do I have a real-life use case for the "access by index" feature.

[sloria]

@lafrech You can use include for that use case:

class MySchema(Schema):
    class Meta:
        include = {
            "my-attribute": fields.Int()
        }

[sloria]

There's some code in List and Tuple to for accessing attributes on the inner objects. This test demonstrates the use case:

    def test_list_field_respect_inner_attribute(self):
        now = dt.datetime.now()
        obj = DateTimeList([now])
        field = fields.List(fields.Int(attribute='day'))
        assert field.serialize('dtimes', obj) == [now.day]

This use case can still be met by using fields.Pluck, though.

    def test_list_field_respect_inner_attribute(self):
        now = dt.datetime.now()
        obj = DateTimeList([now])

        class DateTimeSchema(Schema):
            day = fields.Int()
        field = fields.List(fields.Pluck(DateTimeSchema, 'day'))
        assert field.serialize('dtimes', obj) == [now.day]

This seems like a more consistent API anyway.

[sloria]

Ah, removing attribute would break use cases where you want multiple output fields based on a single attribute:

class UserSchema(Schema):
    created = fields.DateTime()
    created_formatted = fields.DateTime(format='%Y-%m-%d', attribute='created', dump_only=True)
    created_iso = fields.DateTime(format='iso', attribute='created', dump_only=True)

[lafrech]

removing attribute would break use cases where you want multiple output fields based on a single attribute

This could be achieved using pre_dump to duplicate the value.

[sloria]

This could be achieved using pre_dump to duplicate the value.

I guess, but that's awkward. It feels wrong to set arbitrary attributes on an object.

[taion]

It could make more sense just to discourage the use of attribute (and instead recommend data_key) in most cases, just to make the effective API surface area smaller.