Search code examples
pythonattributesschemamarshmallow

What is the difference between a Marshmallow field defined with `data_key` and one defined with `attribute` (with the identifiers reversed?)

Is there a difference between a marshmallow.Schema (v3.0+) where a field with name foo is defined with attribute="bar", and another where a field with name bar is defined with data_key="foo"?

Both seem to serialize and deserialize dictionaries and other simple objects in the same way:

import marshmallow

class MySchema1(marshmallow.Schema):
    foo = marshmallow.fields.String(attribute="bar")

class MySchema2(marshmallow.Schema):
    bar = marshmallow.fields.String(data_key="foo")

schema1 = MySchema1()
schema2 = MySchema2()

dictionary = {"bar": "hello world"}
assert schema1.dump(dictionary) == schema2.dump(dictionary)
assert schema1.load(schema1.dump(dictionary)) == schema2.load(schema2.dump(dictionary))

class Thingy:
    def __init__(self):
        self.bar = "hello world"

instance = Thingy()
assert schema1.dump(instance) == schema2.dump(instance)
assert schema1.load(schema1.dump(instance)) == schema2.load(schema2.dump(instance))

The above passes. This isn't causing any errors in my project currently, but I am curious what the difference is! Thanks in advance.

Solution

  • You are right. Both schemas behave identically.

    This could be seen as redundant API: from your example, you could wonder why have attribute if data_key provides the same functionality.

    In practice, it is useful to have both, because it allows to specify keys that are invalid python variable names for both load and dump keys.

    class MySchema(marshmallow.Schema):
    foo = marshmallow.fields.String(attribute="bar-load", data_key="bar-dump")

    AFAIK, this is the reason we didn't drop attribute in marshmallow. There may be other reasons, but this one seems like a good one already.