How to define a schema with dynamic field names

Hi all,

I have the need to define a schema that basically represents a dict object where I know that the keys are going to be strings and the values are all going to have the same object format.

I know that having this defined as a field of a schema can be done by using the fields.Dict just like

class ValueSchema(Schema):
    att_1 = fields.String()
    att_2 = fields.String()

class RootSchema(Schema):
    root = fields.Dict(keys=fields.String(), values=fields.Nested(ValueSchema))

# dict conforming to the schema
 {
    "root": {
        "id-1":{"att_1":"a", "att_2":"b"},
        "id-2":{"att_1":"c", "att_2":"d"}
    }
}

But what I want is to have a schema that represents just the root object, i.e. without the root field, just like

{
    "id-1": {"att_1":"a", "att_2":"b"},
    "id-2": {"att_1":"c", "att_2":"d"},
    (...)
    "id-N": {"att_1":"x", "att_2":"y"},
}

Is this something that can be done with marshmallow?

Thanks in advance

[deckar01]

You can use Schema.from_dict() after you receive the data, only use the generated schema for that data, then generate a new schema when you get new data.

[deckar01]

The only remaining issue seems to be preferred syntax. I would make a new decorator for from_dict rather than trying to overload a schema class so that you can pass it into a decorator expecting a schema class. Feedback on how you ended up working around it would be appreciated.

I am closing this for now. If you have a proposal that wasn’t already discussed in #972 I am happy to reopen. I still promote unifying the load/dump schema interface with the field interface since it improves many other use cases, but the scope of that would probably justify a new issue.