Assign to __schema__ instead of Schema; load and dump helpers?

It may make sense to assign the generated schema to __schema__ instead of Schema for consistency with dataclasses.dataclass, which only assigns dunder methods.

@dataclass
class Artist:
    name: str

Artist.__schema__  # => Schema class

To improve the DX, you could add dump and load helper methods.

import datetime as dt

from marshmallow_dataclass import dataclass, load

@dataclass
class Album:
    title: str
    release_date: dt.date

album = load(Album, {"title": "Hunky Dory", "release_date": "1971-12-17"})

Schema contructor arguments can be passed as keyword arguments.

albums = load(Album, input_data, many=True)

This could even open the door to conveniences like

albums = load(List[Album], input_data)  # calls Album.__schema__(many=True).load(input_data)

What do you think?

[sloria]

Yep, sounds good–documenting class_schema as preferred reduces the effective API surface. Will work on a docs PR when I have time.

[lovasoa]

I don’t think there is really a need to deprecate @dataclass. It is a small function, and it will not be a pain to maintain. We can just make the change you suggested in the README, making the first example about class_schema, and talking about @dataclass only below. I developed this library for my own needs initially, and it was all about reducing boilerplate. I found that @dataclass helped me maintain short, clean, and well-organized model definitions. I can completely understand why one wouldn’t like it, though. Let’s just make it clear in the README that the central part of this library is class_schema, and that we also provide a few niceties like @add_schema and @dataclass.