dataclass concept extended to support binary serialization/deserialization with numpy

dtypeclass.py

from dataclasses import dataclass, astuple
import numpy as np


def dtypeclass(cls):
    cls = dataclass(cls)

    def frombuffer(cls, buffer, offset=0):
        values = np.frombuffer(buffer, cls._dtype, count=1, offset=offset)
        return cls(*values[0])

    def tobytes(self):
        return np.array(astuple(self), dtype=self._dtype).tobytes()

    setattr(cls, "_dtype", np.dtype(list(cls.__annotations__.items())))
    setattr(cls, "frombuffer", classmethod(frombuffer))
    setattr(cls, "tobytes", tobytes)
    return cls


@dtypeclass
class MyClass:
    b: np.float32
    a: np.int16


if __name__ == "__main__":
    a = MyClass(5.0, 2)
    print(a.tobytes())
    b = MyClass.frombuffer(a.tobytes())
    print(b)

output.txt

b'\x00\x00\xa0@\x02\x00'
MyClass(b=5.0, a=2)

This seems like a more modern approach to what I was trying to do with https://github.com/wrenoud/structobject. Especially because the _field_order member can be dropped because of the dictionary order guarantee introduced in 3.7.