I see your point.

On the other hand, a new parameter would also increase the complexity for the user.

Maybe it should not be seen as re-ordering but just a "zipping" them correctly:


@dataclass
class Parent:
    i: int
    j: int = 0


@dataclass
class Child(Parent):
    k: int
    l: int = 1


The "naive" to define Child's __index__ is:

    __index__(self, i: int, j: int = 0, k: int, l: int = 1):

but wouldn't this make sense (given that it is previsible and deterministic)?


    __index__(self, i: int, k: int, j: int = 0, l: int = 1):