Typing __new__ in dataclass_transform metaclass

Python Help
1

I can’t figure out the correct type annotations for __new__ and a couple of steps within __new__, in a metaclass that wraps dataclasses.dataclass and needs to do some work both before and after the actual creation of the dataclass. Here’s a cut-down version of my code with the errors marked:

import dataclasses
from types import GenericAlias, UnionType
from typing import Any, dataclass_transform

@dataclass_transform()
class DemoMeta(type):
    _attributes: dict[str, dataclasses.Field[Any]]

    def __new__(
        cls,
        cname: str,
        bases: tuple[type, ...],
        body: dict[str, Any]
    ) -> "DemoMeta":
        annotations = body.setdefault("__annotations__", {})
        for fname, ty in annotations.items():
            adj_default = adjust_field(ty, body.get(fname, dataclasses.MISSING))
            if adj_default is not dataclasses.MISSING:
                body[fname] = adj_default
        # error: incompatible types in assignment (expression has type
        # "type[Never]", variable has type "DemoMeta")
        klass: "DemoMeta" = dataclasses.dataclass()(
            # error: Argument 1 has incompatible type "DemoMeta";
            # expected "type[Never]"
            type.__new__(cls, cname, bases, body)
        )
        # error: argument 1 to "fields" has incompatible type "DemoMeta";
        # expected "DataclassInstance | type[DataclassInstance]"
        fields = dataclasses.fields(klass)
        attributes = { field.name: field for field in fields }

        klass._attributes = attributes
        return klass

def adjust_field(ty: type | GenericAlias | UnionType, default: Any) -> Any:
    # in the real program, this is not just a stub
    return default

class Demo(metaclass=DemoMeta):
    errors: list[tuple[str, str]]

If I remove the type annotation on klass, then it stops complaining about the call to dataclasses.fields, but instead it tells me klass needs a type annotation.

This code needs to work, and typecheck, in Python 3.11 and higher.

2

You have to add field from dataclasses as in:

from dataclasses import field

Then, in the Class Demo, you need to redefine the list from:

errors: list[tuple[str, str]] = []

to this:

errors: list[tuple[str, str]] = field(default_factory=list)

3

You are correct that the default value I had for that field normally wouldn’t work as intended. This was an error in cutting down the code; part of what adjust_field actually does is make that work as intended.

However, this has nothing to do with the question I asked. I have removed the distraction.

4

I copied the latest version that you have and an error is no longer being generated. I am using v.3.13.0.

Which Python version are you using?

5

As I said: This code needs to work, and typecheck, in Python 3.11 and higher.