Generically iterating over python dataclass annotations

Question

I have a super dataclass (Settings) and multiple children who inherit the settings of the super dataclass (FSettings, BSettings).

How can I pass a dictionary of "arguments" to a child dataclass in a way where it sets them, if applicable?

from dataclasses import dataclass

@dataclass()
class Settings:
    verbose: bool 
    detailed: bool 



@dataclass()
class FSettings(Settings):
    vals: list 
    def __init__(self, args):
        for arg in args:
            if arg in list(self.__annotations__):
                self.__setattr__(arg, args[arg])
        
if __name__ == '__main__':
    args = {'vals': ["a", "b"], 'verbose': True, 'detailed':False}
    s = FSettings(args)
    print(s)

If I print list(self.__annotations__) it shows ['vals'], ignoring verbose and detailed

Adin D · Answer 1 · 2020-07-10T20:45:06.777

(My own answer after trying for an hour)

This appears to work, but you have to default the values to "none", as the FSettings values initiate after the Settings values. Without a default, hasattr will return false.

The init can also be moved to the parent class since it is generic. However, a super().__init__() is needed in the child otherwise the parent values populate wrong (inconsistent error, not entirely sure why)

from dataclasses import dataclass

@dataclass() 
class Settings:
    verbose: bool = None
    detailed: bool = None
    def __init__(self, args):
        for arg in args:
            if hasattr(self, arg):
                self.__setattr__(arg, args[arg])
            else:
                print(f"Settings class could not locate {arg} as a setting. Skipping...")

@dataclass() 
class FSettings(Settings):
    vals: list = None


    def __init__(self, args):
        super().__init__(args)


if __name__ == '__main__':
    args = {'vals': ["a", "b"], 'verbose': True, 'detailed':False}
    s = FSettings(args)
    print(s)

It seems weird to have an `__init__` method on a dataclass when they are automatically generated for you. You could remove it from the `Settings` class and create an instance simply with `s = Settings(**args)`. You might also look at `__post_init__` a special method just for dataclasses so you don't have to overwrite the autogenerated `__init__` — Davos, Jan 25 '21 at 14:21

Evgeniy_Burdin · Answer 2 · 2020-07-11T09:28:28.973

0

from dataclasses import dataclass
from typing import Optional


@dataclass
class Settings:
    verbose: Optional[bool] = None
    detailed: Optional[bool] = None

    def __post_init__(self) -> None:

        args = []
        if self.verbose is None:
            args.append('verbose')

        if self.detailed is None:
            args.append('detailed')

        if args:
            s = f"Settings class could not locate {args} as a setting."
            s += " Skipping..."
            print(s)


@dataclass
class FSettings(Settings):
    vals: Optional[list] = None


args = {'vals': ["a", "b"], 'verbose': True, 'detailed': False}
f_settings = FSettings(**args)

args = {'vals': ["a", "b"]}
f_settings = FSettings(**args)
# Settings class could not locate ['verbose', 'detailed'] as a setting. Skipping..

edited Jul 11 '20 at 09:28

answered Jul 11 '20 at 08:36

Evgeniy_Burdin

627
5
14

Interesting response, but it is not generic. Your answer requires each variable to be checked in post init. – Adin D Jul 14 '20 at 21:57
1

You can make import: `from dataclasses import fields`. Then in the loop get each field: `for field in fields(self)`. For each field you can get: `field.name`, `getattr(self, field.name)` - field value, `field.type` - field annotation. Based on this data, you can make any generalized check of fields. – Evgeniy_Burdin Jul 15 '20 at 04:50

Generically iterating over python dataclass annotations

2 Answers2