Question r.e. deprecation of Automatic Schema Matching

[sleepwalker2017]

I wrote a yaml file and want to enable type check during runtime.

It seems I can write a data class and register it with the same name with the config file.

I tried it and it worked. But I see the warning.

Will it not be supported in later versions? How can I use it?
Or if it will still be supported, the warning seems a bug?

[sleepwalker2017]

My code is here.

import hydra
from omegaconf import DictConfig, OmegaConf
import time
from dataclasses import dataclass, field
from hydra.core.config_store import ConfigStore

@dataclass
class InferenceConfig:
    backend: str
    batch_size: int
    input_names: list
    output_name: list = field(default_factory = lambda: list())

@dataclass
class TrainConfig:
    batch_size: int
    data_set: str

@dataclass
class Config:
    device: str
    job_name: str
    inference: InferenceConfig
    train: TrainConfig

cs = ConfigStore.instance()
cs.store(name="config", node=Config)

@hydra.main(config_path="conf", config_name="config", version_base='1.2.0')
def main(cfg: Config) -> None:
    print(cfg)
    print(OmegaConf.to_yaml(cfg))
    with open('result', 'w') as f:
        f.write(OmegaConf.to_yaml(cfg))

if __name__ == "__main__":
    main()

[Jasha10]

There feature you've identified is called "Automatic Schema Matching".
It is deprecated; see the docs here for the deprecation rationale and migration suggestions.

Automatic Schema Matching means that a ConfigStore schema with the same name/group as your yaml file will be automatically merged with the contents of your yaml file. The recommend migration is to use a different name (such as config_schema instead of config) when registering your dataclass in the ConfigStore, and then explicitly include the config_schema as an element in the defaults list of your yaml file.