Validate data with a schema

Define a dataclass with the fields and types your application expects, then merge external values into OmegaConf.structured() output. Validation runs as the values are merged:

>>> from dataclasses import dataclass, field
>>> from omegaconf import MISSING, OmegaConf
>>> @dataclass
... class Server:
...     host: str = "localhost"
...     port: int = MISSING
>>> schema = OmegaConf.structured(Server)
>>> cfg = OmegaConf.merge(schema, {"port": 8080})
>>> cfg.port
8080
>>> OmegaConf.merge(schema, {"port": "not a port"})
Traceback (most recent call last):
...
omegaconf.errors.ValidationError: Value 'not a port' of type 'str' could not be converted to Integer
    full_key: port
    object_type=Server

This works with configs loaded from YAML as well as Python dictionaries. Mark required fields with MISSING; reading them before supplying a value raises MissingMandatoryValue. The resulting config retains its schema for later validation.

To keep a dataclass field outside the config, set metadata={"omegaconf_ignore": True} on its dataclasses.field() (or the equivalent attrs field):

>>> @dataclass
... class Example:
...     port: int = 80
...     runtime_only: int = field(default=1, metadata={"omegaconf_ignore": True})
>>> cfg = OmegaConf.structured(Example)
>>> list(cfg.keys())
['port']

See structured configs for the schema model.