The lack of complex types is somewhat intentional to keep the library simple and support importing anything as an environment variable.

Is the goal just to convert yaml from toml? If so, I'd recommend a purpose built tool such as yj.

If you also want to model these in the config, instead of using complex types, I would consider functionality that let you read a config from a specific key prefix in the file. This could work similar to the env_prefix provided by the BaseSettings class.

That would support your use case by allowing you to define multiple GoodConf classes, each with their own key prefix to load from.

Sorry my bad, I made a typo. I meant I was trying to use both yaml and toml config files as sources, a feature that this library readily provides.
My question originates from the fact that GoodConf already loads yaml or toml files with complex structures, such as the following:

class AppConfig(GoodConf):
    class InnerConfig(BaseModel):
        value: str
    inner: InnerConfig
    class Config:
        default_files = ["myproject.yaml", "myproject.toml"]

config = AppConfig()
config.load()

print(config.inner.value)

inner:
  value: "test"

[inner]
a = "test"

I also know that for the env variables the only supported format is a (quite ugly) JSON string, as the documentation of pydantic demonstrates.
So my doubt originates from the fact that only the logic implemented in generating the yaml and toml files is lacking this ability, since the rest of the stack seems to be ready for it.

Thanks anyway for you comment. Do you suggest to hook into the customise_sources method to read only the specific key instead of the whole file?