Allow annotations to define settings

[ziima]

Since Settings are basically smart dataclass, we might use annotations in settings definition to allow even simpler setup.

The goal would be

class MySettings(AppSettings):
    my_foo: str = Setting(default='my_default')
    my_bar: int = Setting(required=True)

or even yet simpler version

class MySettings(AppSettings):
    my_foo: str = None
    my_bar: int

What do you think?

[pawamoy]

You mean

class MySettings(AppSettings):
    my_foo: str = "my_default"
    my_bar: int

?

Well yeah, that would be great 😄 !

We would just need to make sure that every feature (setting) of the app can be translated using type annotations. Then refactor 🙂

Not sure what would validators become though 😕

[stinovlas]

While this can work for the simple type settings, I can't imagine any sensible syntax for nested settings.

I believe that current notation is succinct enough for all practical intents and purposes. However, I wouldn't mind adding this notation as an alternative as long as I still have access to StringSetting et al.

[ziima]

As a middle step, I would propose to translate current Setting to accept annotation-style type argument to check the value against, i.e. add type argument (find better name if possible) and use it to replace TypeValidators.

The problem I was recently facing, was enabling None value in StringSetting. This would help this and most likely other issues.

[pawamoy]

I'm looking again at pydantic: https://pydantic-docs.helpmanual.io/usage/settings/ and https://pydantic-docs.helpmanual.io/usage/validators/. I think you'll find it interesting as well.