Support ForwardRef in config types

[danielgafni]

What's the use case?

Sometimes Python annotations can't be specified as objects existing at runtime.
This includes several cases:

• types not yet defined in the module, for example:

from __future__ import annotations

class Foo:
    bar: "Bar"

class Bar:
    ...

• types coming from optional dependencies which may not be present at runtime:

from typing import TYPE_CHECKING

if TYPE_CHECKING:
    from optional_dependency import Bar

class Foo:
    bar: "Bar"

sometimes these types may be provided by .pyi files (it is not possible to import them at runtime at all), for example:

from typing import TYPE_CHECKING

import dagster as dg

if TYPE_CHECKING:
    from obstore.store import RetryConfig

class StoreConfig(dg.Config):
    retry_config: "RetryConfig | None"

such types from mypy_boto3_* are extensively used in dagster-aws (mostly in Pipes).

Currently Dagster doesn't allow using forward references in dagster.Config or dagster.ConfigurableResource. It's a pity because many libraries provide these kind of type annotations (especially in TypedDict form which I'm not sure if Dagster supports at all) and the ability to use them would enable easier and more correct configuration in Dagster.

Ideas of implementation

No response

Additional information

No response

Message from the maintainers

Impacted by this issue? Give it a 👍! We factor engagement into prioritization.

[danielgafni]

cc @ion-elgreco

[ion-elgreco]

Is it because of dagster or is it due to pydantic?