Should

AssetIn

behave like pydantic’s Field? It would be possible to make assets more readable:

@asset
def my_asset(upstream: pd.DataFrame = AssetIn(metadata={"columns": ["A", "B"]})):
    ...

@sandy @schrockn

Hey @Daniel Gafni thanks for your patience here. I’ve been traveling for the last week and a half. I think using Annotated to co-locate

AssetIn

with the parameters they are associated with is potentially quite awesome actually. I don’t think your original suggestion is possible because that would set the default value to

AssetIn

unless that is leveraging some Python feature that I don’t know about.

Thanks Jick, forgive me for the impatience. I believe using AssetIn for default values would work too - Pydantic, FastAPI, Typer and some other tools are using this pattern. You would just have to handle these defaults in your decorator. Typer example:

import typer


def main(name: str = typer.Argument(...)):
    print(f"Hello {name}")


if __name__ == "__main__":
    typer.run(main)

This CLI program would fail if you don't supply an argument, because Typer treats

typer.Argument(...)

as a missing value for a required field. The

...

is an Ellipsis here (not some missing code). Pydantic example:

from pydantic import BaseModel, Field


class Model(BaseModel):
    a: int
    b: int = ...
    c: int = Field(...)  # Ellipsis again

All

a

,

b

and

c

fields are required. You could do the same by handling default `AssetIn`s correctly. So this shouldn't be a problem. Anyway, that's up to you to decide which of the approaches would be more clear and readable for the end-user and better for Dagster. But both should work. I suggest to take a look at how FastAPI is using

Annotated

.