Should grid be a positional argument to model constructors?

[glwagner]

The argument grid is required --- of course --- in model constructors.

Our current API makes grid a keyword argument. Originally (and it was a long time ago now) it was argued that keyword arguments are just better and that's why we should keep it. Also, we don't dispatch on it.

However, we violate that concept in Simulation, where the model is a positional argument.

I think it's more natural to input grid as a positional argument in the models. It's a relatively minor thing, but it avoids anti-patterns like grid=grid, which appeared in many scripts before ; grid was possible. It looks better.

This has come up because we are adding more Oceananigans-based AbstractModels over in ClimaSeaIce, and my first intuition there was to make grid a required positional argument. Ultimately though, we should strive for all models to have uniform interfaces, so either we change Oceananigans or we change ClimaSeaIce.

[navidcy]

I like kwarg because it’s more verbose.
Is there any other reason why it should be positional?
As you said grid=grid can be avoided although some people like it.

[glwagner]

Hmm in that case, I can consider a grid keyword argument for other models too.

[glwagner]

Is there any other reason why it should be positional?

I guess I'm just arguing that it's not consistent with most of the API, eg we write CenterField(grid).

I find the position representation of required arguments to be a natural and concise interface but perhaps I shouldn't die on that hill