Use BarotropicModel instead of :barotropic etc

[milankl]

Same for ShallowWaterModel, PrimitiveEquationModel.

[milankl]

@maximilian-gelbrecht I thought it would be more consistent to use model::Type{<:ModelSetup} throughout, meaning that we do

run_speedy(model=ShallowWaterModel)

instead of model=:shallowwater. This might give us more scope in the future for externally defined ModelSetups and (most important ofc) it works with auto-completion 😉

[milankl]

Ah same problem as always, we cannot define Parameters without having defined BarotropicModel and vice versa 💩

[maximilian-gelbrecht]

But here we have an abstract type. Shouldn't this just work when defining abstract type ModelSetup{D} end e.g. directly in the SpeedyWeather.jl file (or somewhere else where it is loaded before all other files).

[milankl]

Yeah, but we then also would need a AbstractBarotropicModel... which can be defined before Parameters and used therein

[maximilian-gelbrecht]

Yeah, use in Parameters wouldn't be a problem, one could do the default argument also by calling a function at run time. But you are also using the model types in define_diffusion.jl which always needs to be called before defining the concrete model types.

Mh, ... so either introducing all the abstract types, or maybe we just keep it as it is?

Long-term, an alternative could also be to do something a bit more akin of the syntax of DiffEq.jl etc:

model = ShallowWaterModel(kwargs...)
p, d = solve(model)

Edit: when I think about it, this also wouldn't solve this issue, it's basically the same...

[maximilian-gelbrecht]

I'd assume that use of the model types here, would always require the added abstract types. This file needs to be called before the file that defines the models, as it needs HorizontalDiffusion

[milankl]

What about

abstract type ModelSetup{D} end
abstract type Barotropic{D} <: ModelSetup end
abstract type ShallowWater{D} <: ModelSetup end
abstract type PrimitiveEquation{D} <: ModelSetup end

and then subtype BarotropicModel etc from there?

[maximilian-gelbrecht]

My impression is that other libraries also use more parametric types in the struct because of this problem, and specify the types of the fields in the structs less often. This does definitely come as the cost of less readable code though

[maximilian-gelbrecht]

What about

abstract type ModelSetup{D} end
abstract type Barotropic{D} <: ModelSetup end
abstract type ShallowWater{D} <: ModelSetup end
abstract type PrimitiveEquation{D} <: ModelSetup end

and then subtype BarotropicModel etc from there?

Yeah, I guess that's the easiest way to do it.

[milankl]

Okay that's how it's now!

abstract type ModelSetup end
abstract type Barotropic <: ModelSetup end
abstract type ShallowWater <: ModelSetup end
abstract type PrimitiveEquation <: ModelSetup end

and

struct BarotropicModel <: Barotropic
struct ShallowWaterModel <: ShallowWater
struct PrimitiveEquationModel <: PrimitiveEquation

so without Model it's the abstract supertype that can be used early on, with Model it's the concrete type.