Oceananigans.jl: Why is `w*T` is a `MultiaryOperation`?

[ali-ramadhan]

I'd expect it to be a BinaryOperation?

using Oceananigans, Oceananigans.Grids,Oceananigans.Diagnostics, Oceananigans.AbstractOperations
grid = RegularCartesianGrid(size=(16, 16, 16), extent=(1, 1, 1))
model = IncompressibleModel(grid=grid)
u, v, w = model.velocities
T, S = model.tracers
wT = w*T
````

MultiaryOperation at (Cell, Cell, Face)
├── grid: RegularCartesianGrid{Float64,Periodic,Periodic,Bounded,OffsetArrays.OffsetArray{Float64,1,StepRangeLen{Float64,Base.TwicePrecision{Float64},Base.TwicePrecision{Float64}}}}
│ ├── size: (16, 16, 16)
│ └── domain: x ∈ [0.0, 1.0], y ∈ [0.0, 1.0], z ∈ [-1.0, 0.0625]
└── tree:

• at (Cell, Cell, Face)
  ├── OffsetArray{Float64, 3, Array{Float64,3}}
  └── OffsetArray{Float64, 3, Array{Float64,3}}
  ```

[ali-ramadhan]

Seems that when the fields are at different locations you get a MultiaryOperation and when it's the same location you get a BinaryOperation:

julia> u*T
MultiaryOperation at (Face, Cell, Cell)
├── grid: RegularCartesianGrid{Float64,Periodic,Periodic,Bounded,OffsetArrays.OffsetArray{Float64,1,StepRangeLen{Float64,Base.TwicePrecision{Float64},Base.TwicePrecision{Float64}}}}
│   ├── size: (16, 16, 16)
│   └── domain: x ∈ [0.0, 1.0], y ∈ [0.0, 1.0], z ∈ [-1.0, 0.0625]
└── tree: 

* at (Face, Cell, Cell)
├── OffsetArray{Float64, 3, Array{Float64,3}}
└── OffsetArray{Float64, 3, Array{Float64,3}}

julia> u*v
MultiaryOperation at (Face, Cell, Cell)
├── grid: RegularCartesianGrid{Float64,Periodic,Periodic,Bounded,OffsetArrays.OffsetArray{Float64,1,StepRangeLen{Float64,Base.TwicePrecision{Float64},Base.TwicePrecision{Float64}}}}
│   ├── size: (16, 16, 16)
│   └── domain: x ∈ [0.0, 1.0], y ∈ [0.0, 1.0], z ∈ [-1.0, 0.0625]
└── tree: 

* at (Face, Cell, Cell)
├── OffsetArray{Float64, 3, Array{Float64,3}}
└── OffsetArray{Float64, 3, Array{Float64,3}}

julia> T*S
BinaryOperation at (Cell, Cell, Cell)
├── grid: RegularCartesianGrid{Float64,Periodic,Periodic,Bounded,OffsetArrays.OffsetArray{Float64,1,StepRangeLen{Float64,Base.TwicePrecision{Float64},Base.TwicePrecision{Float64}}}}
│   ├── size: (16, 16, 16)
│   └── domain: x ∈ [0.0, 1.0], y ∈ [0.0, 1.0], z ∈ [-1.0, 0.0625]
└── tree: 

* at (Cell, Cell, Cell) via identity
├── OffsetArray{Float64, 3, Array{Float64,3}}
└── OffsetArray{Float64, 3, Array{Float64,3}}

julia> v*v
BinaryOperation at (Cell, Face, Cell)
├── grid: RegularCartesianGrid{Float64,Periodic,Periodic,Bounded,OffsetArrays.OffsetArray{Float64,1,StepRangeLen{Float64,Base.TwicePrecision{Float64},Base.TwicePrecision{Float64}}}}
│   ├── size: (16, 16, 16)
│   └── domain: x ∈ [0.0, 1.0], y ∈ [0.0, 1.0], z ∈ [-1.0, 0.0625]
└── tree: 

* at (Cell, Face, Cell) via identity
├── OffsetArray{Float64, 3, Array{Float64,3}}
└── OffsetArray{Float64, 3, Array{Float64,3}}

[glwagner]

Possibly we need to change the definition of MultiaryOperation to a, b, c, d.... I guess there can be Varargs of length zero?

Relevant lines are

https://github.com/CliMA/Oceananigans.jl/blob/f13a0a26df78cd1c1495d56ef5e7ea2fbbbfb23c/src/AbstractOperations/multiary_operations.jl#L25-L47

[glwagner]

If two fields are at the same location, that scenario may possibly get shortcut by this method:

https://github.com/CliMA/Oceananigans.jl/blob/f13a0a26df78cd1c1495d56ef5e7ea2fbbbfb23c/src/AbstractOperations/binary_operations.jl#L68

[ali-ramadhan]

I guess there can be Varargs of length zero?

Yup looks like it

julia> f(s, x...) = @show x
f (generic function with 1 method)

julia> f(π);
x = ()

julia> f(π, "∫", "∂");
x = ("∫", "∂")

[glwagner]

That should be an easy fix. Not sure, but BinaryOperation might be preferred over mulitiary because it avoids the ntuple magic unrolling.