`XESMF.regrid!` conflicts with `Oceananigans.Fields.regrid!`

[navidcy]

After #14 I think there might be an issue here because now both Oceananigans and XESMF define a regrid! method. I think ideally one should extends the other's method otherwise we have situations like:

julia> using Oceananigans
[ Info: Oceananigans will use 12 threads

julia> methods(regrid!)
# 2 methods for generic function "regrid!" from Oceananigans.Fields:
 [1] regrid!(a, target_grid, source_grid, b)
     @ ~/Library/CloudStorage/OneDrive-TheUniversityofMelbourne/Documents/Research/Oceananigans.jl-v4/src/Fields/regridding_fields.jl:112
 [2] regrid!(dst_field, src_field)
     @ ~/Library/CloudStorage/OneDrive-TheUniversityofMelbourne/Documents/Research/Oceananigans.jl-v4/src/Fields/regridding_fields.jl:48

julia> using XESMF: regrid!
[ Info: Precompiling OceananigansXESMFExt [94839fb3-6fd9-5496-8fe2-949f90a07c42]
WARNING: ignoring conflicting import of XESMF.regrid! into OceananigansXESMFExt
WARNING: ignoring conflicting import of XESMF.regrid! into Main

How should we proceed. I don't think we want XESMF to depend on Oceananigans. Should we drop the generic regrid! method and leave package extensions to define their own?

cc @simone-silvestri

[navidcy]

Shall we call it regrid_vector! then?

7fe8a3b#commitcomment-166560569

[navidcy]

Not ideal. But can't think of a way to proceed without a different name.

[simone-silvestri]

we could also think about applying the regrid! as a functor of the regrid method. Something like

regridder = Regridder(dst, src)
regridder(dst, src)

which in Oceananigans would become

function regrid!(dst::AbstractField, regridder::Regridder, scr::AbstractField)
      for k in 1:size(dst.grid, 3)
          regridder(vec(interior(dst, :, :, k), vec(interior(src, :, :, k)
      end
end

[simone-silvestri]

This would require something like

# Generic regridding function that does not check the dimensions of the `src` and
# `dst` arrays
function (regrider::Regridder)(dst::AbstractVector, src::AbstractVector)
    regridder.src_temp .= src
    LinearAlgebra.mul!(regridder.dst_temp, regridder.weights, regridder.src_temp)
    dst .= regridder.dst_temp
    return dst
end

[simone-silvestri]

We can also specialize for dense vectors where we can use the multiplication directly

(regrider::Regridder)(dst::DenseArray{T, 1}, src::DenseArray{T, 1}) where T = LinearAlegebra.mul!(dst, regrider.weights, src)

[simone-silvestri]

I ll open a quick PR.

[navidcy]

Yeah