Kernel size calculation for sliced field halo filling seems to be wrong

[glwagner]

To fill the halos of sliced fields we call fill_halo_size:

Oceananigans.jl/src/BoundaryConditions/fill_halo_regions.jl

Lines 335 to 337 in 6b22a79

	@inline fill_halo_size(::OffsetArray, ::WEB, ::Tuple{<:Any, <:Colon, <:Colon}, args...) = :yz
	@inline fill_halo_size(::OffsetArray, ::SNB, ::Tuple{<:Colon, <:Any, <:Colon}, args...) = :xz
	@inline fill_halo_size(::OffsetArray, ::TBB, ::Tuple{<:Colon, <:Colon, <:Any}, args...) = :xy

However, that code incorrectly assumes that an indices of : means that the whole dimension should be filled. This is wrong for fields with Nothing locations, which have only 1 point in the Nothing direction, but can still have indices :.

There is also logic for Nothing locations:

Oceananigans.jl/src/BoundaryConditions/fill_halo_regions.jl

Lines 339 to 352 in 6b22a79

	# If the index is a Colon and the location is _NOT_ a `Nothing` (i.e. not a `ReducedField`),
	# then fill the whole boundary, otherwise fill the size of the corresponding array
	@inline whole_halo(idx, loc) = false
	@inline whole_halo(idx, ::Nothing) = false
	@inline whole_halo(::Colon, ::Nothing) = false
	@inline whole_halo(::Colon, loc) = true
	# Calculate kernel size
	@inline fill_halo_size(c::OffsetArray, ::WEB, idx, bc, loc, grid) =
	@inbounds (ifelse(whole_halo(idx[2], loc[2]), size(grid, 2), size(c, 2)), ifelse(whole_halo(idx[3], loc[3]), size(grid, 3), size(c, 3)))
	@inline fill_halo_size(c::OffsetArray, ::SNB, idx, bc, loc, grid) =
	@inbounds (ifelse(whole_halo(idx[1], loc[1]), size(grid, 1), size(c, 1)), ifelse(whole_halo(idx[3], loc[3]), size(grid, 3), size(c, 3)))
	@inline fill_halo_size(c::OffsetArray, ::TBB, idx, bc, loc, grid) =
	@inbounds (ifelse(whole_halo(idx[1], loc[1]), size(grid, 1), size(c, 1)), ifelse(whole_halo(idx[2], loc[2]), size(grid, 2), size(c, 2)))

But this is never used because of the incorrect assumption about the meaning of indices.

One possibility is to convert indices in Nothing directions to nothing. That would permit dispatching on the type of the indices like we're trying to do. Another possibility is to fix the logic here. Dispatch is annoying because loc is a tuple, but it can be done. A third possibility is just to avoid the symbol sizes :yz, xz and :xy. That last seems maybe like the best choice.

This bug would prevent anyone from using ImmersedBoundaryGrid. So I'm confused.

[glwagner]

@simone-silvestri