ArgumentError: Trailing indices must be 1 when using v[bitarray] .= scalar

[Alexander-Barth]

With DiskArrays v0.4.15 and NCDatasets (47afa1c) I see the following regression:

using NCDatasets
fname = tempname()
ds = NCDataset(fname, "c")
v = defVar(ds, "temperature", zeros(10,11), ("lon", "lat"))
v[rand(10,11) .> 0.5] .= 1
(full error message is below)

@lupemba @felixcremer @rafaqz Do you have any idea what could be the problem?
So far, I am not able to reproduce the issue independently of NCDatasets.

julia> include("/home/abarth/.julia/dev/NCDatasets/test/test_diskarray.jl");
ERROR: LoadError: ArgumentError: Trailing indices must be 1
Stacktrace:
  [1] _resolve_indices(::Tuple{}, i::Tuple{…}, indices_pre::DiskArrays.DiskIndex{…}, strategy::DiskArrays.NoBatch{…})
    @ DiskArrays ~/.julia/packages/DiskArrays/WgDSb/src/diskindex.jl:101
  [2] _resolve_indices (repeats 2 times)
    @ ~/.julia/packages/DiskArrays/WgDSb/src/diskindex.jl:57 [inlined]
  [3] DiskArrays.DiskIndex(a::CommonDataModel.CFVariable{…}, i::NTuple{…}, batch_strategy::DiskArrays.NoBatch{…})
    @ DiskArrays ~/.julia/packages/DiskArrays/WgDSb/src/diskindex.jl:43
  [4] setindex_disk_nobatch!(a::CommonDataModel.CFVariable{…}, values::Array{…}, i::NTuple{…})
    @ DiskArrays ~/.julia/packages/DiskArrays/WgDSb/src/indexing.jl:173
  [5] setindex_disk!(::CommonDataModel.CFVariable{…}, ::Array{…}, ::UnitRange{…}, ::Vararg{…})
    @ DiskArrays ~/.julia/packages/DiskArrays/WgDSb/src/indexing.jl:137
  [6] writeblock!(::CommonDataModel.SubVariable{…}, ::Array{…}, ::UnitRange{…}, ::UnitRange{…}, ::Vararg{…})
    @ DiskArrays ~/.julia/packages/DiskArrays/WgDSb/src/subarray.jl:38
  [7] writeblock_checked!
    @ ~/.julia/packages/DiskArrays/WgDSb/src/indexing.jl:301 [inlined]
  [8] setindex_disk_nobatch!(a::CommonDataModel.SubVariable{…}, values::Array{…}, i::NTuple{…})
    @ DiskArrays ~/.julia/packages/DiskArrays/WgDSb/src/indexing.jl:182
  [9] setindex_disk!
    @ ~/.julia/packages/DiskArrays/WgDSb/src/indexing.jl:137 [inlined]
 [10] setindex!
    @ ~/.julia/packages/DiskArrays/WgDSb/src/indexing.jl:324 [inlined]
 [11] #208
    @ ~/.julia/packages/DiskArrays/WgDSb/src/broadcast.jl:165 [inlined]
 [12] foreach(f::DiskArrays.var"#208#220"{CommonDataModel.SubVariable{…}, Int64}, itr::DiskArrays.GridChunks{4, NTuple{…}})
    @ Base ./abstractarray.jl:3187
 [13] fill!(dest::CommonDataModel.SubVariable{Float64, 2, CommonDataModel.CFVariable{…}, Tuple{…}, false}, value::Int64)
    @ DiskArrays ~/.julia/packages/DiskArrays/WgDSb/src/broadcast.jl:163
 [14] copyto!
    @ ./broadcast.jl:933 [inlined]
 [15] materialize!
    @ ./broadcast.jl:883 [inlined]
 [16] materialize!(dest::CommonDataModel.SubVariable{…}, bc::Base.Broadcast.Broadcasted{…})
    @ Base.Broadcast ./broadcast.jl:880
 [17] top-level scope
    @ ~/.julia/dev/NCDatasets/test/test_diskarray.jl:23

[Alexander-Barth]

I would not expect that the chunks become 4D for a 2D variable

using NCDatasets, DiskArrays
fname = tempname()
ds = NCDataset(fname, "c")
v = defVar(ds, "temperature", zeros(10,11), ("lon", "lat"))
m = rand(10,11) .> 0.5
DiskArrays.eachchunk(view(v,m))
# 1×1×1×1 DiskArrays.GridChunks{4, NTuple{4, DiskArrays.RegularChunks}}:
# [:, :, 1, 1] =
# (1:10, 1:11, 1:10, 1:11)

Also the size is 4D

size(view(v,m))
# (10,11,10,11)

At this line, i is a Tuple{BitMatrix} but it becomes a i2 with the type of Tuple{BitMatrix, BitMatrix}

https://github.com/JuliaGeo/CommonDataModel.jl/blob/6a800752909f78b3d7463a2c71edfd432fb22f8d/src/subvariable.jl#L95

[lupemba]

Looks like I didn't get everything updated correctly after JuliaIO/DiskArrays.jl#260

The Base.view method in CommonDataModel.jl should probably be something like this (Based on DiskArray.@implement_subarray):

function Base.view(a::AbstractVariable,i...) 
    disk_sub_array = DiskArrays.view_disk(a, i...)
    return SubVariable(DiskArrays.subarray(disk_sub_array))
end

Base.vec(a::AbstractVariable) = view(a, :)

Note that the code would be cleaner if the DiskArrays.view_disk just returned the SubArray

I just tested the change locally and I get

v[m] .= 1 # works
DiskArrays.eachchunk(view(v,m)) 
#1-element DiskArrays.GridChunks{1, Tuple{DiskArrays.RegularChunks}}:
#(1:54,)
size(view(v,m))
#(54,)

I will make a real PR later this week.