Unable to index with non-concrete Array{CartesianIndex}

[phipsgabler]

I noticed a weird error when using abstractly typed arrays of CartesianIndex:

julia> graph = rand(3,3)
3×3 Array{Float64,2}:
 0.596119  0.509642  0.0507867
 0.573233  0.895749  0.972909
 0.410628  0.37905   0.0779637

julia> graph[[CartesianIndex(1,1)]]
1-element Array{Float64,1}:
 0.5961186454040548

julia> graph[CartesianIndex[CartesianIndex(1,1)]]
ERROR: ArgumentError: unable to check bounds for indices of type CartesianIndex{2}
Stacktrace:
 [1] checkindex(::Type{Bool}, ::Base.OneTo{Int64}, ::CartesianIndex{2}) at ./abstractarray.jl:561
 [2] checkindex at ./abstractarray.jl:576 [inlined]
 [3] checkbounds_indices at ./abstractarray.jl:532 [inlined]
 [4] checkbounds at ./multidimensional.jl:544 [inlined]
 [5] checkbounds at ./abstractarray.jl:506 [inlined]
 [6] _getindex at ./multidimensional.jl:742 [inlined]
 [7] getindex(::Array{Float64,2}, ::Array{CartesianIndex,1}) at ./abstractarray.jl:1060
 [8] top-level scope at REPL[155]:1

julia> graph[CartesianIndex{2}[CartesianIndex(1,1)]]
1-element Array{Float64,1}:
 0.5961186454040548

julia> versioninfo()
Julia Version 1.5.4
Commit 69fcb5745b (2021-03-11 19:13 UTC)
Platform Info:
  OS: Linux (x86_64-pc-linux-gnu)
  CPU: Intel(R) Core(TM) i7-8565U CPU @ 1.80GHz
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-9.0.1 (ORCJIT, skylake)

This has previously been reported on Discourse: https://discourse.julialang.org/t/surprising-type-error/51943.

[mbauman]

I fear there's not much we can do here but potentially add this to the https://docs.julialang.org/en/v1/manual/arrays/#Cartesian-indices warning and/or improve the error message. The bounds checking infrastructure requires knowing which dimension(s) a particular index indexes into. We've talked some about it here: #35681 (comment)... and I thought there was a more direct issue about this, but I can't find it.

It's kinda like logical indexing with arrays of Bool, but there we explicitly disallow indexing by scalar Bool to help separate its special functionality... and of course that's not something we can do with scalar CartesianIndex.