Possible breakage in 1.6.0-rc2 and nightly in evaluating CartesianIndices for OffsetArrays with BigInt parent ranges

[jishnub]

Unfortunately there's a bug on master in OffsetArrays currently, so this PR is necessary to see the issue:

on 1.5.4:

julia> using OffsetArrays

julia> a = OffsetArray(big(1):big(2), 0);

julia> CartesianIndices(a)
2-element CartesianIndices{1,Tuple{OffsetArrays.IdOffsetRange{Int64,Base.OneTo{Int64}}}} with indices 1:2:
 CartesianIndex(1,)
 CartesianIndex(2,)

on 1.6.0-rc2 and nightly:

julia> CartesianIndices(a)
ERROR: MethodError: no method matching OrdinalRange{Int64, Int64}(::OffsetArrays.IdOffsetRange{BigInt, Base.OneTo{BigInt}})
Closest candidates are:
  OrdinalRange{T1, T2}(::AbstractUnitRange{T1}) where {T1, T2<:Integer} at range.jl:1133
  OrdinalRange{T1, T2}(::StepRange) where {T1, T2<:Integer} at range.jl:1132
  OrdinalRange{T1, T2}(::UnitRange) where {T1, T2<:Integer} at range.jl:1134
  ...
Stacktrace:
 [1] convert(#unused#::Type{OrdinalRange{Int64, Int64}}, r::OffsetArrays.IdOffsetRange{BigInt, Base.OneTo{BigInt}})
   @ Base ./range.jl:215
 [2] (::Base.IteratorsMD.var"#7#8")(r::OffsetArrays.IdOffsetRange{BigInt, Base.OneTo{BigInt}})
   @ Base.IteratorsMD ./multidimensional.jl:271
 [3] map
   @ ./tuple.jl:213 [inlined]
 [4] CartesianIndices(inds::Tuple{OffsetArrays.IdOffsetRange{BigInt, Base.OneTo{BigInt}}})
   @ Base.IteratorsMD ./multidimensional.jl:271
 [5] CartesianIndices(A::OffsetVector{BigInt, UnitRange{BigInt}})
   @ Base.IteratorsMD ./multidimensional.jl:279
 [6] top-level scope
   @ REPL[4]:1

julia> VERSION
v"1.7.0-DEV.707"

This is due to a missing method for the IdOffsetRange axis type defined in OffsetArrays, but this conversion is not used in 1.5.4. The conversion carried out by CartesianIndices appears to have changed from an AbstractUnitRange{Int} in 1.5.4 to an OrdinalRange{Int,Int} in 1.6.

[jishnub]

Looks like this was introduced by #37829

[KristofferC]

Related to #39589?

[jishnub]

It's not unrelated as the same constructor is called eventually, but this seems to be a different issue (the other one is primarily due to how the integer is converted to a range). This issue can be fixed by adding a specialized CartesianIndices constructor for NTuple{N,AbstractUnitRange{<:Integer}}, so perhaps I will make a PR.

[laborg]

This works on 1.6.5, on 1.7.2 and on master - Is there any reason not to close this (and #40038)?

[KristofferC]

If it works on those releases, it seems indeed a good idea to close.

[vtjnash]

Whoops, I tested that PR and it is still needed. Left a comment there about how we could finish that.