CartesianIndices (breaking?) change in 1.6.0-rc1

[akosuas]

Something has changed here:

Julia Version 1.5.3
Commit 788b2c77c1 (2020-11-09 13:37 UTC)
Platform Info:
  OS: Linux (x86_64-pc-linux-gnu)
  LLVM: libLLVM-9.0.1 (ORCJIT, skylake-avx512)
Environment:
  JULIA_PKG_SERVER = pkg.julialang.org
julia> CartesianIndices((3, 1:3))
1×3 CartesianIndices{2,Tuple{UnitRange{Int64},UnitRange{Int64}}}:
 CartesianIndex(3, 1)  CartesianIndex(3, 2)  CartesianIndex(3, 3)

julia> versioninfo()
Julia Version 1.6.0-rc1
Commit a58bdd9010 (2021-02-06 15:49 UTC)
Platform Info:
  OS: Linux (x86_64-pc-linux-gnu)
  LLVM: libLLVM-11.0.1 (ORCJIT, skylake-avx512)
Environment:
  JULIA_PKG_SERVER = pkg.julialang.org
julia> CartesianIndices((3, 1:3,))
3×3 CartesianIndices{2, Tuple{Base.OneTo{Int64}, UnitRange{Int64}}}:
 CartesianIndex(1, 1)  CartesianIndex(1, 2)  CartesianIndex(1, 3)
 CartesianIndex(2, 1)  CartesianIndex(2, 2)  CartesianIndex(2, 3)
 CartesianIndex(3, 1)  CartesianIndex(3, 2)  CartesianIndex(3, 3)
julia> versioninfo()

Is this expected? It is breaking some of my code, and I can't find anything obvious in NEWS that describes this.

[akosuas]

Reading a bit more closely, looks likely to be caused by #37829. I wonder if my single-integer usage of CartesianIndices was "supported" in the first place?

[simeonschaub]

What code is this breaking? CartesianIndex((Base.OneTo(3), 1:3)) does seem more correct to me here than CartesianIndex((1:3, 1:3)). It should behave exactly the same, but IIUC this change was necessary to improve performance of CartesianIndices. Your code might need to be made a little more generic, but since OneTo is an AbstractUnitRange, this should be quite straightforward.

Edit: Sorry, I misread your example, the two behaviors are definitely inconsistent.

[JeffBezanson]

The 1.5 behavior seems to have been inconsistent:

julia> CartesianIndices((2,3))
2×3 CartesianIndices{2,Tuple{Base.OneTo{Int64},Base.OneTo{Int64}}}:
 CartesianIndex(1, 1)  CartesianIndex(1, 2)  CartesianIndex(1, 3)
 CartesianIndex(2, 1)  CartesianIndex(2, 2)  CartesianIndex(2, 3)

julia> CartesianIndices((2,1:3))
1×3 CartesianIndices{2,Tuple{UnitRange{Int64},UnitRange{Int64}}}:
 CartesianIndex(2, 1)  CartesianIndex(2, 2)  CartesianIndex(2, 3)

That can't be right, but I'm not sure what was intended.

[akosuas]

Thanks for investigating guys.

If I was relying on some dodgy / broken behaviour in 1.5, I'm happy to switch to an explicit one-element range, i.e. CartesianIndices((3:3, 1:3)) instead of CartesianIndices((3, 1:3).

[mbauman]

Yes, that behavior was definitely unintentional and buggy. Integers should always describe 1-based lengths; ranges should always specify indices. Sorry this caught you out — perhaps we should indeed add a NEWS item.

[johnnychen94]

FYI, I made that change intentionally in #37829 (comment), and also to LinearIndices in #37928.

[akosuas]

FYI, I made that change intentionally in #37829 (comment), and also to LinearIndices in #37928.

Ah, nicely done, I hadn't spotted that.

[sostock]

I have added a NEWS/HISTORY item for the change: #39784

[johnnychen94]

Close since this is already solved: this change is considered a bug fix with associated NEWS entry.