treat CartesianIndex as a broadcast scalar

NEWS.md

@@ -33,6 +33,7 @@ New library features
 * The `initialized=true` keyword assignment for `sortperm!` and `partialsortperm!`
   is now a no-op ([#47979]). It previously exposed unsafe behavior ([#47977]).
 * `binomial(x, k)` now supports non-integer `x` ([#48124]).
+* A `CartesianIndex` is now treated as a "scalar" for broadcasting ([#47044]).
 
 Standard library changes
 ------------------------

base/broadcast.jl

@@ -724,7 +724,7 @@ julia> Broadcast.broadcastable("hello") # Strings break convention of matching i
 Base.RefValue{String}("hello")
 ```
 """
-broadcastable(x::Union{Symbol,AbstractString,Function,UndefInitializer,Nothing,RoundingMode,Missing,Val,Ptr,AbstractPattern,Pair,IO}) = Ref(x)
+broadcastable(x::Union{Symbol,AbstractString,Function,UndefInitializer,Nothing,RoundingMode,Missing,Val,Ptr,AbstractPattern,Pair,IO,CartesianIndex}) = Ref(x)
 broadcastable(::Type{T}) where {T} = Ref{Type{T}}(T)
 broadcastable(x::Union{AbstractArray,Number,AbstractChar,Ref,Tuple,Broadcasted}) = x
 # Default to collecting iterables — which will error for non-iterables

base/multidimensional.jl

@@ -32,6 +32,10 @@ module IteratorsMD
     A `CartesianIndex` is sometimes produced by [`eachindex`](@ref), and
     always when iterating with an explicit [`CartesianIndices`](@ref).
 
+    An `I::CartesianIndex` is treated as a "scalar" (not a container)
+    for `broadcast`.   In order to iterate over the components of a
+    `CartesianIndex`, convert it to a tuple with `Tuple(I)`.
+
     # Examples
     ```jldoctest
     julia> A = reshape(Vector(1:16), (2, 2, 2, 2))
@@ -61,6 +65,10 @@ module IteratorsMD
     julia> A[CartesianIndex((1, 1, 2, 1))]
     5
     ```
+
+    !!! compat "Julia 1.10"
+        Using a `CartesianIndex` as a "scalar" for `broadcast` requires
+        Julia 1.10; in previous releases, use `Ref(I)`.
     """
     struct CartesianIndex{N} <: AbstractCartesianIndex{N}
         I::NTuple{N,Int}

test/abstractarray.jl

@@ -993,9 +993,9 @@ end
     end
 
     i = CartesianIndex(17,-2)
-    @test CR .+ i === i .+ CR === CartesianIndices((19:21, -1:3))
-    @test CR .- i === CartesianIndices((-15:-13, 3:7))
-    @test collect(i .- CR) == Ref(i) .- collect(CR)
+    @test CR .+ i === i .+ CR === CartesianIndices((19:21, -1:3)) == collect(CR) .+ i
+    @test CR .- i === CartesianIndices((-15:-13, 3:7)) == collect(CR) .- i
+    @test collect(i .- CR) == Ref(i) .- collect(CR) == i .- collect(CR)
 end
 
 @testset "issue #25770" begin

test/broadcast.jl

@@ -1124,6 +1124,11 @@ end
     @test count(A) == count(>=(0), pos)
 end
 
+@testset "issue #38432: make CartesianIndex a broadcast scalar" begin
+    @test CartesianIndex(1,2) .+ (CartesianIndex(3,4), CartesianIndex(5,6)) == (CartesianIndex(4, 6), CartesianIndex(6, 8))
+    @test CartesianIndex(1,2) .+ [CartesianIndex(3,4), CartesianIndex(5,6)] == [CartesianIndex(4, 6), CartesianIndex(6, 8)]
+end
+
 # test that `Broadcast` definition is defined as total and eligible for concrete evaluation
 import Base.Broadcast: BroadcastStyle, DefaultArrayStyle
 @test Base.infer_effects(BroadcastStyle, (DefaultArrayStyle{1},DefaultArrayStyle{2},)) |>