Following this Discourse discussion, I'd like to propose to make CartesianIndexs broadcast like scalars. That is, since

julia> 1 .+ (2, 3)
(3, 4)

works, so should

julia> CartesianIndex(1,1) .+ (CartesianIndex(2,2), CartesianIndex(3,3))

(which currently errors, but should, in my opinion, result in (CartesianIndex(3, 3), CartesianIndex(4, 4)))