[RFC] Adding component iterator

[kimikage]

This is a PoC to clarify the discussion in #184, and is the modernized (Julia v1) version of #7. See also #119

using ColorTypes
import ColorTypes.ColorantN

struct ComponentIterator{C<:Colorant}
    c::C
end

function Base.getproperty(c::C, sym::Symbol) where C <: Colorant
   if sym === :components
      return ComponentIterator{C}(c)
   else # fallback to getfield
      return getfield(c, sym)
   end
end

Base.eltype(itr::ComponentIterator{C}) where {T, C <: Colorant{T}} = T
Base.length(itr::ComponentIterator{C}) where {N, C <: ColorantN{N}} = N

function Base.iterate(itr::ComponentIterator{C}, state::Int=0) where {N, C <: ColorantN{N}}
    state >= N && return nothing
    state += 1
    (itr[state], state)
end

function Base.getindex(itr::ComponentIterator{C}, i::Integer) where {N, C <: ColorantN{N}}
    comp_n = (comp1, comp2, comp3) # FIXME
    (comp_n[i])(itr.c)
end

julia> rgb_f32 = RGB{Float32}(1.0, 0.5, 0.0)
RGB{Float32}(1.0f0,0.5f0,0.0f0)

julia> eltype(rgb_f32.components)
Float32

julia> length(rgb_f32.components)
3

julia> for c in rgb_f32.components
           @show c
       end
c = 1.0f0
c = 0.5f0
c = 0.0f0

julia> rgb_f32.components[2]
0.5f0

julia> RGB((RGB(1.0, 0.5, 0.0).components .+ RGB(0.0, 0.1, 0.2).components)...)
RGB{Float64}(1.0,0.6,0.2)

The above has some problems in practical use, but IMO, it is better than eltypec and lengthc.

[kimikage]

It maight be a good idea to create a new package ColorComponentIterators.jl, as ColorVectorSpace.jl is a separate package.
In fact, such a simple iterator, mapc, and so on are not very useful (cf. #184 (comment)). More complex iterators may be needed.

[timholy]

With constant propagation this might be more useful than previously, and hiding it behind a c.components seems like a pretty clever design. Thanks for submitting this, I will have to think through the implications.

[kimikage]

The problem with providing the iteration interface as a function instead of a property is name collision.
The components is not a color-specific concept. At least, Distributions exports the components. (Of course, I don't think Distributions must have a monopoly on the function.)
https://github.com/JuliaStats/Distributions.jl/blob/a36c613f5434c76ef21d3ed2d4a80fab66068d19/src/Distributions.jl#L178

comps is short, and the risk of name clashes seems to be sufficiently low. It is not easy to understand, though.

[johnnychen94]

Base has fieldnames and fieldtypes for DataType, do you think fields a good name for the instance x::Colorant? We could possibly propose this name and the iterator implementation in Julia Base.

[kimikage]

Thanks for the suggestion.
However, as you say the "fields" is already used as fields of the structure; the difference between ARGB and RGBA or types like ARGB32 will bring confusion.

[kimikage]

The feature itself has been implemented in PR #260, but the design of the public API is still not settled.