similar(::ChannelView, S, dims) has ambiguous intent

[timholy]

As reported on gitter:

julia> a = rand(RGB4{N0f8}, 100, 100);

julia> x = channelview(a);

julia> x[1,:,:]
ERROR: DimensionMismatch("new array has 100 color channels, must have 3")
Stacktrace:
 [1] check_ncolorchan(::Array{ColorTypes.RGB4{FixedPointNumbers.Normed{UInt8,8}},2}, ::Tuple{Int64,Int64}) at /home/tim/.julia/v0.6/ImageCore/src/colorchannels.jl:356
 [2] similar at /home/tim/.julia/v0.6/ImageCore/src/colorchannels.jl:117 [inlined]
 [3] similar(::ImageCore.ChannelView{FixedPointNumbers.Normed{UInt8,8},3,Array{ColorTypes.RGB4{FixedPointNumbers.Normed{UInt8,8}},2}}, ::Tuple{Base.OneTo{Int64},Base.OneTo{Int64}}) at ./abstractarray.jl:521
 [4] macro expansion at ./multidimensional.jl:458 [inlined]
 [5] _unsafe_getindex(::IndexCartesian, ::ImageCore.ChannelView{FixedPointNumbers.Normed{UInt8,8},3,Array{ColorTypes.RGB4{FixedPointNumbers.Normed{UInt8,8}},2}}, ::Int64, ::Base.Slice{Base.OneTo{Int64}}, ::Base.Slice{Base.OneTo{Int64}}) at ./multidimensional.jl:453
 [6] macro expansion at ./multidimensional.jl:442 [inlined]
 [7] _getindex at ./multidimensional.jl:438 [inlined]
 [8] getindex(::ImageCore.ChannelView{FixedPointNumbers.Normed{UInt8,8},3,Array{ColorTypes.RGB4{FixedPointNumbers.Normed{UInt8,8}},2}}, ::Int64, ::Colon, ::Colon) at ./abstractarray.jl:882

This is actually far nastier than it might seem. The offending method is here. The problem is that method is presuming that what you're doing is trying to create another ChannelView, and so it's making sure that the new underlying array has a shape consistent with the first dimension being a color channel. But unfortunately similar gets used in two different ways.

I haven't spent much time thinking about it (deadlines...), but this doesn't seem to be entirely trivial to fix while preserving type-stability.

[timholy]

Note it doesn't happen with RGB because channelview returns an Array in such cases.

[timholy]

Now that we use ReinterpretArray, this is no longer an issue.