`repr` roundtripping. "print it like you build it"

[goretkin]

I'm not sure if there are any functions in the display/show/repr family that are supposed to obey it, but it would be nice if a_rt === a (to the extent that everything is bitstype) in the following

using StaticArrays

a = @SArray [(SA_F32[1,2,3]), SA_F32[2,3,4]]
@show typeof(a)
sa = repr(a)
@show sa
a_rt = eval(Meta.parse(sa))
@show typeof(a_rt)

output:

typeof(a) = SArray{Tuple{2},SArray{Tuple{3},Float32,1,3},1,2}
sa = "SArray{Tuple{3},Float32,1,3}[[1.0, 2.0, 3.0], [2.0, 3.0, 4.0]]"
typeof(a_rt) = Array{SArray{Tuple{3},Float32,1,3},1}

(related a bit to the conversation in #691.

[c42f]

Good issue, thanks! I certainly think we could try changing repr for SArray so that it parses back to being a static array using SA. Currently we have

julia> repr(SA[1,2,3])
"[1, 2, 3]"

julia> repr(SA[SA_F32[1,2,3], SA_F32[4,5,6]])
"SArray{Tuple{3},Float32,1,3}[[1.0, 2.0, 3.0], [4.0, 5.0, 6.0]]"

which is compact but not entirely ideal when printing nested data structures. SA is relatively compact so we could try to get

julia> repr(SA[1,2,3])
"SA[1, 2, 3]"

julia> repr(SA[SA_F32[1,2,3], SA_F32[4,5,6]])
"SA[SA_F32[1.0, 2.0, 3.0], SA_F32[4.0, 5.0, 6.0]]"

or some similar printing which generates the correct type and doesn't imply creating any Array temporaries when it's evaluated. For other StaticArray subtypes we could fall back to using conversion (eg, MArray(SA[1 2; 3 4])) I guess though that's not exactly pretty...

Would you be interested in giving it a go? There's several functions to think about (print / repr / show / display) to compare to how Base handles these things.

[tpapp]

Something as simple as

function Base.show(io::IO, a::SArray{S,T}) where {S,T}
    print(io, "SArray{$(S),$(T)}")
    print(io, Tuple(vec(a)))
end

could take care of this. Special casing SVector and SMatrix would make it nicer, and it generalizes easily for MArray.

I am happy to make a PR (with tests etc) if there is interest.