Storing tuples in StaticArrays

[krcools]

I hit the following inconsistent border case in using StaticArrays/FixedSizeArrays:

julia> using StaticArrays
julia> using FixedSizeArrays
julia> @assert eltype(Vec((1,"a"),(2,"b"))) == Tuple{Int,String}
julia> @assert eltype(Vec((1,"a"))) == Tuple{Int,String}
ERROR: AssertionError: eltype(Vec((1,"a"))) == Tuple{Int,String}
julia> @assert eltype(SVector((1,"a"),(2,"b"))) == Tuple{Int,String}
julia> @assert eltype(SVector((1,"a"))) == Tuple{Int,String}
ERROR: AssertionError: eltype(SVector((1,"a"))) == Tuple{Int,String}
julia>

The problem is that the constructors for both types of static arrays flatten the tuple argument. I understand why, given that the data is stored as a tuple inside those data structures. Still I think given the goal for this package this should be considered a bug.

Btw, for StaticArrays the macro seems not to share this issue:

julia> @SVector [(1,"a")]
1-element StaticArrays.SVector{1,Tuple{Int64,String}}:
 (1,"a")

julia> @fsa [(1,"a")]
Vec(1,a)

My apologies if posting this issue to both projects is annoying.

[andyferris]

Yes, I knew about this. The "real" constructor accepts a tuple of elements, so a single-element SVector of Tuple{Int,String} would be called by SVector(((1,"a"),)).

I provide (as does Simon) a convenience constructor along the lines of SVector(x...) = SVector(x). This makes a single-element array of Tuple{Int,String} ambiguous with a multi-element array of promote_type(Int, String).

I need to revisit all the constructors once Jeff's improved subtyping algorithm lands, since I am working with the quirks of the current system which are changing. Previously, getting anything working was very, very difficult. Perhaps a bit more logic can be applied in the new system (e.g. if promote_type isn't concrete then perhaps the user wanted an array of tuples...).

[c42f]

This is basically the same as issue #518 — closing in favor of that issue which has much more discussion attached.