Zero + StaticVector is ambiguous

[cstjean]

On master,

using StaticArrays, FillArrays
SVector(2,3) + Zeros(2)

@willtebbutt

[dlfivefifty]

Here's the full error:

julia> SVector(2,3) + Zeros(2)
ERROR: MethodError: +(::StaticArrays.SArray{Tuple{2},Int64,1,2}, ::FillArrays.Zeros{Float64,1,Tuple{Int64}}) is ambiguous. Candidates:
  +(a::AbstractArray{T,N}, b::FillArrays.Zeros{V,N,SZ} where SZ) where {T, V, N} in FillArrays at /Users/sheehanolver/.julia/v0.6/FillArrays/src/FillArrays.jl:278
  +(a::StaticArrays.StaticArray, b::AbstractArray) in StaticArrays at /Users/sheehanolver/.julia/v0.6/StaticArrays/src/linalg.jl:12
Possible fix, define
  +(::StaticArrays.StaticArray{S,T,N} where S<:Tuple, ::FillArrays.Zeros{V,N,SZ} where SZ)

[dlfivefifty]

Options:

1. StaticArrays depends on FillArrays
2. FillArrays depends on StaticArrays
3. FillArrays only overrides concrete types, e.g. change to +(a::Array{T, N}, b::Zeros{V, N}) where {T, V, N}

I think we expect this to return an SVector and therefore be allocation-free.

[dlfivefifty]

I'm going to do (3) for now and we can come back to this later.

[willtebbutt]

Ah, yeah. I agree regarding possible solutions. 1 and 2 are only partial fixes though, as we'll keep running into this ambiguity with every other AbstractArray subtype that defines its own notion of addition with other AbstractArrays. I think option 3 might therefore be the best option -- it's a bit limiting but makes it quite unlikely that we'll encounter this ambiguity.

[dlfivefifty]

This is a case where + in Base should be rewritten to use traits I think. A la JuliaLang/julia#25558

[cstjean]

FWIW, the StaticArrays method for addition is quite good already...

julia> @btime $(SVector(2,3)) + $(Zeros(2))
  5.028 ns (0 allocations: 0 bytes)
2-element StaticArrays.SArray{Tuple{2},Float64,1,2}:
 2.0
 3.0

julia> 

julia> @btime identity($(SVector(2,3)))
  2.025 ns (0 allocations: 0 bytes)
2-element StaticArrays.SArray{Tuple{2},Int64,1,2}:
 2
 3