sparse constructor misbehaving on master

[gragusa]

In v.0.3.10

julia> sparse([], [], [1], 0, 0, vcat)
0x0 sparse matrix with 0 Int64 entries:

In 0.4.0-dev+6402

julia> sparse([], [], [1], 0, 0, vcat)
ERROR: MethodError: `sparse` has no method matching sparse(::Array{Any,1}, ::Array{Any,1}, ::Array{Int64,1}, ::Int64, ::Int64, ::Function)
Closest candidates are:
  sparse(::Any, ::Any, ::AbstractArray{T,1}, ::Any, ::Any)
  sparse(::Any, ::Any, ::Number, ::Any, ::Any, ::Union{Base.Func{N},Function})
  sparse(::Any, ::Any, ::AbstractArray{Bool,1}, ::Any, ::Any)
  ...

[jiahao]

This is yet another manifestation of [] being None[] on 0.3 and Any[] on 0.4, similar to #12271

[gragusa]

Yes, indeed:

On 0.4.0-dev+6402

julia> sparse(Int[], Int[], [1], 0, 0, vcat)
0x0 sparse matrix with 0 Int64 entries:```

[jakebolewski]

This is because Vector{None} <: Vector{Integer}. The <: Integer requirement will likely be relaxed in v0.5.

See also #12567

[Sacha0]

What should happen here?

That sparse([], [], [1], 0, 0, vcat) returns a MethodError seems appropriate, Any not being a subtype of Integer. Additionally, were the Any[]s passed in promoted to T[] where T<:Integer, shouldn't that call still fail as sparse's first three arguments (I,J,V) must have equal length when they are vectors? Best!

[tkelman]

Yeah I think 0.3's sparse may have been a bit too permissive. Closing, don't think there's a bug here any more.