NTuple as a type parameter is not invariant

[amitmurthy]

This used to work, doesn't now.

  | | |_| | | | (_| |  |  Version 0.3.0-prerelease+2657 (2014-04-18 03:17 UTC)
 _/ |\__'_|_|_|\__'_|  |  Commit ca44f9b (0 days old master)
|__/                   |  x86_64-linux-gnu

julia> function foo(args::Vector{Tuple})
       end
foo (generic function with 1 method)

julia> foo([(1,2), (3,4)])
ERROR: no method foo(Array{(Int64,Int64),1})

NTuple instead of Tuple dispatches fine.

[simonster]

I'm not really sure this should work for either Tuple or NTuple, given that Julia's type parameters are invariant. Passing a Vector{Tuple} does work as expected:

julia> foo(convert(Vector{Tuple}, [(1,2), (3,4)]))

(although oddly Tuple[(1, 2), (3, 4)] throws an error)

[amitmurthy]

I don't quite understand why an explicit convert is required. Isn't Array{(Int64,Int64),1} a vector of tuples?

[JeffBezanson]

@simonster is quite right.

Tuple[...] isn't defined as an array constructor since tuples (and Tuple) are also indexable. Though it could perhaps be defined for unambiguous cases.

[amitmurthy]

Any explanation of why it is working with NTuple will be helpful.

[JeffBezanson]

Must be a bug.

[Jutho]

The following certainly explains your behaviour:

typeof([(1,2), (3,4)])  # returns  Array{(Int64,Int64),1} or thus Vector{(Int64,Int64)}
Vector{(Int64,Int64)} <: Vector{Tuple}  # returns false

and I do indeed remember from reading somewhere that it is not because T1 <: T2 that Array{T1} <: Array{T2}. However, I don't know how

Vector{(Int64,Int64)} <: Vector{NTuple}  # returns true

fits in this behaviour. I would think only the following strategy could work:

typealias TupleVector{N,T} Vector{NTuple{N,T}}
Vector{(Int64,Int64)} <: TupleVector # returns true