Arrays changes broke JuMP

[mlubin]

using JuMP
m = Model()
@variable(m, tri_3[(i,j)=[(i,i+2) for i in 1:5],k=i:j])

gives

ERROR: UndefRefError: access to undefined reference
 in copy!(::Base.LinearFast, ::Array{Tuple{Vararg{Any,2}},1}, ::Base.LinearFast, ::Array{Tuple{Any,Any},1}) at ./abstractarray.jl:412
 in rehash!(::Dict{Tuple{Vararg{Any,2}},JuMP.Variable}, ::Int64) at ./dict.jl:483
 in _setindex!(::Dict{Tuple{Vararg{Any,2}},JuMP.Variable}, ::JuMP.Variable, ::Tuple{Tuple{Int64,Int64},Int64}, ::Int64) at ./dict.jl:606
 in setindex!(::Dict{Tuple{Vararg{Any,2}},JuMP.Variable}, ::JuMP.Variable, ::Tuple{Tuple{Int64,Int64},Int64}) at ./dict.jl:626
 in setindex!(::JuMP.JuMPDict{JuMP.Variable,2}, ::JuMP.Variable, ::Tuple{Int64,Int64}, ::Int64) at /home/mlubin/.julia/v0.5/JuMP/src/JuMPContainer.jl:49
 in macro expansion; at /home/mlubin/.julia/v0.5/JuMP/src/macros.jl:137 [inlined]
 in anonymous at ./<missing>:?
 in eval(::Module, ::Any) at ./boot.jl:234
 in macro expansion at ./REPL.jl:92 [inlined]
 in (::Base.REPL.##1#2{Base.REPL.REPLBackend})() at ./event.jl:46

This seems to have been broken for quite a while but I haven't been using 0.5. I can spend a bit more time to generate a pure julia error case, but opening this in case the issue is obvious.
CC @IainNZ @joehuchette @tkelman @timholy

Version:

julia> versioninfo()
Julia Version 0.5.0-dev+5247
Commit dad6836 (2016-07-09 01:29 UTC)
Platform Info:
  System: Linux (x86_64-linux-gnu)
  CPU: Intel(R) Core(TM) i5-3320M CPU @ 2.60GHz
  WORD_SIZE: 64
  BLAS: libopenblas (USE64BITINT DYNAMIC_ARCH NO_AFFINITY Sandybridge)
  LAPACK: libopenblas64_
  LIBM: libopenlibm
  LLVM: libLLVM-3.7.1 (ORCJIT, ivybridge)

[timholy]

This actually looks like a type-system error: the key line in the backtrace (obtained with --inline=no) is

in copy!(::Array{Tuple{Vararg{Any,2}},1}, ::Array{Tuple{Any,Any},1}) at ./abstractarray.jl:423

Vararg{Any,2} should be instantiated as Tuple{Any,Any}. Once these are the same, then rather than calling this definition it should call this one.

[timholy]

It seems possible this is the same as #17350, and if so that a possible fix might already be on the way.

[timholy]

To explain a little further, what's happening is that copy! is getting called as a consequence of Dict rehashing, and the array that's being copied has undefined elements in it. The Array-specific copy! method handles #undef.

[JeffBezanson]

Reverting the offending commit from #17350 doesn't seem to fix this.