"Key not found" error with colons

[IainNZ]

julia> m = Model()
julia> @defVar(m, x[i=1:5,j=1:5,k=0:j])
julia> x[:,1,0]
ERROR: KeyError: (Colon(),1,0) not found
 in getindex at /Users/idunning/.julia/v0.4/JuMP/src/JuMPContainer.jl:48

We should probably be providing a better error message here

[mlubin]

We can fix #287 now

[joehuchette]

FWIW:

julia> @defVar(m, x[i=1:5,j=1:5,k=0:5]);

julia> x[:,1,0]
ERROR: MethodError: `isless` has no method matching isless(::Colon, ::Int64)
Closest candidates are:
  isless(::AbstractFloat, ::Real)
  isless(::Real, ::Real)
  isless(::Char, ::Integer)
  ...
 in getindex at /Users/huchette/.julia/v0.4/JuMP/src/JuMPArray.jl:61

[mlubin]

We can give a better error message here, if we assume that nobody would use [Colon()] as an index set...

[joehuchette]

Well we can always make

S = [Colon()]
@defVar(m, x[S])

an error...

[joehuchette]

It's weird, but I'm not sure we can disambiguate the case @IainNZ mentions and this one:

julia> @defVar(m, y[[:]]);

julia> y[:]
1-element Array{JuMP.Variable,1}:
 y[Colon()]

[mlubin]

@joehuchette, not sure I understand. Can we disambiguate it if we disallow colons in keys?

[joehuchette]

I guess? Seems kinda arbitrary and brittle

[mlubin]

Not having slicing work is also brittle and seemingly arbitrary since it works for some kinds of JuMP variables and not others.

[joehuchette]

It's certainly less brittle to reliably throw an error than give completely unexpected behavior if you use a : or Colon() in your index set for some reason. I'm open to the idea of allowing slices for JuMPDicts, but I don't think the implementation will be particularly straightforward.