return_types errors with: _nref has no method matching _nref(::Type{N}, ::Symbol, ::Expr)

[mauro3]

I get these errors after updating my julia master:

   _       _ _(_)_     |  A fresh approach to technical computing
  (_)     | (_) (_)    |  Documentation: http://docs.julialang.org
   _ _   _| |_  __ _   |  Type "help()" for help.
  | | | | | | |/ _` |  |
  | | |_| | | | (_| |  |  Version 0.4.0-dev+1783 (2014-11-24 06:06 UTC)
 _/ |\__'_|_|_|\__'_|  |  Commit 3ccaf3c* (0 days old master)
|__/                   |  x86_64-unknown-linux-gnu

julia> Base.return_types(getindex, (Array{Int64,2}, Any))
ERROR: `_nref` has no method matching _nref(::Type{N}, ::Symbol, ::Expr)
 in func_for_method at ./inference.jl:602
 in return_types at inference.jl:3072

julia> Base.return_types(setindex!, (Array{Int64,2}, Any, Base.IteratorsMD.CartesianIndex))
ERROR: `_nref` has no method matching _nref(::Type{N}, ::Symbol, ::Expr)
 in func_for_method at ./inference.jl:602
 in return_types at inference.jl:3072

julia> Base.return_types(getindex, (Array{Int64,2}, Base.IteratorsMD.CartesianIndex))
ERROR: `_nref` has no method matching _nref(::Type{N}, ::Symbol, ::Expr)
 in func_for_method at ./inference.jl:602
 in return_types at inference.jl:3072

Might be this: 81136af

[timholy]

This specific error is, once again, #8504. However, even if we apply #8973, this still gives an error:

julia> Base.return_types(getindex, (Array{Int64,2}, Any))
ERROR: `anonymous` has no method matching anonymous(::Type{Array{Int64,2}}, ::Type{Any})
 in func_for_method at ./inference.jl:602
 in return_types at inference.jl:3077

Perhaps we need a try/catch block in that loop in Base.return_types?

[timholy]

More info:

julia> Base.return_types(getindex, (Array{Int64,2}, Base.IteratorsMD.CartesianIndex{2}))
1-element Array{Any,1}:                                                                                                                                                                                                                                                        
 Union(Int64,Array{T,N})                                                                                                                                                                                                                                                       

julia> Base.return_types(getindex, (Array{Int64,2}, Base.IteratorsMD.CartesianIndex_2))
ERROR: CartesianIndex_2 not defined                                                                                                                                                                                                                                            

julia> Base.IteratorsMD.gen_cartesian(2)
(:CartesianIndex_2,:IndexIterator_2)                                                                                                                                                                                                                                           

julia> Base.return_types(getindex, (Array{Int64,2}, Base.IteratorsMD.CartesianIndex_2))
1-element Array{Any,1}:
 Int64

[mauro3]

Issue #8504 is now fixed but this still does not work. However, the errors changed, probably due to #10178:

  _       _ _(_)_     |  A fresh approach to technical computing
  (_)     | (_) (_)    |  Documentation: http://docs.julialang.org
   _ _   _| |_  __ _   |  Type "help()" for help.
  | | | | | | |/ _` |  |
  | | |_| | | | (_| |  |  Version 0.4.0-dev+4022 (2015-03-26 19:08 UTC)
 _/ |\__'_|_|_|\__'_|  |  Commit 01a3216* (0 days old master)
|__/                   |  x86_64-unknown-linux-gnu

julia> Base.return_types(getindex, (Array{Int64,2}, Any))
ERROR: 
 in return_types at inference.jl:3364

monkey-patching as suggested by @Jutho in #10178 gives a similar error to when I first reported this:

   _       _ _(_)_     |  A fresh approach to technical computing
  (_)     | (_) (_)    |  Documentation: http://docs.julialang.org
   _ _   _| |_  __ _   |  Type "help()" for help.
  | | | | | | |/ _` |  |
  | | |_| | | | (_| |  |  Version 0.4.0-dev+4022 (2015-03-26 19:08 UTC)
 _/ |\__'_|_|_|\__'_|  |  Commit 01a3216* (0 days old master)
|__/                   |  x86_64-unknown-linux-gnu

julia> function Base.func_for_method(m::Method, tt, env)
           if !m.isstaged
               return m.func.code
           else
               f=ccall(:jl_instantiate_staged,Any,(Any,Any,Any),m,tt,env)
               return f.code
           end
       end
Warning: Method definition func_for_method(Method,Any,Any) in module Base at inference.jl:666 overwritten in module Main at none:2.
func_for_method (generic function with 1 method)

julia> Base.return_types(getindex, (Array{Int64,2}, Any))
ERROR: MethodError: `_ncall` has no method matching _ncall(::Type{N}, ::Expr, ::Symbol, ::Expr)
Closest candidates are:
  _ncall(::Int64, ::Any, ::Any...)
 in func_for_method at none:5
 in return_types at inference.jl:3364

Any change of a fix with #8504 in place?

[timholy]

There's #9139, but I'm not sure if it's a good idea.

[JeffBezanson]

The behavior

julia> Base.return_types(getindex, (Array{Int64,2}, Any))
ERROR: 
 in return_types at inference.jl:3364

is expected, since we can only call staged functions on concrete types. An error message was left out since this was intended only for internal consumption. We can add a message. Look for error() in inference.jl.