diagonal dispatch type-inference error

[vtjnash]

I noticed the following discrepancy while attempting to enable inline_incompletematch_allowed:

julia> Base._methods(convert, (Type{TypeVar(:_,Union(Bool,Function))}, Bool), 1)
1-element Array{Any,1}:
 svec(Tuple{Type{Bool},Bool},svec(),convert(::Type{Bool},x::Bool) at bool.jl:3)

julia> Base._methods(convert, (Type{TypeVar(:_,Union(Bool,Function))}, Bool), -1)
3-element Array{Any,1}:
 svec(Tuple{Type{Bool},Bool},svec(),convert(::Type{Bool},x::Bool) at bool.jl:3)    
 svec(Tuple{Type{Bool},Bool},svec(),convert(::Type{Bool},x::Real) at bool.jl:4)    
 svec(Tuple{Type{Bool},Bool},svec(T,Bool),convert{T}(::Type{T},x::T) at base.jl:38) # this method is missing from the above list

this can trigger a fault in inference given the following method definitions:

julia> type AT
         f::Union(Bool,Function)
       end

julia> g{S}(::Type{S}, ::S) = Int64(1)
g (generic function with 1 method)

julia> f(a::AT) = g(Base.fieldtype(typeof(a), :f), true)
f (generic function with 1 method)

julia> g(::Type{Bool}, ::Bool) = 2.0
g (generic function with 2 methods)

julia> code_typed(f,(AT,))
1-element Array{Any,1}:
 :($(Expr(:lambda, Any[:a], Any[Any[],Any[Any[:a,AT,0]],Any[],Any[]], :(begin  # none, line 1:
        return g(((top(getfield))(Base,:fieldtype))(typeof(a::AT)::Type{AT},:f)::Type{_<:Union(Bool,Function)},true)::Float64
    end::Float64))))

julia> f(AT(true))
5.0e-324
#oops: should be 1 but got `reinterpret(Float64,Int64(1))`

(note that g(fieldtype, x) is a derived from the lowered form of a field-assignment, so this situation is actually fairly plausible to create in practice)`

[vtjnash]

I assume this will be covered by #8974 in v0.6?