ambiguity error with non-applicable methods

[rfourquet]

julia> g(::Type{<:AbstractVector{T}}, ::Int) where {T} = 1

julia> g(Vector, 1)
ERROR: MethodError: no method matching g(::Type{Array{T,1} where T}, ::Int64)
Closest candidates are:
  g(::Type{#s1} where #s1<:AbstractArray{T,1}, ::Int64) where T at REPL[1]:1
Stacktrace:
 [1] top-level scope at REPL[2]:1

julia> g(::Type{Vector}, ::Any) = 2

julia> g(Vector, 1)
ERROR: MethodError: g(::Type{Array{T,1} where T}, ::Int64) is ambiguous. Candidates:
  g(::Type{#s1} where #s1<:AbstractArray{T,1}, ::Int64) where T in Main at REPL[1]:1
  g(::Type{Array{T,1} where T}, ::Any) in Main at REPL[3]:1
Possible fix, define
  g(::Type{Array{T,1} where T}, ::Int64)
Stacktrace:
 [1] top-level scope at REPL[4]:1

julia> f() = g(Vector, 1); f() # works on Julia 1.1 ! But errors on master
2

The last line with f is the reason why tests in my package where passing on 1.1 but not on master.

[Keno]

The error on master is a bugfix, so it probably doesn't count a regression (just that one bug canceled out another). @JeffBezanson will have to judge the type system issue.

[rfourquet]

Yes agreed that it's not a regression, I inserted the example with f just as contextual information.

[JeffBezanson]

Type intersection used to give exactly the first signature as its result, even though they should be disjoint. Now it gives Type{Vector}, Int which is not really better or worse, but doesn't equal either signature hence the ambiguity.