false ambiguity detection?

[Sacha0]

Reduced from #20970,

julia> versioninfo()
Julia Version 0.6.0-pre.alpha.126
Commit 0615ea2 (2017-03-12 13:00 UTC)
Platform Info:
  OS: macOS (x86_64-apple-darwin15.6.0)
  CPU: Intel(R) Core(TM) i7-3520M CPU @ 2.90GHz
  WORD_SIZE: 64
  BLAS: libopenblas (USE64BITINT DYNAMIC_ARCH NO_AFFINITY Sandybridge)
  LAPACK: libopenblas64_
  LIBM: libopenlibm
  LLVM: libLLVM-3.9.1 (ORCJIT, ivybridge)

julia> begin
           bar(x::AbstractMatrix{T}, y::Vector{Complex{T}}) where T<:Float64 = 1
           bar(x::Matrix{T}, y::Vector) where T<:Float64 = 2
           bar(x::Matrix{Float64}, y::Vector{Complex{Float64}}) = 3

           for (ma, mb) in Test.detect_ambiguities(Main; imported = true, allow_bottom = false)
               println("ma: ", ma)
               println("mb: ", mb)
               println("=> ", typeintersect(ma.sig, mb.sig))
           end
       end
ma: bar(x::AbstractArray{T,2}, y::Array{Complex{T},1}) where T<:Float64 in Main at REPL[3]:2
mb: bar(x::Array{T,2}, y::Array{T,1} where T) where T<:Float64 in Main at REPL[3]:3
=> Tuple{#bar,Array{T,2},Array{Complex{T},1}} where T<:Float64

julia> bar(rand(4, 4), complex.(rand(4), rand(4)))
3

julia> bar(x::Matrix{T}, y::Vector{Complex{T}}) where T<:Float64 = 4
bar (generic function with 4 methods)

julia> for (ma, mb) in Test.detect_ambiguities(Main; imported = true, allow_bottom = false)
           println("ma: ", ma)
           println("mb: ", mb)
           println("=> ", typeintersect(ma.sig, mb.sig))
       end

julia> bar(rand(4, 4), complex.(rand(4), rand(4)))
3

Best!

[yuyichao]

Seems like a missing ambiguity error.

[Sacha0]

Seems like a missing ambiguity error.

How so? You mean the behavior after introducing the fourth bar definition?

[yuyichao]

NVM, I think I was confused by allow_bottom and which behavior is false...........

[StefanKarpinski]

That keyword argument should probably be called ambiguous_bottom::Bool where true means that bottom is ambiguous while false means that bottom is not considered ambiguous.

[timholy]

See an alternative proposal at the bottom of #20982 (comment). If people don't think it's too late to change this for 0.6...

[JeffBezanson]

I believe this is correct behavior. The case bar(rand(4, 4), complex.(rand(4), rand(4))) is covered by the third method definition, which is not ambiguous. The case bar(Array{Union{},2}(0,0), Array{Complex{Union{}},1}(0)) however does fall in the ambiguity, and indeed gives an error, so I think we're good here.