Type inference problem with binary operators on Julia master

[KristofferC]

Just a heads up that something is broken in type inference when using Julia master.

Using this simple function:

function foo(x)
    println(eltype(x))
    b = 5.0
    c = x / b
    println(eltype(c))

    d = x * (1/b)
    println(eltype(d))
    return 3 * c
end

gives on 0.4.1 the normal:

julia> ForwardDiff.jacobian(foo, rand(2))
ForwardDiff.GradientNumber{2,Float64,Tuple{Float64,Float64}}
ForwardDiff.GradientNumber{2,Float64,Tuple{Float64,Float64}}
ForwardDiff.GradientNumber{2,Float64,Tuple{Float64,Float64}}
2x2 Array{Float64,2}:
 0.6  0.0
 0.0  0.6

However, on 0.5 this gives:

julia> ForwardDiff.jacobian(foo, rand(2))
ForwardDiff.GradientNumber{2,Float64,Tuple{Float64,Float64}}
########
ForwardDiff.GradientNumber{N,T,C} # <------ NOTE
########
ForwardDiff.GradientNumber{2,Float64,Tuple{Float64,Float64}}
ERROR: no promotion exists for Int64 and ForwardDiff.GradientNumber{N,T,C}
 [inlined code] from promotion.jl:160
 in .* at arraymath.jl:118
 in foo at none:9
 in _calc_jacobian at /home/kristoffer/.julia/v0.5/ForwardDiff/src/api/jacobian.jl:101
 in jacobian at /home/kristoffer/.julia/v0.5/ForwardDiff/src/api/jacobian.jl:84
 in eval at ./boot.jl:263

It seems that the division x/b cause julia to lose the type inference for the array.

[KristofferC]

Some warntype stuff shows it as well:

Good with *:

julia> @code_warntype *(Array(ForwardDiff.GradientNumber{2,Float64,Tuple{Float64,Float64}}),1.0)
Variables:
  A::Array{ForwardDiff.GradientNumber{2,Float64,Tuple{Float64,Float64}},0}
  B::Float64

Body:
  begin  # abstractarraymath.jl, line 55:
      return A::Array{ForwardDiff.GradientNumber{2,Float64,Tuple{Float64,Float64}},0} .* B::Float64::Array{ForwardDiff.GradientNumber{2,Float64,Tuple{Float64,Float64}},0}
  end::Array{ForwardDiff.GradientNumber{2,Float64,Tuple{Float64,Float64}},0}

Bad with /:

julia> @code_warntype /(Array(ForwardDiff.GradientNumber{2,Float64,Tuple{Float64,Float64}}),1.0)
Variables:
  A::Array{ForwardDiff.GradientNumber{2,Float64,Tuple{Float64,Float64}},0}
  B::Float64

Body:
  begin  # abstractarraymath.jl, line 57:
      return A::Array{ForwardDiff.GradientNumber{2,Float64,Tuple{Float64,Float64}},0} ./ B::Float64::Array{ForwardDiff.GradientNumber{N,T,C},0}
  end::Array{ForwardDiff.GradientNumber{N,T,C},0}

[KristofferC]

Doing @code_warntype on ./ shows the following instability in one of the lines:

      GenSym(4) = x::ForwardDiff.GradientNumber{2,Float64,Tuple{Float64,Float64}} ./ y::Float64::ForwardDiff.GradientNumber{N,T,C}

[KristofferC]

Seems to be a regression in type inference?
Same code: http://imgur.com/vcUHMHC

[KristofferC]

Tracked at JuliaLang/julia#14294

[jrevels]

Very nice detective work, thanks!

[KristofferC]

Confirmed fixed on master

julia> Base.return_types(*, (ForwardDiff.GradientNumber{2,Float64,Tuple{Float64,Float64}},Float64))
1-element Array{Any,1}:
 ForwardDiff.GradientNumber{2,Float64,Tuple{Float64,Float64}}

[mlubin]

There was no regression test added upstream, could we put one here?

[KristofferC]

I don't have a small repo right now.

[Ken-B]

It seems there was a regression on Julia master for this, could someone else verify?

julia> Base.return_types(*, (ForwardDiff.GradientNumber{2,Float64,Tuple{Float64,Float64}},Float64))
1-element Array{Any,1}:
 ForwardDiff.GradientNumber{N,T,C}

julia> versioninfo()
Julia Version 0.5.0-dev+3977
Commit 957f1d1 (2016-05-08 00:28 UTC)
Platform Info:
  System: Darwin (x86_64-apple-darwin15.4.0)
  CPU: Intel(R) Core(TM) i7-4980HQ CPU @ 2.80GHz
  WORD_SIZE: 64
  BLAS: libopenblas (USE64BITINT DYNAMIC_ARCH NO_AFFINITY Haswell)
  LAPACK: libopenblas64_
  LIBM: libopenlibm
  LLVM: libLLVM-3.7.1 (ORCJIT, haswell)