Optim.jl Causes a Strange Type Instability in norm

[cortner]

import Optim
function test_norm(R::Vector{Float64})
#         r = Base.LinAlg.vecnorm2(R)
        r = norm(R)
    return r
end
R1 = rand(3);
@code_warntype test_norm(R1)

Output is

Variables:
  #self#::#test_norm
  R::Array{Float64,1}
  r::Any

Body:
  begin
      r::Any = $(Expr(:invoke, LambdaInfo for vecnorm(::Array{Float64,1}, ::Int64), :(Base.LinAlg.vecnorm), :(R), 2)) # line 4:
      return r::Any
  end::Any

Comments:

• With vecnorm2 the type of r is inferred correctly.
• If I don't import Optim first, then the type is also inferred correctly.
• This happens on v0.5-rc4, but not on v0.4.6
• It happens on Optim master as well as the current release.

[pkofod]

I'm thinking JuliaLang/julia#18465 JuliaLang/julia#18449 perhaps?

[cortner]

yes, it is fixed in the nightly.

[pkofod]

Well... yeah... Good to know it is fixed at least :)

[cortner]

Before closing here, can we wait until it is clear that 0.5 will backport the fix?

[pkofod]

Yes, let's keep it open.

[pkofod]

Well, there doesn't seem to be any 0.5.x before 0.6.0. It's fixed on v0.6 so I guess I might as well just close this, as v0.6 should be "just around the corner".