Possible type related bug in Dict

[1oly]

Hi there,

By a coincidence I stumbled upon a possible bug in Dict.
See MWE below:

A test function that copies keys and values from one dict to another:

function foo(e::T,c::S) where {S,T}
    mydict1 = Dict{String,T}()
    mydict1["a"] = zero(T)
    mydict1["b"] = one(T)

    mydict2 = Dict{String,Array{T,1}}()
    N = 3
    a = zeros(S,N)
    for (key,value) in mydict1
        for i in 1:N
            a[i] = value
        end
        mydict2[key] = a
    end
    return mydict2
end

If T==S then the output is wrong:

julia> foo(1.,1.)
Dict{String,Array{Float64,1}} with 2 entries:
  "b" => [0.0, 0.0, 0.0]
  "a" => [0.0, 0.0, 0.0]

Or

julia> foo(1,1)
Dict{String,Array{Float64,1}} with 2 entries:
  "b" => [0, 0, 0]
  "a" => [0, 0, 0]

But if T != S then I get the correct output:

julia> foo(1,1.)
Dict{String,Array{Int64,1}} with 2 entries:
  "b" => [1, 1, 1]
  "a" => [0, 0, 0]

julia> foo(1.,1)
Dict{String,Array{Float64,1}} with 2 entries:
  "b" => [1.0, 1.0, 1.0]
  "a" => [0.0, 0.0, 0.0]

Tried to track it down further but could not find the source. Am I doing something silly here or is it a bug?

Can try to look more into if I can get some guidance :)

julia> versioninfo()
Julia Version 0.6.0
Commit 903644385b (2017-06-19 13:05 UTC)
Platform Info:
  OS: macOS (x86_64-apple-darwin13.4.0)
  CPU: Intel(R) Core(TM) i7-6700HQ CPU @ 2.60GHz
  WORD_SIZE: 64
  BLAS: libopenblas (USE64BITINT DYNAMIC_ARCH NO_AFFINITY Haswell)
  LAPACK: libopenblas64_
  LIBM: libopenlibm
  LLVM: libLLVM-3.9.1 (ORCJIT, skylake)

Best, Oliver

[yuyichao]

This is not a bug. There's no implicit copying so all values will point to the same object.

[yuyichao]

And ref #12441 though the aliasing behavior is definatly the expected one here.

[1oly]

Thanks for the quick reply and reference!