Multiplication by -im has inconsistent imaginary part sign compared with floating point arithmetic

[dlfivefifty]

I found an inconsistency:

julia> -im * big(0.0-10.0im)
-10.0 - 0.0im

julia> -im * (0.0-10.0im)
-10.0 - 0.0im

julia> -im * ComplexDF64(0.0-10.0im) # has opposite imaginary sign
-10.0 + 0.0im

This causes the following to give different results:

julia> log( -im * (0.0-10.0im))
2.302585092994046 - 3.141592653589793im

julia> log(-im * ComplexDF64(0.0-10.0im))
2.30258509299404568401799145468437419 + 3.14159265358979323846264338327950588im

[araujoms]

This boils down to

julia> Double64(0) * Double64(-1)
0.0

julia> 0.0 * -1.0
-0.0

I don't think it is fixable. FWIW the sign disappears in this line:

DoubleFloats.jl/src/math/ops/op_dddd_dd.jl

Line 37 in f00b992

hi, lo = two_hilo_sum(hi, t)