RFC: Implement complex(::Type)

[garrison]

Currently there is a method real(::Type) that converts a type to its "real" part:

julia> real(Float64)
Float64

julia> real(Complex128)
Float64

This pull request adds a similar method complex(::Type) that behaves as follows:

julia> complex(Float64)
Complex{Float64}

julia> complex(Complex128)
Complex{Float64}

This is much more clear than writing Complex{real(T)}, which was necessary to get this behavior before.

[johnmyleswhite]

This is very reasonable, but I worry that it puts us down the path of lowercase int and lowercase float again.

[ScottPJones]

I don't think so (that these are like the lowercase constructors int and float),
these seem to me to be more like trait functions such as eltype.

[garrison]

I will plan to merge this in a few days unless anybody objects more strongly. I do think this is distinct from lowercase int and float.

[nalimilan]

Why not add a method to the uppercase Complex constructor? AFAICT there should be no ambiguities.

[garrison]

This function mirrors lowercase real(::Type), so having it use uppercase Complex seems like a needless inconsistency to me.

[johnmyleswhite]

Yes, the distinction is real and important. I was not thinking clearly when I made that comment earlier.

[andreasnoack]

I'm using real(::Type{T}) all the time so I understand the usefulness of this PR. However, real predates the posibility of using constructors for abstract types so I'm wondering if we should start using Real, AbstractFloat etc. instead of the small letter versions. One problem is, that the float conversion actually represents an abstract type we don't have, i.e. Union{AbstractFloat,Complex{T<:AbstractFloat}}.

These changes can easily be made later though, so as long as we have real, I think is fine to have complex(::Type).

[nalimilan]

My suggestion was actually similar to what @andreasnoack said (somewhat more convincingly). So +1 to that.