zip() returns scalar instead of tuple when passed a single argument in 0.4

[nalimilan]

Is this intended or a regression?

In 0.3:

julia> i = zip(1:2)
Zip{(UnitRange{Int64},)}((1:2,))

julia> start(i)
(1,)

julia> next(i, (1,))
((1,),(2,))

In 0.4:

julia> i = zip(1:2)
2-element UnitRange{Int64}:
 1,2

julia> start(i)
1

julia> next(i, 1)
(1,2)

This means you cannot handle single-argument and multiple-argument cases with the same code. I encountered this issue when doing something similar to:

function (x...) 
...
    vtypes = map(eltype, typeof(x).parameters)
    d = Dict{Tuple{vtypes...}, Any}()
    for el in zip(x...)
        # Error in convert()
        d[el] = true
    end
...
end

[nalimilan]

A line zip(a) = a was introduced by @JeffBezanson in eaf5a95. Since there is no mention of this particular change, this looks like an oversight to me. That line also isn't exercised by the tests at all.

See PR #13990