fix length for zips with cycles

[mschauer]

This fixes a small issue with Zips containing infinite components which prevented constructions like

[i for i in zip(1:10, cycle(1:2))]

[hayd]

An alternative to doing it this way is to define the length of Cycle and Repeated to be infinite:

Base.length(z::Base.Zip2) = Int(min(length(z.a), length(z.b)))   # change
Base.length(c::Base.Cycle) = Inf                                 # add
Base.length(r::Base.Repeated) = Inf                              # add

This would generalize but may be controversial (as length is otherwise an Int).

[mschauer]

[edit: moved the argument over to #11977]

[ivarne]

See also discussion in #11977

[mschauer]

Sorry, @simonbyrne , there was no intention to pre-empt the introduction of Alephs into julia with this PR ;-)

[stevengj]

You could also define the length of an infinite iterable to be typemax(Int), I suppose.

[mschauer]

@stevengj My first reaction is that this would be not a good idea. Producing an error when asking for the length of an infinite object is the current convention. That also sounds really prone to produce overflow when doing innocent computations like n = length(l)+1.

[mschauer]

#11977 is closed, can we go ahead here?

[hayd]

How's about at the top we do

typealias InfiniteIterators Union{Cycle,Repeated}

and add a TODO that we should do this with traits once we have #13222.

[JeffBezanson]

That's a good suggestion.

[mschauer]

Can be done in a much nicer way with the new Iterator traits.