Iterate on it – Mike Innes
Any Julia object – an array, a linked list, an infinite generator, whatever – can then implement iterate(xs) and iterate(xs, state) to make iterating over it with for possible, as well as getting a few other utilities like collect.
This is a lot nicer than Julia’s original, pre-1.0 iteration protocol which used start, next and done instead of the current Optional/Maybe-like pattern. The problem with the former approach is that some iterators, like Channels, don’t know whether they have a next value or not until you try them, which meant that many iterators awkwardly did all their work in done. So the 1.0 design above made it far easier to define asynchronous iterators like channels.
Unfortunately, the new protocol still has an issue: it can leak memory. Notice that xs must remain alive over the entire loop. In many cases that makes total sense (if you iterate over an array, you want the array to stick around so you can index it). If you have a linked list, though, it’s unnecessary: you only need the current element of the list, not the first, to carry on iterating, and if you let go of the head you can let the GC clean up the list as you go along.
In many cases, this is a minor missed optimisation opportunity. If you have very large functional data structures, even infinite ones (the functional equivalent of an indefinitely long generator / channel), it’s a disaster.
