Chapter VIII: Advanced topics

It's dangerous to go alone...

The goal of this tutorial was to introduce generators as the centerpiece of iteration in ES6. With that done, we'll briefly look at a few advanced use cases. This isn't meant to be an exhaustive tour of everything about generators, just a few items to take with you as you leave and explore the world of generators on your own.

A word of warning

Recall that iterators are stateful objects. When an iterator comes from a collection, it "remembers" its place in that collection at each step. Hence the rule: Never modify a collection while iterating it. Otherwise, things can go haywire since the iterator's state could be invalidated.

// BAD!
for (const val of collection) {

POJOs aren't iterable

Given that you can for/in loop over keys in a plain old JavaScript object (POJO), you could be forgiven for thinking POJOs are iterable. Well, they're not. You can use a Map instead, or a write yourself a utility generator function.

function* iterObj(ob) {
  for (let key in ob) {
    if (ob.hasOwnProperty(key)) {
      yield [ key, ob[key] ];

for (const [ key, val ] of iterObj(obj)) { ... }

The generator-detection antipattern

If you work with generators much, eventually you'll find yourself thinking, "it sure would be handy to detect whether this function is a generator." Don't do it!

While it may be possible to exploit various quirks of your local JavaScript engine to sniff out a generator, the intent is that, to the outside world, generators should be indistinguishable from normal functions that return iterators. For example, suppose you have a function that accepts a callback:

function runGenerator(generator) {
  if (!isGenerator(generator)) {
    throw new Error('requires a generator!');
  } else {
    for (const x of generator()) { ... }

// intended usage
runGenerator(function*() { ... });

That would break for this perfectly legitimate use case, since Function#bind() returns a new function:

runGenerator(function*() { ... }.bind(this));

If you need to do some sort of validation, consider instead just inspecting the return value of the function:

function runGenerator(generator) {
  var itr = generator();
  if (!isIterator(itr)) {
    throw new Error('must return an iterator!');
  } else {
    for (const x of itr) { ... }

Functional programming over sequences

Considering that iteration introduces a conceptual shift from collections to abstract sequences, collection-oriented libraries like lodash start to seem incomplete. Operations like map, filter, and reduce can just as easily operate on infinite or lazy sequences, for example. The wu library offers these kinds of capabilities, and can be thought of as "lodash for iterators":

const evenSquares = wu(range(0, Infinity))
.map(n => n * n)
.filter(n => n % 2 === 0);

for (let n of evenSquares) {
// => 0
// => 4
// => 16
// => 36
// => ...

Two-way communication

It's syntactically valid for yield or yield x to appear anywhere an expression is expected. For example:

var myString = `Hello ${yield 4}`;

This raises the question: what end up being the contents of myString?

So far we've only looked at consuming information from generators. But here's yet another twist: if you have an iterator "itr" gotten from a generator, you can pass in an argument: Inside the generator, what this looks like is that a yield expression evaluates to someValue.

This is significant, because now instead of a one-way consumer/producer relationship, we have a two-way, "ping-pong" sort of relationship, with the generator sending things to its runner, and the runner sending things back to the generator.

// generator
function* foo() {
  console.log(`Hello ${yield 1}!`); // 'Hello world!'

// runner
var itr = foo();
console.log(; // => 1'world');

Not only can information be sent into a generator via next(), but so can errors via throw(), which inside the generator you handle like any other error, treating a yield as something that might throw an exception.

function* foo() {
  try { yield 1; }
  catch(ex) { console.log(ex); } // 'oops'

var itr = foo();
console.log(; // 1

Note that these capabilities aren't part of the iterator protocol. Rather, they're extra powers that generator-created iterators have. Read more about it here.

Async generators

This two-way communication between a generator and its runner opens up new vistas in async flow control. A generator can yield promises out to a special runner, which resolves it and then "bounces" the resolved value right back into the generator, asynchronously. What this looks like inside a generator is that a yield promise expression evaluates to the resolved value of that promise, or else throws an exception.

Here's an example using the fetch API, which is promise-based:

// in the generator
var resp = yield fetch('/users/123');
var user = yield resp.json();

This "trampoline" technique gives us callback-free async flow control. All we need is a runner function, which we'll call coroutine(), and a generator function to pass to it:

const getUsername = coroutine(function*(id) {
  let resp = yield fetch(`/users/${id}`);
  var obj = yield resp.json();
  if (resp.status === 200) return obj.username;
  else throw new Error(obj.errorMessage);

const getGreeting = coroutine(function*() {
  const myId = localStorage.getItem('my_id');
  if (!myId) throw new Error('not logged in');
  else return `Hello, ${yield getUsername(myId)}`;

  greeting => alert(greeting),
  error => alert(error.message)


The upshot is that we retain the power of asynchronous programming, while gaining the power of imperative, synchronous-style programming. No more callback hell, or thenable heck!

All of this of course depends on having a coroutine() function that does the right thing. Fortunately, there are multiple libraries to choose from here, including co, Bluebird.coroutine, and Q.spawn.

Finally, it's worth noting that this approach is so powerful that it inspired the async functions proposal. As of early 2016, it's a stage 3 EcmaScript proposal.


Copyright © 2016 by Greg Reimer (github, twitter). Submit issues to the GitHub issues page.