Iteration protocols

The iterable protocol allows JavaScript objects to define or customize their iteration behavior, such as what values are looped over in a for...of construct. Some built-in types are built-in iterables with a default iteration behavior, such as Array or Map, while other types (such as Object) are not.

In order to be iterable, an object must implement the @@iterator method, meaning that the object (or one of the objects up its prototype chain) must have a property with a @@iterator key which is available via constant Symbol.iterator:

[Symbol.iterator]

A zero-argument function that returns an object, conforming to the iterator protocol.

Whenever an object needs to be iterated (such as at the beginning of a for...of loop), its @@iterator method is called with no arguments, and the returned iterator is used to obtain the values to be iterated.

Note that when this zero-argument function is called, it is invoked as a method on the iterable object. Therefore inside of the function, the this keyword can be used to access the properties of the iterable object, to decide what to provide during the iteration.

This function can be an ordinary function, or it can be a generator function, so that when invoked, an iterator object is returned. Inside of this generator function, each entry can be provided by using yield.

The iterator protocol defines a standard way to produce a sequence of values (either finite or infinite), and potentially a return value when all values have been generated.

An object is an iterator when it implements a next() method with the following semantics:

next()

A function that accepts zero or one argument and returns an object conforming to the IteratorResult interface (see below). If a non-object value gets returned (such as false or undefined) when a built-in language feature (such as for...of) is using the iterator, a TypeError ("iterator.next() returned a non-object value") will be thrown.

All iterator protocol methods (next(), return(), and throw()) are expected to return an object implementing the IteratorResult interface. It must have the following properties:

done Optional

A boolean that's false if the iterator was able to produce the next value in the sequence. (This is equivalent to not specifying the done property altogether.)

Has the value true if the iterator has completed its sequence. In this case, value optionally specifies the return value of the iterator.

value Optional

Any JavaScript value returned by the iterator. Can be omitted when done is true.

In practice, neither property is strictly required; if an object without either property is returned, it's effectively equivalent to { done: false, value: undefined }.

If an iterator returns a result with done: true, any subsequent calls to next() are expected to return done: true as well, although this is not enforced on the language level.

The next method can receive a value which will be made available to the method body. No built-in language feature will pass any value. The value passed to the next method of generators will become the value of the corresponding yield expression.

Optionally, the iterator can also implement the return(value) and throw(exception) methods, which, when called, tells the iterator that the caller is done with iterating it and can perform any necessary cleanup (such as closing database connection).

return(value) Optional

A function that accepts zero or one argument and returns an object conforming to the IteratorResult interface, typically with value equal to the value passed in and done equal to true. Calling this method tells the iterator that the caller does not intend to make any more next() calls and can perform any cleanup actions. When built-in language features call return() for cleanup, value is always undefined.

throw(exception) Optional

A function that accepts zero or one argument and returns an object conforming to the IteratorResult interface, typically with done equal to true. Calling this method tells the iterator that the caller detects an error condition, and exception is typically an Error instance. No built-in language feature calls throw() for cleanup purposes — it's a special feature of generators for the symmetry of return/throw.

It is very easy to make an iterator also iterable: just implement an [@@iterator]() method that returns this.

// Satisfies both the Iterator Protocol and Iterable
const myIterator = {
  next() {
    // ...
  },
  [Symbol.iterator]() {
    return this;
  },
};

Such object is called an iterable iterator. Doing so allows an iterator to be consumed by the various syntaxes expecting iterables — therefore, it is seldom useful to implement the Iterator Protocol without also implementing Iterable. (In fact, almost all syntaxes and APIs expect iterables, not iterators.) The generator object is an example:

const aGeneratorObject = (function* () {
  yield 1;
  yield 2;
  yield 3;
})();

console.log(typeof aGeneratorObject.next);
// "function" — it has a next method (which returns the right result), so it's an iterator

console.log(typeof aGeneratorObject[Symbol.iterator]);
// "function" — it has an @@iterator method (which returns the right iterator), so it's an iterable

console.log(aGeneratorObject[Symbol.iterator]() === aGeneratorObject);
// true — its @@iterator method returns itself (an iterator), so it's an iterable iterator

All built-in iterators inherit from Iterator.prototype, which implements the [@@iterator]() method as returning this, so that built-in iterators are also iterable.

However, when possible, it's better for iterable[Symbol.iterator] to return different iterators that always start from the beginning, like Set.prototype[@@iterator]() does.

There are another pair of protocols used for async iteration, named async iterator and async iterable protocols. They have very similar interfaces compared to the iterable and iterator protocols, except that each return value from the calls to the iterator methods is wrapped in a promise.

An object implements the async iterable protocol when it implements the following methods:

[Symbol.asyncIterator]

A zero-argument function that returns an object, conforming to the async iterator protocol.

An object implements the async iterator protocol when it implements the following methods:

next()

A function that accepts zero or one argument and returns a promise. The promise fulfills to an object conforming to the IteratorResult interface, and the properties have the same semantics as those of the sync iterator's.

return(value) Optional

A function that accepts zero or one argument and returns a promise. The promise fulfills to an object conforming to the IteratorResult interface, and the properties have the same semantics as those of the sync iterator's.

throw(exception) Optional

A function that accepts zero or one argument and returns a promise. The promise fulfills to an object conforming to the IteratorResult interface, and the properties have the same semantics as those of the sync iterator's.

The language specifies APIs that either produce or consume iterables and iterators.

String, Array, TypedArray, Map, Set, and Segments (returned by Intl.Segmenter.prototype.segment()) are all built-in iterables, because each of their prototype objects implements an @@iterator method. In addition, the arguments object and some DOM collection types such as NodeList are also iterables. There is no object in the core JavaScript language that is async iterable. Some web APIs, such as ReadableStream, have the Symbol.asyncIterator method set by default.

Generator functions return generator objects, which are iterable iterators. Async generator functions return async generator objects, which are async iterable iterators.

The iterators returned from built-in iterables actually all inherit from a common class Iterator, which implements the aforementioned [Symbol.iterator]() { return this; } method, making them all iterable iterators. The Iterator class also provides additional helper methods in addition to the next() method required by the iterator protocol. You can inspect an iterator's prototype chain by logging it in a graphical console.

console.log([][Symbol.iterator]());

Array Iterator {}
  [[Prototype]]: Array Iterator     ==> This is the prototype shared by all array iterators
    next: ƒ next()
    Symbol(Symbol.toStringTag): "Array Iterator"
    [[Prototype]]: Object           ==> This is the prototype shared by all built-in iterators
      Symbol(Symbol.iterator): ƒ [Symbol.iterator]()
      [[Prototype]]: Object         ==> This is Object.prototype

There are many APIs that accept iterables. Some examples include:

• Map()
• WeakMap()
• Set()
• WeakSet()
• Promise.all()
• Promise.allSettled()
• Promise.race()
• Promise.any()
• Array.from()
• Object.groupBy()
• Map.groupBy()

const myObj = {};

new WeakSet(
  (function* () {
    yield {};
    yield myObj;
    yield {};
  })(),
).has(myObj); // true

Some statements and expressions expect iterables, for example the for...of loops, array and parameter spreading, yield*, and array destructuring:

for (const value of ["a", "b", "c"]) {
  console.log(value);
}
// "a"
// "b"
// "c"

console.log([..."abc"]); // ["a", "b", "c"]

function* gen() {
  yield* ["a", "b", "c"];
}

console.log(gen().next()); // { value: "a", done: false }

[a, b, c] = new Set(["a", "b", "c"]);
console.log(a); // "a"

When built-in syntaxes are iterating an iterator, and the last result's done is false (i.e. the iterator is able to produce more values) but no more values are needed, the return method will get called if present. This can happen, for example, if a break or return is encountered in a for...of loop, or if all identifiers are already bound in an array destructuring.

const obj = {
  [Symbol.iterator]() {
    let i = 0;
    return {
      next() {
        i++;
        console.log("Returning", i);
        if (i === 3) return { done: true, value: i };
        return { done: false, value: i };
      },
      return() {
        console.log("Closing");
        return { done: true };
      },
    };
  },
};

const [a] = obj;
// Returning 1
// Closing

const [b, c, d] = obj;
// Returning 1
// Returning 2
// Returning 3
// Already reached the end (the last call returned `done: true`),
// so `return` is not called

for (const b of obj) {
  break;
}
// Returning 1
// Closing

The for await...of loop and yield* in async generator functions (but not sync generator functions) are the only ways to interact with async iterables. Using for...of, array spreading, etc. on an async iterable that's not also a sync iterable (i.e. it has [@@asyncIterator]() but no [@@iterator]()) will throw a TypeError: x is not iterable.

Because iteration involves transferring control back and forth between the iterator and the consumer, error handling happens in both ways: how the consumer handles errors thrown by the iterator, and how the iterator handles errors thrown by the consumer. When you are using one of the built-in ways of iteration, the language may also throw errors because the iterable breaks certain invariants. We will describe how built-in syntaxes generate and handle errors, which can be used as a guideline for your own code if you are manually stepping the iterator.

Errors may happen when acquiring the iterator from the iterable. The language invariant enforced here is that the iterable must produce a valid iterator:

• It has a callable [@@iterator]() method.
• The [@@iterator]() method returns an object.
• The object returned by [@@iterator]() has a callable next() method.

When using built-in syntax to initiate iteration on a non-well-formed iterable, a TypeError is thrown.

const nonWellFormedIterable = { [Symbol.iterator]: 1 };
[...nonWellFormedIterable]; // TypeError: nonWellFormedIterable is not iterable
nonWellFormedIterable[Symbol.iterator] = () => 1;
[...nonWellFormedIterable]; // TypeError: [Symbol.iterator]() returned a non-object value
nonWellFormedIterable[Symbol.iterator] = () => ({});
[...nonWellFormedIterable]; // TypeError: nonWellFormedIterable[Symbol.iterator]().next is not a function

For async iterables, if its @@asyncIterator property has value undefined or null, JavaScript falls back to using the @@iterator property instead (and wraps the resulting iterator into an async iterator by forwarding the methods). Otherwise, the @@asyncIterator property must conform to the above invariants too.

This type of errors can be prevented by first validating the iterable before attempting to iterate it. However, it's fairly rare because usually you know the type of the object you are iterating over. If you are receiving this iterable from some other code, you should just let the error propagate to the caller so they know an invalid input was provided.

Most errors happen when stepping the iterator (calling next()). The language invariant enforced here is that the next() method must return an object (for async iterators, an object after awaiting). Otherwise, a TypeError is thrown.

If the invariant is broken or the next() method throws an error (for async iterators, it may also return a rejected promise), the error is progated to the caller. For built-in syntaxes, the iteration in progress is aborted without retrying or cleanup (with the assumption that if the next() method threw the error, then it has cleaned up already). If you are manually calling next(), you may catch the error and retry calling next(), but in general you should assume the iterator is already closed.

If the caller decides to exit iteration for any reason other than the errors in the previous paragraph, such as when it enters an error state in its own code (for example, while handling an invalid value produced by the iterator), it should call the return() method on the iterator, if one exists. This allows the iterator to perform any cleanup. The return() method is only called for premature exits—if next() returns done: true, the return() method is not called, with the assumption that the iterator has already cleaned up.

The return() method might be invalid too! The language also enforces that the return() method must return an object and throws a TypeError otherwise. If the return() method throws an error, the error is propagated to the caller. However, if the return() method is called because the caller encountered an error in its own code, then this error overrides the error thrown by the return() method.

Usually, the caller implements error handling like this:

try {
  for (const value of iterable) {
    // ...
  }
} catch (e) {
  // Handle the error
}

The catch will be able to catch errors thrown when iterable is not a valid iterable, when next() throws an error, when return() throws an error (if the for loop exits early), and when the for loop body throws an error.

Most iterators are implemented with generator functions, so we will demonstrate how generator functions typically handle errors:

function* gen() {
  try {
    yield doSomething();
    yield doSomethingElse();
  } finally {
    cleanup();
  }
}

The lack of catch here causes errors thrown by doSomething() or doSomethingElse() to propagate to the caller of gen. If these errors are caught within the generator function (which is equally advisable), the generator function can decide to continue yielding values or to exit early. However, the finally block is necessary for generators that keep open resources. The finally block is guaranteed to run, either when the last next() is called or when return() is called.

Some built-in syntaxes wrap an iterator into another iterator. They include the iterator produced by Iterator.from(), iterator helpers (map(), filter(), take(), drop(), and flatMap()), yield *, and a hidden wrapper when you use async iteration (for await...of, Array.fromAsync) on sync iterators. The wrapped iterator is then responsible for forwarding errors between the inner iterator and the caller.

• All wrapper iterators directly forward the next() method of the inner iterator, including its return value and thrown errors.
• Wrapper iterators generally directly forward the return() method of the inner iterator. If the return() method doesn't exist on the inner iterator, it returns { done: true, value: undefined } instead. In the case of iterator helpers: if the iterator helper's next() method has not been called, after trying to call return() on the inner iterator, the current iterator always returns { done: true, value: undefined }. This is consistent with generator functions where execution hasn't entered the yield * expression yet.
• yield * is the only built-in syntax that forwards the throw() method of the inner iterator. For information on how yield * forwards the return() and throw() methods, see its own reference.

You can make your own iterables like this:

const myIterable = {
  *[Symbol.iterator]() {
    yield 1;
    yield 2;
    yield 3;
  },
};

console.log([...myIterable]); // [1, 2, 3]

Iterators are stateful by nature. If you don't define it as a generator function (as the example above shows), you would likely want to encapsulate the state in a closure.

function makeIterator(array) {
  let nextIndex = 0;
  return {
    next() {
      return nextIndex < array.length
        ? {
            value: array[nextIndex++],
            done: false,
          }
        : {
            done: true,
          };
    },
  };
}

const it = makeIterator(["yo", "ya"]);

console.log(it.next().value); // 'yo'
console.log(it.next().value); // 'ya'
console.log(it.next().done); // true

State encapsulation can be done with private properties as well.

class SimpleClass {
  #data;

  constructor(data) {
    this.#data = data;
  }

  [Symbol.iterator]() {
    // Use a new index for each iterator. This makes multiple
    // iterations over the iterable safe for non-trivial cases,
    // such as use of break or nested looping over the same iterable.
    let index = 0;

    return {
      // Note: using an arrow function allows `this` to point to the
      // one of `[@@iterator]()` instead of `next()`
      next: () => {
        if (index < this.#data.length) {
          return { value: this.#data[index++], done: false };
        } else {
          return { done: true };
        }
      },
    };
  }
}

const simple = new SimpleClass([1, 2, 3, 4, 5]);

for (const val of simple) {
  console.log(val); // 1 2 3 4 5
}

For example, a String is a built-in iterable object:

const someString = "hi";
console.log(typeof someString[Symbol.iterator]); // "function"

String's default iterator returns the string's code points one by one:

const iterator = someString[Symbol.iterator]();
console.log(`${iterator}`); // "[object String Iterator]"

console.log(iterator.next()); // { value: "h", done: false }
console.log(iterator.next()); // { value: "i", done: false }
console.log(iterator.next()); // { value: undefined, done: true }

You can redefine the iteration behavior by supplying our own @@iterator:

// need to construct a String object explicitly to avoid auto-boxing
const someString = new String("hi");

someString[Symbol.iterator] = function () {
  return {
    // this is the iterator object, returning a single element (the string "bye")
    next() {
      return this._first
        ? { value: "bye", done: (this._first = false) }
        : { done: true };
    },
    _first: true,
  };
};

Notice how redefining @@iterator affects the behavior of built-in constructs that use the iteration protocol:

console.log([...someString]); // ["bye"]
console.log(`${someString}`); // "hi"

• Iterators and generators guide
• function*
• Symbol.iterator
• Iterator