12. Generators, advanced iteration
Asynchronous iteration allow us to iterate over data that comes asynchronously, on-demand. Like, for instance, when we download something chunk-by-chunk over a network. And asynchronous generators make it even more convenient.
Let's see a simple example first, to grasp the syntax, and then review a real-life use case.
Let's recall the topic about iterables.
The idea is that we have an object, such as range here:
let range = {
from: 1,
to: 5
};
...And we'd like to use for..of loop on it, such as for(value of range), to get values from 1 to 5.
In other words, we want to add an iteration ability to the object.
That can be implemented using a special method with the name Symbol.iterator:
for..of construct when the loop is started, and it should return an object with the next method.next() method is invoked for the next value.next() should return a value in the form {done: true/false, value:<loop value>}, where done:true means the end of the loop.Here's an implementation for the iterable range:
let range = {
from: 1,
to: 5,
*!*
[Symbol.iterator]() { // called once, in the beginning of for..of
*/!*
return {
current: this.from,
last: this.to,
*!*
next() { // called every iteration, to get the next value
*/!*
if (this.current <= this.last) {
return { done: false, value: this.current++ };
} else {
return { done: true };
}
}
};
}
};
for(let value of range) {
alert(value); // 1 then 2, then 3, then 4, then 5
}
If anything is unclear, please visit the chapter , it gives all the details about regular iterables.
Asynchronous iteration is needed when values come asynchronously: after setTimeout or another kind of delay.
The most common case is that the object needs to make a network request to deliver the next value, we'll see a real-life example of it a bit later.
To make an object iterable asynchronously:
Symbol.asyncIterator instead of Symbol.iterator.next() method should return a promise (to be fulfilled with the next value).
async keyword handles it, we can simply make async next().for await (let item of iterable) loop.
await word.As a starting example, let's make an iterable range object, similar like the one before, but now it will return values asynchronously, one per second.
All we need to do is to perform a few replacements in the code above:
let range = {
from: 1,
to: 5,
*!*
[Symbol.asyncIterator]() { // (1)
*/!*
return {
current: this.from,
last: this.to,
*!*
async next() { // (2)
*/!*
*!*
// note: we can use "await" inside the async next:
await new Promise(resolve => setTimeout(resolve, 1000)); // (3)
*/!*
if (this.current <= this.last) {
return { done: false, value: this.current++ };
} else {
return { done: true };
}
}
};
}
};
(async () => {
*!*
for await (let value of range) { // (4)
alert(value); // 1,2,3,4,5
}
*/!*
})()
As we can see, the structure is similar to regular iterators:
Symbol.asyncIterator (1).next() method returning a promise (2).next() method doesn't have to be async, it may be a regular method returning a promise, but async allows us to use await, so that's convenient. Here we just delay for a second (3).for await(let value of range) (4), namely add "await" after "for". It calls range[Symbol.asyncIterator]() once, and then its next() for values.Here's a small table with the differences:
| Iterators | Async iterators | |
|---|---|---|
| Object method to provide iterator | Symbol.iterator |
Symbol.asyncIterator |
next() return value is |
any value | Promise |
| to loop, use | for..of |
for await..of |
````warn header="The spread syntax ... doesn't work asynchronously"
Features that require regular, synchronous iterators, don't work with asynchronous ones.
For instance, a spread syntax won't work:
alert( [...range] ); // Error, no Symbol.iterator
That's natural, as it expects to find Symbol.iterator, not Symbol.asyncIterator.
It's also the case for for..of: the syntax without await needs Symbol.iterator.
## Recall generators
Now let's recall generators, as they allow to make iteration code much shorter. Most of the time, when we'd like to make an iterable, we'll use generators.
For sheer simplicity, omitting some important stuff, they are "functions that generate (yield) values". They are explained in detail in the chapter [](info:generators).
Generators are labelled with `function*` (note the star) and use `yield` to generate a value, then we can use `for..of` to loop over them.
This example generates a sequence of values from `start` to `end`:
```js run
function* generateSequence(start, end) {
for (let i = start; i <= end; i++) {
yield i;
}
}
for(let value of generateSequence(1, 5)) {
alert(value); // 1, then 2, then 3, then 4, then 5
}
```
As we already know, to make an object iterable, we should add `Symbol.iterator` to it.
```js
let range = {
from: 1,
to: 5,
*!*
[Symbol.iterator]() {
return <object with next to make range iterable>
}
*/!*
}
```
A common practice for `Symbol.iterator` is to return a generator, it makes the code shorter, as you can see:
```js run
let range = {
from: 1,
to: 5,
*[Symbol.iterator]() { // a shorthand for [Symbol.iterator]: function*()
for(let value = this.from; value <= this.to; value++) {
yield value;
}
}
};
for(let value of range) {
alert(value); // 1, then 2, then 3, then 4, then 5
}
```
Please see the chapter [](info:generators) if you'd like more details.
In regular generators we can't use `await`. All values must come synchronously, as required by the `for..of` construct.
What if we'd like to generate values asynchronously? From network requests, for instance.
Let's switch to asynchronous generators to make it possible.
## Async generators (finally)
For most practical applications, when we'd like to make an object that asynchronously generates a sequence of values, we can use an asynchronous generator.
The syntax is simple: prepend `function*` with `async`. That makes the generator asynchronous.
And then use `for await (...)` to iterate over it, like this:
```js run
*!*async*/!* function* generateSequence(start, end) {
for (let i = start; i <= end; i++) {
*!*
// Wow, can use await!
await new Promise(resolve => setTimeout(resolve, 1000));
*/!*
yield i;
}
}
(async () => {
let generator = generateSequence(1, 5);
for *!*await*/!* (let value of generator) {
alert(value); // 1, then 2, then 3, then 4, then 5 (with delay between)
}
})();
```
As the generator is asynchronous, we can use `await` inside it, rely on promises, perform network requests and so on.
````smart header="Under-the-hood difference"
Technically, if you're an advanced reader who remembers the details about generators, there's an internal difference.
For async generators, the `generator.next()` method is asynchronous, it returns promises.
In a regular generator we'd use `result = generator.next()` to get values. In an async generator, we should add `await`, like this:
```js
result = await generator.next(); // result = {value: ..., done: true/false}
```
That's why async generators work with `for await...of`.
Regular generators can be used as Symbol.iterator to make the iteration code shorter.
Similar to that, async generators can be used as Symbol.asyncIterator to implement the asynchronous iteration.
For instance, we can make the range object generate values asynchronously, once per second, by replacing synchronous Symbol.iterator with asynchronous Symbol.asyncIterator:
let range = {
from: 1,
to: 5,
// this line is same as [Symbol.asyncIterator]: async function*() {
*!*
async *[Symbol.asyncIterator]() {
*/!*
for(let value = this.from; value <= this.to; value++) {
// make a pause between values, wait for something
await new Promise(resolve => setTimeout(resolve, 1000));
yield value;
}
}
};
(async () => {
for *!*await*/!* (let value of range) {
alert(value); // 1, then 2, then 3, then 4, then 5
}
})();
Now values come with a delay of 1 second between them.
Technically, we can add both `Symbol.iterator` and `Symbol.asyncIterator` to the object, so it's both synchronously (`for..of`) and asynchronously (`for await..of`) iterable.
In practice though, that would be a weird thing to do.
So far we've seen basic examples, to gain understanding. Now let's review a real-life use case.
There are many online services that deliver paginated data. For instance, when we need a list of users, a request returns a pre-defined count (e.g. 100 users) - "one page", and provides a URL to the next page.
This pattern is very common. It's not about users, but just about anything.
For instance, GitHub allows us to retrieve commits in the same, paginated fashion:
fetch in the form https://api.github.com/repos/<repo>/commits.Link header.For our code, we'd like to have a simpler way to get commits.
Let's make a function fetchCommits(repo) that gets commits for us, making requests whenever needed. And let it care about all pagination stuff. For us it'll be a simple async iteration for await..of.
So the usage will be like this:
for await (let commit of fetchCommits("username/repository")) {
// process commit
}
Here's such function, implemented as async generator:
async function* fetchCommits(repo) {
let url = `https://api.github.com/repos/${repo}/commits`;
while (url) {
const response = await fetch(url, { // (1)
headers: {'User-Agent': 'Our script'}, // github needs any user-agent header
});
const body = await response.json(); // (2) response is JSON (array of commits)
// (3) the URL of the next page is in the headers, extract it
let nextPage = response.headers.get('Link').match(/<(.*?)>; rel="next"/);
nextPage = nextPage?.[1];
url = nextPage;
for(let commit of body) { // (4) yield commits one by one, until the page ends
yield commit;
}
}
}
More explanations about how it works:
We use the browser fetch method to download the commits.
https://api.github.com/repos/<repo>/commits, and the next page will be in the Link header of the response.fetch method allows us to supply authorization and other headers if needed -- here GitHub requires User-Agent.The commits are returned in JSON format.
We should get the next page URL from the Link header of the response. It has a special format, so we use a regular expression for that (we will learn this feature in Regular expressions).
https://api.github.com/repositories/93253246/commits?page=2. It's generated by GitHub itself.Then we yield the received commits one by one, and when they finish, the next while(url) iteration will trigger, making one more request.
An example of use (shows commit authors in console):
(async () => {
let count = 0;
for await (const commit of fetchCommits('javascript-tutorial/en.javascript.info')) {
console.log(commit.author.login);
if (++count == 100) { // let's stop at 100 commits
break;
}
}
})();
// Note: If you are running this in an external sandbox, you'll need to paste here the function fetchCommits described above
That's just what we wanted.
The internal mechanics of paginated requests is invisible from the outside. For us it's just an async generator that returns commits.
Regular iterators and generators work fine with the data that doesn't take time to generate.
When we expect the data to come asynchronously, with delays, their async counterparts can be used, and for await..of instead of for..of.
Syntax differences between async and regular iterators:
| Iterable | Async Iterable | |
|---|---|---|
| Method to provide iterator | Symbol.iterator |
Symbol.asyncIterator |
next() return value is |
{value:…, done: true/false} |
Promise that resolves to {value:…, done: true/false} |
Syntax differences between async and regular generators:
| Generators | Async generators | |
|---|---|---|
| Declaration | function* |
async function* |
next() return value is |
{value:…, done: true/false} |
Promise that resolves to {value:…, done: true/false} |
In web-development we often meet streams of data, when it flows chunk-by-chunk. For instance, downloading or uploading a big file.
We can use async generators to process such data. It's also noteworthy that in some environments, like in browsers, there's also another API called Streams, that provides special interfaces to work with such streams, to transform the data and to pass it from one stream to another (e.g. download from one place and immediately send elsewhere).