Coroutines and web components

First publication date:

In the we learned what coroutines are and saw some patterns they can help implement. In this article, we will see how coroutines can be used to model web components in a different way, and why you might like it.

Rendering loop

Among other things, coroutines have a few properties that we will use in this short essay:

Introduction example

Consider the following generator:

function* someComponent({$host}) {
    while (true) {
        const {content = ''} = yield;
        $host.textContent = content;
    }
}

It takes a $host DOM element and has a rendering loop. You can wrap this generator with a function that produces a render function:

const createComponent = (generator) => ({$host}) => {
    const gen = generator({$host});
    gen.next(); // we initiate the component by entering inside the rendering loop
    return (state = {}) => gen.next(state);
};

const HelloWorldComponent = createComponent(function* ({$host}) {
    while (true) {
        const {name = ''} = yield;
        $host.textContent = `hello ${name}`;
    }
});

// div is some DOM element
const render = HelloWorldComponent({
    $host: div
});

render({name: 'Laurent'});
render({name: 'Bernadette'});

The power of functions

For now, the rendering loop is a piece of imperative code, but it can use any rendering library you want (react and so on). The first point above says that functions (and therefore coroutines) are very versatile in Javascript. We could easily go back to a known paradigm if we wanted to. For example, we use lit-html to have a declarative view instead of a bunch of imperative code:

import {render, html} from 'lit-element';

const HelloWorldComponent = createComponent(function* ({$host}) {
    while (true) {
        const {name=''} = yield
        const template = html`<p>hello ${name}</p>`;
        render($host, template);
    }
});

you can draw the template into a function:

import {html} from 'lit-element';

const template = ({name = ''} = {}) => html`<p>hello ${name}</p>`;

And compose with a new higher order function:

import {render} from 'lit-element';

const withView = (templateFn) => function* ({$host}) {
    while (true) {
        render($host, templateFn(yield));
    }
};

const HelloWorldComponent = createComponent(withView(template));

All right, we are on familiar ground: our component is now a simple function of the state

({name}) => html`<p>hello ${name}</p>`; 

Maintaining a state

Having an infinite rendering loop to model our component can actually be more interesting than it seems at first: you can have a state in the closure of that loop.

If we first modify the higher-level createComponent function a little to bind the render function to the host element:

const createComponent = (generator) => ({$host}) => {
    const gen = generator({$host});
    gen.next();
    $host.render = (state = {}) => gen.next(state);
    return $host;
};

We can now make the component trigger its own rendering:

const CountClick = createComponent(function *({$host}){
   let clickCount = 0;
   
   $host.addEventListener('click', () => {
       clickCount+=1;
       $host.render();
   });
   
   while(true) {
       $host.textContent = `I have been clicked ${clickCount} time(s)`
       yield;
   }
});

In frameworks like React, where you only have access to the equivalent of what is inside the loop, you rely on the framework extension points (the hooks in the case of React) to build this sort of mechanism, and have very little control over rendering scheduling.

More HOF function to reduce the coupling.

The component embeds its view and some logic at the same time. Again, we can easily decouple them so that we can reuse either the view or the logic: All we need to do is take advantage of the third property of coroutines mentioned in the introduction, and a simple delegation mechanism inherent to generators: yield*.

const countClickable = (view) => function *({$host}) {
    let clickCount = 0;

    $host.addEventListener('click', () => {
        clickCount+=1;
        $host.render({count: clickCount});
    });
    
    yield* view({$host}); 
}

This type of mixin is responsible for holding the state and triggering the rendering of any view. Rendering is left to the view thanks to delegation, while the state is passed whenever the view coroutine is paused and requires a new render:

const CountClick = createComponent(countClickable(function* ({$host}) {
    while (true) {
        const {count = 0} = yield;
        $host.textContent = `I have been clicked ${count} time(s)`;
    }
}));

Neat ! You can now use the “clickable” behaviour independently, on different views. In the same way, you can plug the view into a different controller logic, as long as it passes the expected data interface ({ count: number | string}): note how the data comes from the yield assignation.

We will see more patterns like this in future articles.

Web components and lifecycle mapping

So far we have designed our component to be a function of the host. We can go further and ensure that the rendering routine is actually private to the host, so that the rendering code is encapsulated inside along with any potential behaviour enhancements (the countClickable mixin for example), while both remain reusable.

Let’s look at another way of modelling custom elements. To enhance your HTML document, you can teach the browser new ones using its registry and the define method.

customElements.define('hello-world', class extends HTMLElement {
    connectedCallback() {
        this.textContent = `hello ${this.getAttribute('name')}`
    }
}) // (define takes a third optional argument we won't consider for the moment)

And then use the hello-world tag in the markup like any other regular HTML tag.

<hello-world name="Laurent"></hello-world>

Instead of using a class that extends the HTMLElement class (or any other valid built-in element class), we want the second argument to be a generator function. This means our custom define would need to turn the generator into a class.

const define = (tag, gen) => {
    customElements.define(tag, class extends HTMLElement {
        #loop;

        constructor() {
            super();
            this.#loop = gen.bind(this)({
                $host: this
            });
            this.render = this.render.bind(this);
            this.#loop.next();
        }

        connectedCallback() {
            this.render();
        }

        render(state = {}) {
            this.#loop.next(state);
        }

    });
};

define('hello-world', function* ({$host}) {
    while(true) {
        yield;
        $host.textContent = `hello ${$host.getAttribute('name')}`
    }
});

Using a class expression, we create the custom element class on the fly. The #loop rendering routine is instantiated inside the constructor and advanced to its first yield point. Note that we pass the host as a parameter to the routine, although the routine is specifically bound to the host so that we could just use this inside the generator to refer to the host. This is a personal preference as I find the use of this in Javascript very error-prone.

When the connectedCallback is called (this happens when the component is mounted into the DOM). We call next again, which in our previous example corresponds to the first iteration of the loop. Then, whenever the component needs to be rendered (when render is called) again, we continue the loop.

This is very interesting because we are able to match the different component lifecycles to a location within the generator function:

function* comp({$host}) {

    console.log('I am being instantiated');

    yield;

    console.log('Iam being mounted');
    $host.textContent = 'I have just been mounted';

    yield;

    while (true) {
        console.log('I am being rendered');
        $host.textContent = `hello ${$host.getAttribute('name')}`;
        yield 'I have been rendered';
    }
}

Yet, one important lifecycle remains to be implemented. When the component is unmounted, the disconnectedCallbak of the class definition is normally called, allowing us to run cleanup code and avoid memory leaks for example.

In the generator we can force the exit of the loop into a finally clause. This is as simple as calling the loop’s return function instead of the usual next.

Altogether:

const define = (tag, gen) => {
    customElements.define(tag, class extends HTMLElement {
        #loop;

        constructor() {
            super();
            this.#loop = gen.bind(this)({
                $host: this
            });
            this.render = this.render.bind(this);
            this.#loop.next();
        }

        connectedCallback() {
            this.render();
        }
        
        disconnectedCallback() {
            this.#loop.return();
        }

        render(state = {}) {
            this.#loop.next(state);
        }

    });
};

function* comp({$host}) {

    console.log('I am being instantiated');

    yield;

    console.log('Iam being mounted');
    $host.textContent = 'I have just been mounted'

    yield;

    try {
        while (true) {
            console.log('I am being rendered');
            $host.textContent = `hello ${$host.getAttribute('name')}`;
            yield;
        }
    } finally {
        console.log('cleanup here !!!');
    }
}

This “linear” representation of the component and its lifetime makes things easier to reason about: there are no surprises when a callback or a hook is called, everything is read from top to bottom!

Concurrent updates

Before we conclude, we can illustrate the fourth point mentioned in the introduction: if you try to advance a generator function while it is already advancing, you will get an error. In the component world, this means that concurrent rendering is impossible by design!

This code:

function* comp({$host}) {
    while (true) {
        console.log('I am being rendered');

        $host.render(); // yet I try to render ...

        $host.textContent = `hello ${$host.getAttribute('name')}`;
        yield;
    }
}

will trigger an error Uncaught TypeError: already executing generator.

conclusion

We have seen throughout this article that the functional nature of a generator combined with its intrinsic properties can be useful to build a very flexible and simple abstraction of UI component, with the ability to split behaviour and view into reusable bits, to maintain internal state or to have at reach all component lifecycles in the same place.

In , we will see how we can further improve and optimise our generator-to-class conversion.