Glimmer Components

Glimmer components have some huge benefits:

• These new components give you all the benefits described in Native Classes above
• They don't extend from EmberObject at all, which means that they don't need EmberObject APIs, such as reopenClass, extend. You can safely use constructor for all setup code.
• Lifecycle hooks are greatly simplified and easier to use
• They don't have that wrapping HTML element that got in the way of CSS styling and layout

Arguments are also namespaced on this.args within Glimmer components, which is an immutable object. This means that:

• It's clear when you are accessing arguments passed to the component, and when you are accessing fields and properties of the component itself.
• Arguments always refer to the original value that was passed in, so you don't have to track down confusing code in hooks or computed property definitions that modifies the value of the argument.
• There is no confusing two-way data binding for arguments via the component class, data can only flow in one direction.

These components have 2 lifecycle hooks:

• constructor
• willDestroy

These can be used to setup the class and tear it down, respectively. Other lifecycle hooks, like didInsertElement and didUpdate don't have equivalents. Instead, you should use modifiers to fill their use cases. These are discussed in more detail later on.

Components also have 3 properties:

• args - the arguments that the component receives when invoked. These are passed to and assigned in the constructor, so they're available then for any setup code that is needed.
• isDestroying - Set to true when the component has been marked for destruction.
• isDestroyed - Set to true when the component has been fully destroyed.

These components don't have a wrapping element. This is referred to as outer HTML semantics, and it means that whatever you see in the template is what you get in the final rendered DOM:

{{!-- template.hbs --}}
<div>
  Hello, {{this.worldName}}!
</div>

<!-- rendered -->
<div>
  Hello, Earth!
</div>

This means that you no longer have to customize your component using any of the following APIs:

• tagName
• classNames
• classNameBindings
• attributeBindings

Instead, you can do these directly in your template. Here are some before and after examples of each API, converted from classic components:

• tagName

  Before:

  import Component from '@ember/component';

  export default Component.extend({
    tagName: 'button',
    text: 'Hello, world!',

    click() {
      console.log('Hello, world!');
    }
  });

  {{this.text}}

After:

  import Component from '@glimmer/component';
  import { action } from '@ember/object';

  export default class HelloButtonComponent extends Component {
    text = 'Hello, world!';

    @action
    sayHello() {
      console.log('Hello, world!');
    }
  }

    <button {{on "click" this.sayHello}}>
      {{this.text}}
    </button>

• classNames

  Before:

  import Component from '@ember/component';

  export default Component.extend({
    classNames: ['hello-world'],
    text: 'Hello, world!'
  });

  {{this.text}}

After:

  import Component from '@glimmer/component';

  export default class HelloComponent extends Component {
    text = 'Hello, world!';
  }

    <div class="hello-world">
      {{this.text}}
    </div>

• classNameBindings

  Before:

  import Component from '@ember/component';

  export default Component.extend({
    classNameBindings: ['darkMode:dark-mode'],
    darkMode: false,
    text: 'Hello, world!'
  });

    {{this.text}}

After:

  import Component from '@glimmer/component';
  import { tracked } from '@glimmer/tracking';

  export default class HelloComponent extends Component {
    text = 'Hello, world!';
    @tracked darkMode = false;
  }

    <div class={{if this.darkMode "dark-mode"}}>
      {{this.text}}
    </div>

• attributeBindings

  Before:

  import Component from '@ember/component';

  export default Component.extend({
    attributeBindings: ['role'],
    role: 'button',
    text: 'Hello, world!'
  });

    {{this.text}}

After:

  import Component from '@glimmer/component';

  export default class HelloComponent extends Component {
    text = 'Hello, world!';
    role = 'button';
  }

    <div role={{this.role}}>
      {{this.text}}
    </div>

To sum it up, the new mental model is that the "wrapping" element is just like any other element in your template, and you interact with it in exactly the same way. This means that when converting a classic component, you will need to add the wrapping element that was there previously to the template (unless it was a tagless component, e.g. tagName: '').

When you pass standard HTML attributes to a component (like class, alt, role, etc), you need to tell the template where to put them. Remember, there's no wrapping element anymore! The way you show where to apply the attributes is by using ...attributes in the template.

For example, here we pass a class to a component:

<MyComponent class="my-class" />

And in that component, we can apply the class to the paragraph using ...attributes:

{{!--
  The paragraph gets the attributes, and not the h1
--}}
<h1>
  Hello, world!
</h1>
<p ...attributes>
  Lorem Ipsum...
</p>

Attributes can be applied to multiple elements as well:

{{!-- Both elements get the attributes --}}
<h1 ...attributes>
  Hello, world!
</h1>
<p ...attributes>
  Lorem Ipsum...
</p>

You can apply ...attributes to elements that have explicit attributes as well. If ...attributes comes after another attribute, then it'll be possible for the user to override them:

<p
  data-overridable="you can override me"
  ...attributes
  data-non-overridable="but you can't override me!"
>
  ...
</p>

Finally, if you don't apply ...attributes to any elements, then Ember will throw an error if someone tries to use attributes when invoking your component:

{{!-- components/uncustomizable-button.hbs --}}
<button class="btn">Do a thing!</button>

{{!-- This throws an error --}}
<UncustomizableButton class="customized-button-class"/>

Attributes are also available to classic components, and ...attributes is applied automatically to the wrapping element. If you're converting a component from classic components, you should be sure to add ...attributes to the wrapper element.

Before:

import Component from '@ember/component';

export default Component.extend({
  text: 'Hello, world!'
});

{{this.text}}

After:

import Component from '@glimmer/component';

export default class HelloComponent extends Component {
  text = 'Hello, world!';
}

<div ...attributes>
  {{this.text}}
</div>

In class components, arguments are assigned directly to the class instance. This has caused a lot of issues over the years, from methods and actions being overwritten, to unclear code where the difference between internal class values and arguments is hard to reason about.

New components solve this by placing all arguments in an object available as the args property.

Before:

import Component from '@ember/component';
import { computed } from '@ember/object';

export default Component.extend({
  width: 0,
  height: 0,

  aspectRatio: computed('width', 'height', function() {
    return this.width / this.height;
  })
});

{{!-- Usage --}}
<Image @width="1920" @height="1080" />

After:

import Component from '@glimmer/component';

export default class ImageComponent extends Component {
  get aspectRatio() {
    return this.args.width / this.args.height;
  }
}

{{!-- Usage --}}
<Image @width="1920" @height="1080" />

args and its values are automatically tracked, so there is no need to annotate them, the aspectRatio getter will invalidate properly when they change and the component will rerender (if aspectRatio is used in the template).

Additionally, args is not mutable, and is frozen in development modes. This is partially to prevent folks from trying to accomplish two-way bindings (which doesn't work, this is discussed in more detail below) and partially to ensure that args always stays in sync with the arguments passed to the component, so it can be the canonical "single source of truth". If you want to provide defaults to an argument, you should use a getter.

Before:

import Component from '@ember/component';
import { computed } from '@ember/object';

export default Component.extend({
  width: 0,
  height: 0,

  aspectRatio: computed('width', 'height', function() {
    return this.width / this.height;
  })
});

After:

import Component from '@glimmer/component';

export default class ImageComponent extends Component {
  get width() {
    return this.args.width ?? 0;
  }

  get height() {
    return this.args.height ?? 0;
  }

  get aspectRatio() {
    return this.width / this.height;
  }
}

Classic component's arguments are two-way bound. This means that when you set a value in the component, it also changes the value in the parent component:

// components/parent.js
import Component from '@ember/component';

export default Component.extend({
  value: 'Hello, world!'
});

{{!-- templates/components/parent.hbs --}}
<Child @value={{this.value}} />

// components/child.js
import Component from '@ember/component';

export default Component.extend({
  click() {
    this.set('value', 'Hello, moon!');
  }
});

{{!-- templates/components/child.hbs --}}
<button>
  Change value
</button>

In this setup, when we click the child component's button, it'll update the value in both the child component and the parent component. This feature led to many problematic data patterns in classic components, where mutations would occur seemingly randomly. It was hard to figure out what was causing changes, and to debug them.

For Glimmer components, arguments are one-way bound. There is no way to directly mutate a value on a parent component from the child component, even if it is passed as an argument. Instead, you must send an action upward to mutate the value:

// components/parent.js
import Component from '@glimmer/component';

export default class ParentComponent extends Component {
  value = 'Hello, world!';

  @action
  updateValue(newValue) {
    this.value = newValue;
  }
}

{{!-- templates/components/parent.hbs --}}
<Child @value={{this.value}} @onClick={{this.updateValue}} />

// components/child.js
import Component from '@ember/component';

export default class ChildComponent extends Component {}

{{!-- templates/components/child.hbs --}}
<button {{on "click" (fn @onClick 'Hello, moon!')}}>
  Change value
</button>

In our new setup, the parent component has an action which sets the new value. We pass this action to the child component, and the child component directly assigns it to the click of the button, using the {{on}} modifier. It also passes the value we want to call the @onClick using the fn helper. We don't need any additional logic in the child class itself - in fact, this could become a template-only component at this point.

This pattern is known as Data-Down, Actions Up, or unidirectional data flow. For these new components, this pattern is enforced - all mutations must occur through actions. This clarifies the data flow, because it's immediately possible to see where all of the mutations are occurring.

As we mentioned above, components only have two lifecycle hooks, constructor and willDestroy. There were a number of other lifecycle hooks that existed on classic components which were generally related to updating component state or DOM manipulation:

• willInsertElement
• didInsertElement
• willDestroyElement
• didDestroyElement
• willRender
• didRender
• willUpdate
• didUpdate
• didReceiveAttrs
• didUpdateAttrs

These can generally be replaced either by using getters, in cases where they are related to updating component state, or by using modifiers. For example, installing the @ember/render-modifiers addon will give you the ability to use {{did-insert}} and {{did-update}}. You can also write your own modifiers! Keep reading below to learn more.

If you previously did something like this in your didReceiveAttrs or didUpdateAttrs hooks:

import Component from '@ember/component';

export default Component.extend({
  didUpdateAttrs() {
    this._super(...arguments);

    if (this.disabled) {
      // clear input value
      this.set('value', '');
    }
  },

  @action
  updateValue(newValue) {
    this.set('value', newValue);

    if (this.onChange) {
      this.onChange(newValue);
    }
  }
});

You can instead model this through getters and setters, deriving the value from the state of your component:

import Component from '@glimmer/component';
import { tracked } from '@glimmer/tracking';

export default class TextComponent extends Component {
  @tracked _value;

  get value() {
    if (this.args.disabled) {
      return (this._value = '');
    }

    return this._value;
  }

  @action
  updateValue(newValue) {
    this._value = newValue;

    if (this.args.onChange) {
      this.args.onChange(newValue);
    }
  }
}

You'll notice that this getter is mutating the value when the Text component is disabled. If this feels like a code smell to you, it probably is, and is a sign that we're managing state at the wrong level. In this case, for instance, we should instead consider converting the text component to be a stateless component, and mutate the value in the same place where the disabled is set: The Parent component.

// components/form.js
import Component from '@glimmer/component';
import { tracked } from '@glimmer/tracking';

export default class FormComponent extends Component {
  @tracked text;
  @tracked disabled;

  @action
  updateText(text) {
    this.text = text;
  }

  @action
  updateDisabled(disabled) {
    this.disabled = disabled;

    if (disabled) {
      this.text = '';
    }
  }
}

{{!-- templates/components/form.hbs --}}
<Text
  @value={{this.text}}
  @disabled={{this.disabled}}
  @onChange={{this.updateText}}
/>
<button {{on "click" (fn this.updateDisabled (not this.disabled))}}>
  Toggle Disabled
</button>

import Component from '@glimmer/component';

export default class TextComponent extends Component {
  @action
  updateValue(newValue) {
    if (this.args.onChange) {
      this.args.onChange(newValue);
    }
  }
}

Now the Text component doesn't have any internal state, it defers to the parent Form component, and when the Form component toggles its disabled state, it clears the state of the text. The mutation of state is centralized in the action where it occurs, making our program easier to reason about as a whole.

In cases when you were using the hooks to manipulate the DOM, you can instead update to use modifiers. For instance, let's say you were adding an event listener to the element in your component's didInsertElement hook, and removing it in willDestroyElement:

import Component from '@ember/component';

export default Component.extend({
  didInsertElement() {
    this._super(...arguments);

    this.listener = e => {
      this.set('scrollOffset', e.clientY);
    };

    this.element.addEventListener(`scroll`, this.listener);
  },

  willDestroyElement() {
    this.element.removeEventListener(`scroll`, this.listener);

    this._super(...arguments);
  }
});

This could be rewritten using the {{did-insert}} and {{will-destroy}} modifiers, if you install ember-render-modifiers in your app:

import Component from '@glimmer/component';
import { tracked } from '@glimmer/tracking';
import { action } from '@ember/object';

export default class ScrollComponent extends Component {
  @tracked scrollOffset;

  @action
  listener(e) {
    this.scrollOffset = e.clientY;
  }

  @action
  registerListener(element) {
    element.addEventListener('scroll', this.listener);
  }

  @action
  unregisterListener(element) {
    element.removeEventListener('scroll', this.listener);
  }
}

<div
  {{did-insert this.registerListener}}
  {{will-destroy this.unregisterListener}}
>
  ...
</div>

These modifiers run the function passed to them when the element they are applied to is inserted into or removed from the DOM. This makes the hooks explicit in the element they are acting on. There is also a did-update modifier, which does not run on insertion, but runs whenever any of its passed values change, allowing you to update the element:

import Component from '@glimmer/component';
import { tracked } from '@glimmer/tracking';
import { action } from '@ember/object';

export default class ScrollComponent extends Component {
  @action
  setColor(element, color) {
    element.style.color = color;
  }
}

<div
  {{did-insert this.setColor @color}}
  {{did-update this.setColor @color}}
>
  ...
</div>

These three modifiers are basic modifiers that allow you to cover most of the functionality that lifecycle hooks contained.

There are also community APIs available for writing your own modifiers, such as ember-modifier. Ember itself has low level APIs known as modifier managers which can be used to write these higher level APIs. In general, it's recommended to use a community addon to write modifiers, and not to write your own modifier manager.

Let's see what our first example would look like if we were to write it as a modifier using ember-modifier:

import { modifier } from 'ember-modifier';

export default modifier((element, [eventName, listener]) => {
  element.addEventListener(eventName, listener);

  return () => element.removeEventListener(eventName, listener);
});

import Component from '@glimmer/component';
import { tracked } from '@glimmer/tracking';
import { action } from '@ember/object';

export default class ScrollComponent extends Component {
  @tracked scrollOffset;

  @action
  listener(e) {
    this.scrollOffset = e.clientY;
  }
}

<div {{add-event-listener "scroll" this.listener}}>
  ...
</div>

This modifier generalizes the functionality that the component implemented using lifecycle hooks before, so we can use this modifier whenever we need to in any component. This is a much better solution than manually managing event listeners every time we need one! At this point, the modifier is effectively the same as the {{on}} modifier as well, so we could get rid of it altogether and replace it with on:

<div {{on "scroll" this.listener}}>
  ...
</div>

In Octane, template-only components have only an hbs file and no JavaScript file. Behind the scenes, template-only components inherit from '@glimmer/component'.

They can be thought of as functional components, in the sense that their output (the rendered template) is a pure function of their inputs (their arguments). The fact that they can't have state makes them much easier to reason about in general, and less prone to errors.

Template-only components have no backing class instance, so this in their templates is null. This means that you can only reference passed in arguments via named argument syntax (e.g. {{@arg}}):

{{!--
  This does not work, since `this` does not exist
--}}
<label for="title">Title</label>
<Input @value={{this.value}} id="title" />

Additionally, the mut helper generally can't be used for the same reason:

{{!-- This does not work --}}
<input
  value={{this.value}}
  onkeyup={{action (mut this.value) target="value"}}
/>