React Hooks + Three.js

Motivation/Background

With three.js, you need to instantiate several objects each time you use it. You need to create a Scene, a camera, a renderer, some 3D objects, and maybe a canvas. Beyond that, if you need controls, maybe you setup DAT.gui like the examples use.

The renderer needs a reference to the camera and the Scene. You need a reference to the Scene to add any 3D objects you create. For camera controls you need reference to the canvas DOM element. It's up to you how to structure everything cleanly.

In contrast, React devs are now used to the ease and simplicity of using create-react-app to try things out. It would be great to have a React component you could throw some random three.js code into and have it work!

Taking the idea further, what if we could model three.js scene objects like React components and write code like this:

<Scene>
  <cameraControls />
  <Cube h={12} w={12} d={12} />
</Scene>

The pattern I explain below aims to allow this in a very minimally obtrusive way.

What is the end result?

• A component: ThreeJSManager
• A custom hook: useThree

Combining them allows you to create React-ish three.js components without losing any control of three.js

API

ThreeJSManager:

This component takes 3 props:

• getCamera (Function): Function that returns a three.js Camera. Function called with a canvas element
• getRenderer (Function): Function that returns a three.js Renderer. Function called with a canvas element
• getScene (Function): Function that returns a three.js Scene.

The output of these functions, along with a canvas and timer, are made available in a React Context. Any of the components you add the useThree hook to need to be a child of this component.

useThree Custom Hook:

useThree(setup, [destroy])

useThree custom hook relies on the context provided by ThreeJSManager, so it can only be used in components that have ThreeJSManager as an ancestor.

Arguments:

• setup (Function): This function will be called when the component mounts. It gets called with the context value provided by ThreeJSManager. This function is where you setup the 3D objects to use in your component. Whatever you return here will be available to you later as the output of the getEntity function, which is on the object returned by useThree.
• destroy (Function): Optional. This function will be called when the component is unmounted. It gets called with 2 arguments: the context values provided by ThreeJSManager, and a reference to whatever you returned from setup. If the destroy param isn't passed, scene.remove is called with the return value of setup by default. Note: the return value of setup is the same as the output of getEntity described below

Returns: (Object)

useThree returns an object which has all the values from the context provided by ThreeJSManager, and a getEntity function which returns a reference to the return value of the setup function. Whatever is returned from setup is stored internally inside the hook for you to access with getEntity whenever you need to, ie when props change.

How does it work? React Hooks.

There are a few features in React 16.x that make using three.js (or any external library) in a React app a lot cleaner. Those are forwardRef, and some of the new, experimental Hooks: useRef, and most importantly useEffect.

Read the docs above, but in a nutshell what the features allow is the ability to create function components which:

• Use useEffect to configurably run arbitrary JS (such as calling a third party library)
• Use useRef to store a reference to any arbitrary value (such as third-party classes)
• Combine the above with forwardRef to create a component that makes any arbitrary reference available to it's parent

Here is how you could use these hooks to make a Cube functional component:

function Cube = (props) {
  const entityRef = useRef()
  const scene = getScene()

  useEffect(
    () => {
      entityRef.current = createThreeJSCube(scene, props)
      return () => {
        cleanupThreeJSCube(scene, entityRef.current)
      }
    },
    [],
  )

  return null
}

Ignoring for now where the component gets scene from, what we've done is created a React component, which, when mounted, calls createThreeJSCube and stores a reference to the return value, and when unmounted, calls cleanupThreeJSCube. It renders null, so doesn't effect the DOM; it only has side effects. Interesting.

In case you haven't read up on useEffect yet, the 2nd argument is the hook's dependencies, by specifying an empty array, we're indicating this hook doesn't have dependencies and should only be run once. Omitting the argument indicates it should be run on every render, and adding references into the array will cause the hook to run only when the references have changed.

Using this knowledge, we can add a second hook to our Cube component to run some effects when props change. Since we stored the output from our three.js code into entityRef.current, we can now access it from this other hook:

function Cube(props) {
  …
  useEffect(
    () => {
      updateCubeWithProps(entityRef.current, props)
    },
    [props]
  )
}

We now have a React component which adds a 3D object to a three.js scene, alters the object when it gets new props, and cleans itself up when we unmount it. Awesome! Now we just need to make scene available in our component so that it actually works.

forwardRef

Before discussing how we get scene available in our component, let's discuss another newer React feature that will help us setup three.js in the way we need: forwardRef. Remember, before we can even get to adding 3D objects to the scene, we still need to setup our canvas, renderer, camera, and all of that.

Consider the fact that, in three.js, several things need reference to the canvas element. In more vanilla usages this canvas is created by THREE code itself, but we want more control, so we're going to render it from a React component so we can encapsulate resize actions and anything else specific to the canvas in that component. Now we have a problem though, in that, the DOM element is only available in that component. How do we solve this? forwardRef! With forwardRef, we can create a Canvas component, that renders a canvas element, and forwards it's ref to it. So anyone for anyone rendering <Canvas ref={myRef} />, myRef will point to the canvas HTML element itself, not the Canvas React component. Cool!

const Canvas = forwardRef((props, ref) => {
  …
  return (
    <canvas ref={ref} … />
  )
})

Also remember that, from the React docs, refs are not just for DOM element references! We could set and forward a ref to any value.

ThreeJSManager Component

Using the above techniques, we can create a ThreeJSManager component that has refs to everything we need to use three.js: we'll pass it functions that return the camera, renderer, scene objects, and we'll use our Canvas component to reference the canvas DOM element.

However, we'll still need to make these objects available to child components. For this, we'll have ThreeJSManager render a Context.Provider with all these values. Most components will only need scene, but the canvas and camera object will be useful for components that render for example camera controls.

Now with the context Provider setup, we can use the useContext hook to access the scene in our React/three.js components:

function Cube = (props) {
  const context = useContext(ThreeJSContext)
  const { scene } = context
  …
}

In a nutshell, ThreeJSManager abstracts away the base-level three.js tasks you need to do to before you can add 3D objects. Here's how its return value might look:

<ThreeJSContext.Provider
  value={{
    scene,
    camera,
    canvas,
  }}
>
  { props.children }
</ThreeJSContext.Provider>

And here's how we might use it in our app:

<SceneManager
  getCamera={getCamera}
  getRenderer={getRenderer}
  getScene={getScene}
>
  <Ground />
  <Lights />
  <CameraControls />
  <Cube color={Number(`0x${color}`)} />
</SceneManager>

With Ground, Lights, CameraControls, and Cube being components that make use of the useThree hook.

useThree Custom Hook

Let's look at what useThree does:

• Accesses the scene and other three.js objects with useContext
• Initializes a placeholder that will store the 3D object with useRef (entityRef)
• Runs code on mount that instantializes the 3D object, assigns it to entityRef, adds it to the scene, and returns a cleanup function that removes it from scene with useEffect
• Returns an object with a getEntity function that can be used in other effects (such as when props change) to update the 3D object.

Here's the code for our custom hook:

import { ThreeJSContext } from './ThreeJSManager';

const useThree = (setup, destroy) => {
  const entityRef = useRef();
  const context = useContext(ThreeJSContext);

  const getEntity = () => entityRef.current;

  useEffect(
    () => {
      entityRef.current = setup(context);

      return () => {
        if (destroy) {
          return destroy(context, getEntity());
        }
        context.scene.remove(getEntity());
      };
    },
    [],
  );

  return {
    getEntity,
    ...context,
  };
}

Here's how you'd use it to add a simple grid object to the scene:

const Grid = () => {
  useThree(({ scene }) => {
    const grid = new THREE.GridHelper(1000, 100);
    scene.add(grid);

    return grid;
  });

  return null;
};

Notice a few things here:

• Our setup param method signature destructures scene since that's all we care about
• We didn't pass destroy param, so useThree will just call scene.remove with grid, since that's what we returned from setup.
• The component renders null, otherwise React will throw an error.
• We don't care about props changing, so we don't store the return value of useThree (which would give us access to grid object through getEntity).

If we did care about the props changing, we could destructure getEntity from the return value of useThree and use it in another effect that triggers when props change:

const Grid = props => {
  const { color } = props
  const { getEntity } = useThree(…)

  useEffect(
    () => {
      const grid = getEntity()
      grid.material.color.set(color)
    },
    [color],
  )
  …
}

If we wanted to do something specific on unmount, we can pass a destroy function as the 2nd argument. Perhaps our component is complex and has several three.js objects and our setup function returned an object containing all of them:

const ComplexThreeComponent = () => {
  const getEntity = useThree(
    ({ scene }) => {
      …
      return {
        arms,
        body,
        leg,
      }
    },
    ({ scene }, entity) => {
      const { arms, body, leg } = entity
      scene.remove(arms)
      scene.remove(body)
      scene.remove(leg)
    }
  })
  …
};

If we want to setup a camera control, we can create a component that uses camera and canvas in its setup function:

const CameraControls = () => {
  useThree(({ camera, canvas }) => {
    const controls = new OrbitControls(camera, canvas)
    …
  })
  …
}

Animations / requestAnimationFrame

ThreeJSManager has a component state variable called timer which it provides on the context. We can create effects that use this, same as what we've already done for props. Here's how it looks to rotate our simple cube:

const Cube = props => {
  const { getEntity, timer } = useThree(…)

  useEffect(
    () => {
      const cube = getEntity()
      cube.rotation.x += .01
      cube.rotation.z += .01
    },
    [timer],
  )
  …
}

Summary

At a high level what we've done is created React components that don't render anything and just have side effects, in this case side effects all relating to calling the three.js library, but the same concept could be applied to anything.

To manage the framework of side effects we created a component which provides a bunch of objects in a React Context that we can perform our side effects on.

We used React's new experimental hooks feature to separate the concerns of the different side effects, and control when each gets run with a high level of granularity. We have could have achieved similar results with the classic lifecycle methods, but not as declaratively.

Known Limitations

• Currently there's no way to switch the scene outside of rendering a different ThreeJSManager component
• Changing the props for ThreeJSManager doesn't have any effect since it only uses them on mount
• The scene is always rerendered with requestAnimationFrame, it's needed for the props changing on a useThree component to take effect.
• The function passed to requestAnimationFrame doesn't actually trigger the timer effects in our components directly, so profiling the rendering performance could be harder

Caveat

This is mostly an experiment to see what can be done with the new React hooks, but not intended for production-level use, given the "experimental" status of hooks.

Inspiration

• How to use plain Three.js in your React apps