overreactedby Dan Abramov

Why Isn’t X a Hook?

January 26, 2019

Since the first alpha version of React Hooks was released, there is a question that keeps coming up in discussions: “Why isn’t <some other API> a Hook?”

To remind you, here’s a few things that are Hooks:

But there are some other APIs, like React.memo() and <Context.Provider>, that are not Hooks. Commonly proposed Hook versions of them would be noncompositional or antimodular. This article will help you understand why.

Note: this post is a deep dive for folks who are interested in API discussions. You don’t need to think about any of this to be productive with React!


There are two important properties that we want React APIs to preserve:

  1. Composition: Custom Hooks are largely the reason we’re excited about the Hooks API. We expect people to build their own Hooks very often, and we need to make sure Hooks written by different people don’t conflict. (Aren’t we all spoiled by how components compose cleanly and don’t break each other?)

  2. Debugging: We want the bugs to be easy to find as the application grows. One of React’s best features is that if you see something wrong rendered, you can walk up the tree until you find which component’s prop or state caused the mistake.

These two constraints put together can tell us what can or cannot be a Hook. Let’s try a few examples.


A Real Hook: useState()

Composition

Multiple custom Hooks each calling useState() don’t conflict:

function useMyCustomHook1() {
  const [value, setValue] = useState(0);
  // What happens here, stays here.
}
 
function useMyCustomHook2() {
  const [value, setValue] = useState(0);
  // What happens here, stays here.
}
 
function MyComponent() {
  useMyCustomHook1();
  useMyCustomHook2();
  // ...
}

Adding a new unconditional useState() call is always safe. You don’t need to know anything about other Hooks used by a component to declare a new state variable. You also can’t break other state variables by updating one of them.

Verdict:useState() doesn’t make custom Hooks fragile.

Debugging

Hooks are useful because you can pass values between Hooks:

function useWindowWidth() {
  const [width, setWidth] = useState(window.innerWidth);
  // ...
  return width;
}
 
function useTheme(isMobile) {
  // ...
}
 
function Comment() {
  const width = useWindowWidth();
  const isMobile = width < MOBILE_VIEWPORT;
  const theme = useTheme(isMobile);
  return (
    <section className={theme.comment}>
      {/* ... */}
    </section>
  );
}

But what if we make a mistake? What’s the debugging story?

Let’s say the CSS class we get from theme.comment is wrong. How would we debug this? We can set a breakpoint or a few logs in the body of our component.

Maybe we’d see that theme is wrong but width and isMobile are correct. That would tell us the problem is inside useTheme(). Or perhaps we’d see that width itself is wrong. That would tell us to look into useWindowWidth().

A single look at the intermediate values tells us which of the Hooks at the top level contains the bug. We don’t need to look at all of their implementations.

Then we can “zoom in” on the one that has a bug, and repeat.

This becomes more important if the depth of custom Hook nesting increases. Imagine we have 3 levels of custom Hook nesting, each level using 3 different custom Hooks inside. The difference between looking for a bug in 3 places versus potentially checking 3 + 3×3 + 3×3×3 = 39 places is enormous. Luckily, useState() can’t magically “influence” other Hooks or components. A buggy value returned by it leaves a trail behind it, just like any variable. 🐛

Verdict:useState() doesn’t obscure the cause-effect relationship in our code. We can follow the breadcrumbs directly to the bug.


Not a Hook: useBailout()

As an optimization, components using Hooks can bail out of re-rendering.

One way to do it is to put a React.memo() wrapper around the whole component. It bails out of re-rendering if props are shallowly equal to what we had during the last render. This makes it similar to PureComponent in classes.

React.memo() takes a component and returns a component:

function Button(props) {
  // ...
}
export default React.memo(Button);

But why isn’t it just a Hook?

Whether you call it useShouldComponentUpdate(), usePure(), useSkipRender(), or useBailout(), the proposal tends to look something like this:

function Button({ color }) {
  // ⚠️ Not a real API
  useBailout(prevColor => prevColor !== color, color);
 
  return (
    <button className={'button-' + color}>  
      OK
    </button>
  )
}

There are a few more variations (e.g. a simple usePure() marker) but in broad strokes they have the same flaws.

Composition

Let’s say we try to put useBailout() in two custom Hooks:

function useFriendStatus(friendID) {
  const [isOnline, setIsOnline] = useState(null);
 
  // ⚠️ Not a real API
  useBailout(prevIsOnline => prevIsOnline !== isOnline, isOnline);
 
  useEffect(() => {
    const handleStatusChange = status => setIsOnline(status.isOnline);
    ChatAPI.subscribe(friendID, handleStatusChange);
    return () => ChatAPI.unsubscribe(friendID, handleStatusChange);
  });
 
  return isOnline;
}
 
function useWindowWidth() {
  const [width, setWidth] = useState(window.innerWidth);
  
  // ⚠️ Not a real API
  useBailout(prevWidth => prevWidth !== width, width);
 
  useEffect(() => {
    const handleResize = () => setWidth(window.innerWidth);
    window.addEventListener('resize', handleResize);
    return () => window.removeEventListener('resize', handleResize);
  });
 
  return width;
}

Now what happens if you use them both in the same component?

function ChatThread({ friendID, isTyping }) {
  const width = useWindowWidth();
  const isOnline = useFriendStatus(friendID);
  return (
    <ChatLayout width={width}>
      <FriendStatus isOnline={isOnline} />
      {isTyping && 'Typing...'}
    </ChatLayout>
  );
}

When does it re-render?

If every useBailout() call has the power to skip an update, then updates from useWindowWidth() would be blocked by useFriendStatus(), and vice versa. These Hooks would break each other.

However, if useBailout() was only respected when all calls to it inside a single component “agree” to block an update, our ChatThread would fail to update on changes to the isTyping prop.

Even worse, with these semantics any newly added Hooks to ChatThread would break if they don’t also call useBailout(). Otherwise, they can’t “vote against” the bailout inside useWindowWidth() and useFriendStatus().

Verdict: 🔴 useBailout() breaks composition. Adding it to a Hook breaks state updates in other Hooks. We want the APIs to be antifragile, and this behavior is pretty much the opposite.

Debugging

How does a Hook like useBailout() affect debugging?

We’ll use the same example:

function ChatThread({ friendID, isTyping }) {
  const width = useWindowWidth();
  const isOnline = useFriendStatus(friendID);
  return (
    <ChatLayout width={width}>
      <FriendStatus isOnline={isOnline} />
      {isTyping && 'Typing...'}
    </ChatLayout>
  );
}

Let’s say the Typing... label doesn’t appear when we expect, even though somewhere many layers above the prop is changing. How do we debug it?

Normally, in React you can confidently answer this question by looking up. If ChatThread doesn’t get a new isTyping value, we can open the component that renders <ChatThread isTyping={myVar} /> and check myVar, and so on. At one of these levels, we’ll either find a buggy shouldComponentUpdate() bailout, or an incorrect isTyping value being passed down. One look at each component in the chain is usually enough to locate the source of the problem.

However, if this useBailout() Hook was real, you would never know the reason an update was skipped until you checked every single custom Hook (deeply) used by our ChatThread and components in its owner chain. Since every parent component can also use custom Hooks, this scales terribly.

It’s like if you were looking for a screwdriver in a chest of drawers, and each drawer contained a bunch of smaller chests of drawers, and you don’t know how deep the rabbit hole goes.

Verdict: 🔴 Not only useBailout() Hook breaks composition, but it also vastly increases the number of debugging steps and cognitive load for finding a buggy bailout — in some cases, exponentially.


We just looked at one real Hook, useState(), and a common suggestion that is intentionally not a Hook — useBailout(). We compared them through the prism of Composition and Debugging, and discussed why one of them works and the other one doesn’t.

While there is no “Hook version” of memo() or shouldComponentUpdate(), React does provide a Hook called useMemo(). It serves a similar purpose, but its semantics are different enough to not run into the pitfalls described above.

useBailout() is just one example of something that doesn’t work well as a Hook. But there are a few others — for example, useProvider(), useCatch(), or useSuspense().

Can you see why?

(Whispers: Composition… Debugging…)


Discuss on 𝕏  ·  Edit on GitHub