Prefer Higher Order Component for Branching Logic

August 9, 2020

Table of Contents

Hooks have taken the React world by storm and they’re here to stay whether you like it or not. I can understand the appeal of their simplicity since it’s hard to be more concise than const [foo, setFoo] = useState(). But there are use cases where I think that higher-order component (HOC) fare better than hooks and one of them is branching logic.

The Contagious Effect of Defensive Coding

Imagine a component a that makes a network request and depending on the result, you either render a loading spinner or an entire subtree of components.

const Foo = (url) => {
  const [data, loading] = useRequest(url);

  if (loading) return <Loading />;
  else return <Subtree data={data} />;
};

Where this style of writing your components falls apart is once you start adding more hooks to Foo.

const Foo = (url) => {
  const [data, loading] = useRequest(url);

  const [state, setState] = useState(data.whatever);

  useSomeEffect(data);

  if (loading) return <Loading />;
  else return <Subtree data={data} />;
};

Both useState and useSomeEffect expect the request to have succeeded and data to be available. But that’s not the case if the request is still loading and/or if it failed. You have two options for handling this: make both hooks fail gracefully in case of missing data or not render the hooks at all. Option 1 is what I see most people reach for instinctively. It’s the safe and easy thing to do. But I consider if !x then else an anti-pattern and the worst thing about it is that it spreads like wildfire. The next person who works on this code will see the defensive coding and conclude that all bets are off and data can be whatever, so they to will litter their code with if !x then else.

Therefore I try to use option 2 as much as possible. But you can’t conditionally return before rendering a hook, it just doesn’t work.

Conditionally Rendering Hooks

What you can do however is conditionally render a component containing all the hooks.

const FooWithHooks = (data) => {
  const [state, setState] = useState(data.whatever)

  useSomeEffect(data)

  return <Subtree data={data} />
}

const Foo = (url) => {
  const [data, loading] = useRequest(url)

  if (loading) return <Loading />
  else return <FooWithHooks data={data}>
}

What’s unsatisfying about these two components is that Foo contains very little code but it’s hardcoded to always render FooWithHooks (and Loading). Luckily HOCs make it trivial to extract this logic.

const withFoo = WrappedComponent = props => {
  const { url } = props

  const [data, loading] = useRequest(url)

  if (loading) return <Loading />
  else return <WrappedComponent data={data}>
}

const Foo = withFoo(FooWithHooks)

I suspect that this pattern isn’t exactly news to many people, since it’s how many libraries used to expose their functionality before hooks (think graphql HOC from Apollo). But I know very well that it can be super attractive to throw out the old and replace it with the new One True Way of doing things. And in many cases, hooks will probably make your life easier. But HOCs are still a really important design pattern and an essential part of every React developer’s toolkit. Right tool for the job and all.