Discussion on: Mutable and immutable useRef semantics with React & TypeScript

[Nicholas Boll]

I think explicitly typing everything makes it harder to read and now you have to understand the nuances of when you should or should not explicitly type.

const foo = 'foo' // 'foo'
const bar: string = 'bar' // string

Not only is the second one harder to read and parse, it actually widened our type.

[Frederic CHAPLIN]

This is a really simple assignation example and I agree with you on this (except for a little typo) . But for function returns, and libs specific types, making a rule of typing explicitly everything WILL help.

Example:

const timerRef : React.MutableRefObject<number | null> = React.useRef<number | undefined>();
//mutable

or

const inputRef: React.RefObject<HTMLInputElement> = React.useRef<HTMLInputElement>(null);
//not mutable

By explicitly typing, you give explicitly the mutability information to other developpers (or to you in a month or two). So you improve readability.

And if you try

const inputRef: React.MutableRefObject<HTMLInputElement> = React.useRef<HTMLInputElement>(null);
//TS2322: Type 'RefObject<HTMLInputElement>' is not assignable to type MutableRefObject<HTMLInputElement>'.

Here, typescript tell you instantly you're making a mistake: "No, it's not mutable!".

I know there are many sources that says you can use implicit types, but if you use them too much, you may lose some typescript gifts.

[Nicholas Boll]

I'd probably argue there should be a useMutableRef and useRef rather than complicated types to communicate intent. I often have these small functions that map to normal functions to more clearly communicate intent:

const mutableRef = useMutableRef(false) // mutable, default assigned
const immutableRef = useRef<HTMLInputElement>(null) // React handles this, no default assigned

/**
 * Alias to `useEffect` that intentionally is only run on mount/unmount
 */
const useMount = (callback?: () => void) => {
  React.useEffect(callback, [])
}

It is even possible to create nice utility functions that make element refs easier to work with:

// util file
function useElementRef<E extends keyof ElementTagNameMap>(element: E) {
  return React.useRef<ElementTagNameMap[E]>(null)
}

// usage
const ref = useElementRef('div') // React.RefObject<HTMLDivElement>

Notice the theme where the Typescript types start to disappear for normal usage? This means you can still get the benefits of Typescript without explicitly using Typescript. Even JavaScript users of your code can benefit. This technique works better for libraries, especially if you have JavaScript users of your library. You can use JSDoc to explicitly type JS code, but that is a pain for non-primitive types.

I say there doesn't need to be a tradeoff between Typescript gifts and expressing intent. If your team only uses Typescript and understands all the types in use, maybe you don't need to spend any extra time communicating intent through functions. But it is very useful for JavaScript users in addition to Typescript users who don't spend time finding out all the nuances of Typescript type differences like useRef. You have to learn something extra either way (type differences or which function to use), but why not communicate intent explicitly through names vs types?

[Frederic CHAPLIN]

Because in this example case Typescript may :

• throw exceptions at compile time
• and give intent to the reader
• without adding more code.

[Dr. Derek Austin 🥳]

I actually never type anything in TypeScript unless I have to, and I consider explicit types to be an antipattern.

In my opinion, it's easy to check VSCode's Intellisense to make sure that the right type was inferred.

In React, for example, I've never had to actually use the FC type or explicitly return JSX.Element; if I write a function component, then TypeScript catches it 100% of the time.

There are definitely certain cases where I type function returns, such as if I'm using a "pseudo enum" (union type of strings) and want to coerce the function return down from string to either "thingOne" | "thingTwo" -- so I do see your point.

Overall, I don't think it's useful for productivity or type safety to explicitly type things when the implicit type was correct, so I try to avoid it.