Fix exported types so that they work for react-redux

[marqu3z]

It seems the typing for react redux hook useSelector relies on the types of the equality function to infer the return type:

// @types/react-redux/index.d.ts
export function useSelector<TState, TSelected>(
    selector: (state: TState) => TSelected,
    equalityFn?: (left: TSelected, right: TSelected) => boolean
): TSelected;

Using isEqual as equality function the compiler will infer any as TSelected type because:

declare module 'react-fast-compare' {
  const isEqual: (a: any, b: any) => boolean
  export default isEqual
}

So

const selector = (state: State): DeepObject => {a: { b: [1,2,3] } }
const value = useSelector(selector) // typed correctly as DeepObject
const value = useSelector(selector, isEqual) // typed as any

Declaring isEqual with generic types should solve the issue

declare module 'react-fast-compare' {
  const isEqual: <A, B>(a: A, b: B) => boolean
  export default isEqual
}

[chrisbolin]

thanks @marqu3z! I'll look into this as soon as I can. I might cut a beta release—would you be willing to test that?

[kitten]

@chrisbolin We may be able to retain support as it was before but satisfy typings like these by using: <A = any, B = any>(a: A, b: B) => boolean

[marqu3z]

thanks @marqu3z! I'll look into this as soon as I can. I might cut a beta release—would you be willing to test that?

Sure, no problem

[chrisbolin]

see https://github.com/FormidableLabs/react-fast-compare/pull/62 as well. We may be able to use kitten's <A = any, B = any>(a: A, b: B) => boolean, but we may have to use (a: any, b: any): boolean. Some testing will be needed, and I think we should release a beta and have folks try it.

[chrisbolin]

Important note: in order to fix the types, we need to give up the naive hope that we could ever types that statically do a comparison. e.g.

<Thing = any>(<a: Thing, b: Thing>) => boolean

We've dealt with this earlier, so I want to make sure it doesn't come back to bite us.

[kale-stew]

@chrisbolin the more I'm reading about namespace declaration, the more it seems our types are definitely broken as-is. With the namespace declaration being dead code, I'm not confident it's working the way we think it is?

More info about ambient namespaces / namespacing in general

[cdierkens]

kale-stew moved this from In progress to Done in May 2020 on May 28, 2020

@ryan-roemer this might be closable as well. We may not have had the automation turned on to set this ticket to closed when moving the card on the project board.

[ryan-roemer]

Sounds good!