Andrei Calazans

ReasonML, Flow, and TypeScript function inference

☕️ 3 min read

It’s interesting how in ReasonML we don’t need to worry much about writing types, its inference engine is nicely capable of infering most behavior. Take the following example:

Imagine we want to use a function which takes another function as a callback parameter and calls it on its return.

Plain JavaScript

const myFunction = () => 'some-string';

const useSomeFunction = (anyFunction) => {
  return anyFunction();
}

const result = useSomeFunction(myFunction); 
/*
  * Expect result to be equal to 'some-string',
  * therefore its type should infer to string. 
*/

Based on the JavaScript snippet above, result should always infer its type to the return value of the function passed to useSomeFunction. Now let’s see how the different type checkers for JavaScript behave in this scenario.

TypeScript

const myFunction = () => 'some-string';

const useSomeFunction = <FnType extends (...args: any[]) => any>(anyFunction: FnType): ReturnType<FnType> => {
  return anyFunction();
}

const result = useSomeFunction(myFunction);
// result is correctly infered to `string`

With TypeScript we can no longer just write the function, we must help TS to infer its bits. There are a couple things going on here.

First to be able to infer any function passed as an argument to useSomeFunction we must define a generic, which we called FnType, cast it as an extension of a function which can have any arguments and return anything (...args: any[]) => any. At the end to infer the return, we must get the return value of our generic function FnType by calling it with ReturnType.

Flow

const myFunction = () => 'some-string';

function useSomeFunction<FnType: Function>(anyFunction: FnType): $Call<FnType> {
  return anyFunction();
}

const result = useSomeFunction(myFunction);
// result is correctly infered to `string`

With Flow it’s not so different, a generic must also be defined, but instead of extending it we define its type to be of a Function. Again the return is defined as $Call<FnType>. $Call is an utility helper just like ReturnType which gets the type of that function’s return.

ReasonML

let myFunction = () => "some-string";

let useSomeFunction = (anyFunction) => anyFunction();

/* result is correctly infered to `string` */
let result = useSomeFunction(myFunction);

With ReasonML no type annotation is required, its engine is capable of infering the type based on the use. For example.

It knows anyFunction is a function because we are calling it inside useSomeFunction. If we instead added a number:

let myFunction = () => "some-string";

let useSomeFunction = (anyFunction) => anyFunction + 2;

let result = useSomeFunction(myFunction);

The above code would throw the following error:

This has type:
  unit -> string
But somewhere wanted:
  int

It knows that anyFunction has type of unit -> string but you were doing an addition so you wanted an int.

This ability of infering types based on usage is an amazing feature of ReasonML, it lets you code faster and confidently. It’s no wonder why Developers coming from other JavaScript Static types fall in love with ReasonML, Static typing JS application can be become verbose and troublesome with TypeScript and Flow since one must know multiple helpers to achieve inference correctly.

This post was originally a Tweet, you can find it here: https://twitter.com/Andrei_Calazans/status/1098596699092779008