@ncjamieson

RxJS: Improving the Static pipe Function

May 21, 2018 • 4 minute read

Pencil sharpening
Photo by Angelina Litvin on Unsplash

My previous article looked at using the static pipe function to compose reusable combinations of operators.

Most of the time, the pipe function’s TypeScript overload signatures will infer the desired type for the returned function. However, sometimes it’s desirable to have a generic type inferred and the current overload signatures will not do that.

Let’s look at why you might want a generically-typed function to be returned and what needs to be changed for that to happen.

Specifically-typed operators

The static pipe function can be used to compose a reusable, debouncing operator that can be used with the changes emitted by input elements:

import { pipe } from "rxjs";
import { debounceTime, distinctUntilChanged } from "rxjs/operators";

export const debounceInput = pipe(
  debounceTime<string>(400),
  distinctUntilChanged()
);

The static pipe function does not have a source observable, so there is nothing from which the first type parameter can be inferred.

If an explicit type parameter is not specified, it will be inferred as {} — an empty interface — because that’s how TypeScript behaves if a type parameter is not specified and cannot be inferred.

If a type parameter is not specified — and {} is inferred — applying the composed operator to a source observable will see the resultant observable inferred as Observable<{}> — losing the source observable’s type.

The type parameter is specified on the debounceTime operator — rather than on the pipe call itself — as the pipe call has more than one type parameter and specifying only the first will see the remaining type parameters inferred as {}. And specifying them all is too tedious.

A string type parameter means that the composed operator can only be used with observable sources that emit string values. Here, that’s unlikely to be a problem, as input elements will be emitting string values when they change.

However, sometimes you don’t want to restrict composed operators to observables of a specific type and that’s when you need pipe to return a generically-typed function.

Generically-typed operators

To synchronise the rendering of changes with the browser, it’s common to use the auditTime operator with the animationFrameScheduler.

To avoid having to specify the arguments in each call, we could compose auditFrame, like this:

import { animationFrameScheduler } from "rxjs";
import { auditTime } from "rxjs/operators";

export const auditFrame = auditTime(0, animationFrameScheduler);

However, for the reasons outlined above, when auditFrame is applied to a source observable, the source observable’s type will be lost.

How can we compose a generically-typed auditFrame? That is, how can we compose an auditFrame function that will have a signature something like this?

import { Observable } from "rxjs";

declare const auditFrame: <T>(source: Observable<T>) => Observable<T>;

If the static pipe function were to return a generically-typed function, we could compose auditFrame like this:

import { animationFrameScheduler, pipe } from "rxjs";
import { auditTime } from "rxjs/operators";

export const auditFrame = pipe(auditTime(0, animationFrameScheduler));

And could then use auditFrame — in a type-safe way — with source observables of any type.

So how can we change pipe to return a generically-typed function?

Overload signatures

There are numerous operators — like auditTime, debounceTime, and distinctUntilChanged— that don’t change emitted values and don’t change the source observable’s type.

It’s simple to add an overload signature that matches their use, as each operator accepts and returns an observable of the same type. Such an overload signature — that returns generically-typed function — would look something like this:

import { MonoTypeOperatorFunction, Observable, UnaryFunction } from "rxjs";

declare module "rxjs/internal/util/pipe" {
  export function pipe<T>(
    ...operators: MonoTypeOperatorFunction<T>[]
  ): <R extends T>(source: Observable<R>) => Observable<R>;
}

With the overload signature declared, whenever pipe is passed one or more operators that deal only with observables of type T, the returned function can be used with any observable of type R — where R extends T.

That means that the auditFrame composed using pipe can be used with source observables of any type, without the source observable’s type being lost. And it also means that debounceInput doesn’t have to have an explicit type parameter specified.

Interestingly, this particular overload signature needs to precede the others. That means you don’t have to wait for the signature to be incorporated into RxJS and can instead perform the declaration merging done in the snippet to add the overload signature to a current codebase — as merged signatures are placed first.

Or you can write your own, as I’ve done here.


If you are wondering whether or not there is a PR for the additional overload signature, there isn’t, but I’ll be opening one once some type-checking issues with the RxJS test suite are resolved.


Nicholas Jamieson’s personal blog.
Mostly articles about RxJS, TypeScript and React.
TwitterGitHubSponsor

© 2020 Nicholas Jamieson All Rights ReservedRSS