Understanding Angular 2 change detection

TL;DR Angular 2 introduces a reinvented change detection system that drops digest cycles in favor of one-way flow. Additionally, change detection can now be controlled and fine-tuned by developers to get the most of the framework's performance.

Introduction to Change Detection

Angular 2 final has been released. You've probably heard about some of the changes due to the major version bump: it's has been completely rewritten, TypeScript was selected as the language of choice, reinvented forms, RxJS, a completely new router, etc. In my opinion, the most valuable thing is the redesign of the core change detection system. As you may remember, the digest loop performance of AngularJS (aka Angular 1) was problematic. Now it's not.

Why Do We Need Change Detection?

Why bother? Generally, the power of modern JavaScript frameworks works something like this: an event changes in the model and forcing a change in the UI. This is change detection, the system that monitors events and acts on them. Something has to trigger this propagation to the view. As mentioned before, in Angular 1 we had digest loops that checked every single reference that was set to be watched for value changes. When Angular found out that everything was stable (no infinite loops, etc.), it propagated changes to the view. Although this was not efficient, it worked for a long time. Also, the problem was tracking asynchronous events. You also probably used

$scope.$apply(...)

How Javascript works

The JavaScript runtime works on a single threaded engine. You've probably heard about the stack (possibly from other programming languages). Let's take the code below:

console.log('Hey')
setTimeout(() => {
   console.log('Hello from timeout!')
}, 1000);
console.log('Hi')

We'll see this in a console as:

Hey
Hi
Hello from timeout!

Moreover, nothing is blocked during the one-second wait period. So how would the JS engine do this with a single thread?

Synchronous code

Let's go step by step. If you have code like this:

console.log('1')
console.log('2')
console.log('3')

every instruction will be put onto the stack and will run one by one. There's no possibility of seeing 3 before 2 or 1. So we'll end up with the following:

1
2
3

Every time. Everywhere.

Asynchronous code

But let's go back to the timeout:

console.log('1')
setTimeout(() => {
  console.log('2')
}, 0)
console.log('3')

What happens now? On the stack, we'll have:

console.log
setTimeout
console.log

The trick here is how

setTimeout

setTimeout

What does it mean? Two things:

• Everything that's inside an asynchronous callback (as in

  setTimeout

  ) will be run after any other synchronous code; this is why hacks like

  setTimeout(() => {}, 0)

  work.
• We have no way of ensuring 1000ms is exactly 1000ms (but we know it's at least 1000ms).

For complete understanding of event loop and what's going on in the browser, I encourage you to take a look at this Philip Roberts talk.

How Zones Relate to Change Detection

How does all of this relate to Angular and change detection? Tracking objects with synchronous code is fairly easy. However, when it comes to asynchronous code, things get complicated. That's because Angular 1 forced us to use

$scope.$apply(...)

$timeout

$http

On the other hand, we now have Angular 2. It dropped all of the stuff connected to digest cycles and now uses Zones. Zones are able to track the context of asynchronous actions by monkey-patching them (i.e., overwriting them with its own code), which then invokes the desired action but with some additional information attached. This additional information is the context. This way, Angular will know which component the asynchronous action was invoked from.

The big win of this approach is that we can use the browser APIs natively, and Angular will know what's going on without forcing us to manually tell it a change has occured. The drawback is that Zones overwrite asynchronous actions, which is kind of a hacky solution and may affect other (existing) code if we're not relying only on Angular in the app.

But exactly how is Angular notified of the change? Angular uses its own version of the Zone called

NgZone

onTurnDone

Make Use of Change Detection In Your App

Everything described above is going on under the hood. Equally important is how we can make use of it. Unlike Angular 1, Angular 2 gives us the possibility of controlling the change detection. However, the Angular team claims that even without any performance tweaking, it's 3 to 10 times faster than the previous one, and for most apps this will be fast enough. But it can be much faster. Let's look at an example.

Here's a very typical problem: rendering a list. There's one component containing a list of other components that have some input data. Generally, we have a container with data and a dumb component just for rendering a single list item. Nothing fancy here, just the getter and

ngOnChange

ngOnChange

rowData

This means the getter is fired by Angular itself. And guess what happens? We have a single change on the input, but there are hundreds of getter logs over there.

Why is that?

Angular is notified about the change from some component and has to check how that affects the current state, so it checks all the values for the change. Actually, the team says it can make thousands of such checks in milliseconds, but it's still a waste of time and can even harm our big data-driven application.

Immutability

The cool thing about the new change detection system is that now we can tune it. Let's take a break from Angular and consider following code:

const users = [{
  name: 'John',
  age: 27
}, {
  name: 'Anna',
  age: 23
}]

users.push({
  name: 'Max',
  age: 30
})

The most important thing here is the

const

users

const

push

Array

const user = {
  name: 'Max',
  age: 30
}

user.age = 31

The same thing applies. Although we can't modify the whole object to make it be another one (reference change), we still can change part of the object!

This is why the checks we discussed before aren't so good. If you want to check whether the object is the same as it was before, you have to deeply check all of its properties. It's not efficient.

How can we force the object to be a new one with the changed property? It's actually quite easy with the new ECMAScript Object spread properties proposal:

const user = {
  name: 'Max',
  age: 30
}

const modifiedUser = { ...user, age: 31 }

Change Detection Strategies

The good part about all this is that now we can say to Angular that we know what we're doing. To modify the change detection behavior, we can use the

ChangeDetectionStrategy

OnPush

Let's add the

OnPush

import {ChangeDetectionStrategy, Component, Input} from '@angular/core';

@Component({
  selector: 'row',
  template: `
    <pre>{{ rowData }}</pre>
  `,
  changeDetection: ChangeDetectionStrategy.OnPush
})
export class RowComponent {
  ...
}

You can try it on Plunker and see the difference.

The huge improvement is that there's now only one getter call for one change! We didn't need anything more as our input data are strings that are being changed, so that reference on input changes. The reference for the rest of the components hasn't changed, so Angular doesn't even look at it.

App Structure

How can we build a highly performing app? With Angular 2, it's actually quite easy. As in all of the component frameworks nowadays, you should have dumb and smart components. The dumb components, which are meant only for displaying data from the input or handling user events, are ideal volunteers for having the

OnPush

OnPush

auth0-spa-js

npm install @auth0/auth0-spa-js

environment.ts

src/environments

CLIENT_DOMAIN

CLIENT_ID

// src/environments/environment.ts

export const environment = {
  production: false,
  auth: {
    CLIENT_DOMAIN: 'YOUR_DOMAIN',
    CLIENT_ID: 'YOUR_CLIENT_ID',
  },
};

export const config = {};

AuthService

ng generate service auth

src/app/auth.service.ts

//src/app/auth.service.ts

import { Injectable } from '@angular/core';
import createAuth0Client from '@auth0/auth0-spa-js';
import Auth0Client from '@auth0/auth0-spa-js/dist/typings/Auth0Client';
import {
  from,
  of,
  Observable,
  BehaviorSubject,
  combineLatest,
  throwError,
} from 'rxjs';
import { tap, catchError, concatMap, shareReplay } from 'rxjs/operators';
import { Router } from '@angular/router';
import { environment } from './../environments/environment';

@Injectable({
  providedIn: 'root',
})
export class AuthService {
  // Create an observable of Auth0 instance of client
  auth0Client$ = (from(
    createAuth0Client({
      domain: environment.auth.CLIENT_DOMAIN,
      client_id: environment.auth.CLIENT_ID,
      redirect_uri: `${window.location.origin}`,
    }),
  ) as Observable<Auth0Client>).pipe(
    shareReplay(1), // Every subscription receives the same shared value
    catchError((err) => throwError(err)),
  );
  // Define observables for SDK methods that return promises by default
  // For each Auth0 SDK method, first ensure the client instance is ready
  // concatMap: Using the client instance, call SDK method; SDK returns a promise
  // from: Convert that resulting promise into an observable
  isAuthenticated$ = this.auth0Client$.pipe(
    concatMap((client: Auth0Client) => from(client.isAuthenticated())),
    tap((res) => (this.loggedIn = res)),
  );
  handleRedirectCallback$ = this.auth0Client$.pipe(
    concatMap((client: Auth0Client) => from(client.handleRedirectCallback())),
  );
  // Create subject and public observable of user profile data
  private userProfileSubject$ = new BehaviorSubject<any>(null);
  userProfile$ = this.userProfileSubject$.asObservable();
  // Create a local property for login status
  loggedIn: boolean = null;

  constructor(private router: Router) {
    // On initial load, check authentication state with authorization server
    // Set up local auth streams if user is already authenticated
    this.localAuthSetup();
    // Handle redirect from Auth0 login
    this.handleAuthCallback();
  }

  // When calling, options can be passed if desired
  // https://auth0.github.io/auth0-spa-js/classes/auth0client.html#getuser
  getUser$(options?): Observable<any> {
    return this.auth0Client$.pipe(
      concatMap((client: Auth0Client) => from(client.getUser(options))),
      tap((user) => this.userProfileSubject$.next(user)),
    );
  }

  private localAuthSetup() {
    // This should only be called on app initialization
    // Set up local authentication streams
    const checkAuth$ = this.isAuthenticated$.pipe(
      concatMap((loggedIn: boolean) => {
        if (loggedIn) {
          // If authenticated, get user and set in app
          // NOTE: you could pass options here if needed
          return this.getUser$();
        }
        // If not authenticated, return stream that emits 'false'
        return of(loggedIn);
      }),
    );
    checkAuth$.subscribe();
  }

  login(redirectPath: string = '/') {
    // A desired redirect path can be passed to login method
    // (e.g., from a route guard)
    // Ensure Auth0 client instance exists
    this.auth0Client$.subscribe((client: Auth0Client) => {
      // Call method to log in
      client.loginWithRedirect({
        redirect_uri: `${window.location.origin}`,
        appState: { target: redirectPath },
      });
    });
  }

  private handleAuthCallback() {
    // Call when app reloads after user logs in with Auth0
    const params = window.location.search;
    if (params.includes('code=') && params.includes('state=')) {
      let targetRoute: string; // Path to redirect to after login processed
      const authComplete$ = this.handleRedirectCallback$.pipe(
        // Have client, now call method to handle auth callback redirect
        tap((cbRes) => {
          // Get and set target redirect route from callback results
          targetRoute =
            cbRes.appState && cbRes.appState.target
              ? cbRes.appState.target
              : '/';
        }),
        concatMap(() => {
          // Redirect callback complete; get user and login status
          return combineLatest([this.getUser$(), this.isAuthenticated$]);
        }),
      );
      // Subscribe to authentication completion observable
      // Response will be an array of user and login status
      authComplete$.subscribe(([user, loggedIn]) => {
        // Redirect to target route after callback processing
        this.router.navigate([targetRoute]);
      });
    }
  }

  logout() {
    // Ensure Auth0 client instance exists
    this.auth0Client$.subscribe((client: Auth0Client) => {
      // Call method to log out
      client.logout({
        client_id: environment.auth.CLIENT_ID,
        returnTo: `${window.location.origin}`,
      });
    });
  }
}

ng generate component login-button

src/app/login-button/login-button.component.ts

//src/app/login-button/login-button.component.ts

import { Component, OnInit } from '@angular/core';
import { AuthService } from '../auth.service';

@Component({
  selector: 'app-login-button',
  templateUrl: './login-button.component.html',
  styleUrls: ['./login-button.component.css'],
})
export class LoginButtonComponent implements OnInit {
  constructor(public auth: AuthService) {}

  ngOnInit() {}
}

src/app/login-button/login-button.component.html

<!-- src/app/login-button/login-button.component.html -->
<div>
  <button (click)="auth.login()" *ngIf="!auth.loggedIn">Log In</button>
  <button (click)="auth.logout()" *ngIf="auth.loggedIn">Log Out</button>
</div>

<app-login-button></app-login-button>

src/app/app.component.html

Conclusions

Performance Advantages

One of the big advantages of using stricter change detection is the performance gains. Angular is meant to be used for large applications that can handle lots of dynamic data. The Angular team has given the developer the tools necessary to fine-tune and improve performance from the get-go. By default, every change should be reflected on the UI, as Angular takes care of that, but the price is lower performance. Immutable or reactive code is harder to write but easier to maintain and reason out. The choice is yours.

Eventually Angular can be tweaked

The good thing is that we have a choice with Angular 2. In Angular 1, it was impossible to get away from the digest cycle. At a certain point, it was advantageous to use React or another library to render UI instead of Angular templates, as it was too slow when handling a large amount of dynamic data. Now, you have a complete solution with much more control over the internal behavior. This, in combination with the other changes made to Angular 2, makes the learning curve of the framework steeper but worth it.