Converting an Angular Webapp to Svelte

Written — Updated

Confidence: Pretty confident in this. Not necessarily the ideal solution.

My company in the process of converting our large web application based on AngularJS 1.x and Angular Material to use Svelte and Tailwind CSS. Here are my experiences so far.

Converting Templates đź”—

Converting Angular components into Svelte is largely a mechanical process. For the most part, each Angular template feature has a direct corollary in Svelte. Some things are simpler and some are more complex but overall it’s pretty easy to do. We don’t use “attribute directives” much which makes things easier.

The main gotcha there is that AngularJS templates silently drop any exceptions that happen. This is convenient for writing the templates but bad for finding and tracling down bugs.

Directives like ng-if="" sort of work in Angular if is undefined, in that the ng-if becomes false. But the Svelte equivalent {#if} throws an error. For now we’re using lodash get in places where we need to and looking forward to Svelte support for optional chaining.

Component Features and Lifecycle đź”—

Svelte slots are much easier to use and reason about than Angular transclude, especially in cases where you don’t want an extra wrapper element around the slot content.

It’s also much easier to reason about the lifecycle of a Svelte component. No need to deal with $onChanges happening before $onInit, or even special handling of changes at all, since it’s all taken care of with Svelte’s $: syntax.

Likewise, $postLink simply turns into either use: directives or bind:this={element} on the relevant DOM node.

Asynchronous Code đź”—

When calling asynchronous code in an Angular controller, you need to make sure that something triggers Angular to update the DOM once the callback or promise is done. Otherwise the DOM may not be updated with your latest changes.

Angular calls this a “digest update,” and it provides various methods of doing this, as well as its own promise implementation which automates performing the updates. But you can still run into strange race conditions where the browser displays stale data depending on if some other unrelated code has caused a digest update to run after the buggy code or not.

Svelte doesn’t have this problem, because the compiler sees where you assign to variables and automatically marks them dirty and schedules an update. (Of course, Svelte has its own gotchas around detecting mutation of variables in ways not obvious to the compiler.)

Watchers đź”—

Much of Angular’s update detection is done by using watchers. A watcher runs an expression, and if the value of that expression has changed, Angular updates the value.

Watchers can become a performance issue because every active watcher needs to be run on every digest to see if a change is required. Angular provides some methods of getting around this, such as prefixing a template expression with :: to indicate that you don’t need a watcher for that expression. But watchers are often unavoidable.

As with asynchronous code, Svelte’s advantage here is that it indicates the need for an update at the place where the associated data is updated, instead of at each place the data is used. Then each template expression of reactive statement is able to check very quickly if it needs to rerender or not.

Sharing Code Between Svelte and AngularJS đź”—

Angular 1.x uses dependency injection to distribute services throughout the codebase.

// Define a service
export default ng.module('configsModule').factory('UserService', function() {
  return {
    doThis: () => ...,
    doThat: () => ...,

// And elsewhere, use it

import configsModule from './configs';
ng.module('anotherModule', [configsModule]).run(function(ConfigsService) {
  // A bundler plugin uses this magic string to set up the
  // data for Angular to inject the services listed in
  // the function arguments.


As you can guess, this doesn’t work for Svelte components since they can’t interact with Angular’s dependency injection. We are converting our own services to be directly importable as ES6 modules:

export function doThis { ... };
export function doThat { ... };

// And elsewhere...
import { doThis } from './configs';

But for third-party Angular packages, we can’t easily do this. Svelte components sometimes need access to things like ui-router to create links to other places in the app, or $mdDialog to show dialogs using the existing system.

Eventually all of these third-party services will be replaced with more modern ones that aren’t dependent on Angular, but for now we created a hack solution by defining a services object in a file. The Angular module-level run function fills that object in with the various services, and then Svelte components can import that object and access the services they need. It’s a horrible hack, but it works fine. Over time, we are converting our Angular services into normal modules that can be imported from anywhere.

import { services as svelteServices } from './svelte-services';
ng.module('mainModule', [...allTheDependentModules]).run(function($mdDialog, $state) {
  Object.assign(services, {
    mdDialog: $mdDialog,
    state: $state,

Direct Interaction Between Svelte and Angular đź”—

A lot of Angular asynchronous code returns objects where the promise lives under the $promise field, so we added a function to wrap regular promises where legacy Angular code interacts with promises returned from Svelte code.

function wrapRegularPromise(p) {
  if(p && p.$promise) {
    return p.$promise;

  return p;

It’s really helpful that Svelte stores are easy to use in plain JS. We can change a state store over completely to Svelte and make the Angular components subscribe to that store as well without needing to maintain and sync 2 copies of the state.

Embedding Svelte inside Angular is pretty easy, for the most part. I wrote a function that would take in a Svelte component and generate an Angular controller class. Just have to repeat the bindings and event declarations. It’s too long to post here but I created a Github Gist with the contents. The class does a few things:

  1. Use $onChanges to pass on property changes to the Svelte component.
  2. In $postLink, instantiate the Svelte component.
  3. Listen on the Svelte component’s events and call the associated Angular & function binding.
  4. Destroy the Svelte component in $onDestroy.

Then to use it, you just create an Angular component like so:

import svelteShim from './svelte-shim.ts';
import MyComponent from './MyComponent.svelte';
export default ng.module('modulename').component('myComponent', {
  controller: svelteShim(MyComponent,
  // Events
    change: 'onChange',
  bindings: {
    class: '@',
    data: '<',
    onChange: '&',

We haven’t tried (and won’t try) embedding Angular code inside Svelte, which means that any Angular component we port to Svelte also requires us to port all the components it uses, directly or indirectly, to Svelte as well. This means that sometimes we have to tackle porting certain components or services earlier than we otherwise would, but in the end it doesn’t make a huge difference.

We haven’t yet tackled the page routing system. That’ll probably be the trickiest part.

Using Tailwind and Angular Material at the same time đź”—

This is a pretty smooth transition despite both frameworks defining a lot of classes.

The biggest problem is when using the Angular Material flex directive. This adds a class called flex to the component, which acts very much like Tailwind’s flex-1 class. But Tailwind also has a flex class which sets display:flex. Elements whose children are not supposed to be laid out via flexbox end up looking strange.

This can be worked around by simply using the flex-1 class instead of the flex directive. Angular Material’s flex=NUMBER directive is still ok to use in markup not yet converted to Tailwind, since it applies a class name that does not overlap with Tailwind.

This does mean that anything with the class flex will also essentially have flex-1 applied, but that can be overridden as needed with additional classes to specify whatever behavior you actually want.

Thanks for reading! If you have any questions or comments, please send me a note on Twitter.