yarn workbox-build 6.0.0-alpha.1
Workbox v6.0.0-alpha.1

Overview of Workbox v6

We're happy to announce the first alpha release of Workbox v6!

🎉 What's New?

workbox-strategies improvements

The best way to ensure third-party developers have the power to extend Workbox in ways that fully meet their needs is to base our own strategies on top of the extensibility mechanisms we expose to third-party developers.

Specifically, this proposal introduces a new way for third-party developers to define their own Workbox strategies, and all of our built-in strategies have been rewritten on top of this mechanism.

This change also allowed us to rewrite the workbox-precaching codebase to use workbox-strategies as a base. This should not result in any breaking changes, and should lead to better long-term consistency in how the two modules access the network and cache.

See #2446, #2459 and #2569 for more details.

New strategy base class

In v6, all Workbox strategy classes (both built-in strategies as well as custom, third-party strategies) must extend the new Strategy base class.

The Strategy base class is responsible for two primary things:

  • Invoking plugin lifecycle callbacks common to all strategy handlers (e.g. when they start, respond, and end).
  • Creating a "handler" instance, that can manage state for each individual request a strategy is handling.

A new "handler" class

We previously had internal modules call fetchWrapper and cacheWrapper, which (as their name implies) wrap the various fetch and cache APIs with hooks into their lifecycle. This is the mechanism that currently allows plugins to work, but it's not exposed to developers.

The new "handler" class (which this proposal calls StrategyHandler) will expose these methods so custom strategies can call fetch() or cacheMatch() and have any plugins that were added to the strategy instance automatically invoked.

This class would also make it possible for developers to add their own custom, lifecycle callbacks that might be specific to their strategies, and they would "just work" with the existing plugin interface.

New plugin lifecycle state

In Workbox v5, plugins are stateless. That means if a request for /index.html triggers both the requestWillFetch and cachedResponseWillBeUsed callbacks, those two callbacks have no way of communicating with each other or even knowing that they were triggered by the same request.

In this proposal, all plugin callbacks will also be passed a new state object. This state object will be unique to this particular plugin object and this particular strategy invocation (i.e. the call to handle()).

This allows developers to write plugins where one callback can conditionally do something based on what another callback in the same plugin did (e.g. compute the time delta between running requestWillFetch and fetchDidSucceed or fetchDidFail).

New plugin lifecycle callbacks

In order to fully leverage the plugin lifecycle state (mentioned above), you need to know when the lifecycle of a given strategy invocation starts and finishes.

To address this need (and others), the following new plugin lifecycle callbacks will be added:

  • handlerWillStart: called before any handler logic starts running. This callback can be used to set the initial handler state (e.g. record the start time).
  • handlerWillRespond: called before the strategies handle() method returns a response. This callback can be used to modify that response before returning it to a route handler or other custom logic.
  • handlerDidRespond: called after the strategy's handle() method returns a response. This callback can be used to record any final response details, e.g. after changes made by other plugins.
  • handlerDidComplete: called after all extend lifetime promises added to the event from the invocation of this strategy have settled. This callback can be used to report on any data that needs to wait until the handler is done in order to calculate (e.g. cache hit status, cache latency, network latency).
  • handlerDidError: called if the handler was unable to provide a valid response from any source. This callback can be used to provide "fallback" content as an alternative to a network error.

Developers implementing their own custom strategies do not have to worry about invoking these callbacks themselves; that's all handled by a new Strategy base class.

More accurate TypeScript types for handlers

TypeScript definitions for various callback methods have been normalized. This should lead to a better experience for developers who use TypeScript and write their own code to implement or call handlers.

See #2548.

workbox-window improvements

New messageSkipWaiting() method

A new method, messageSkipWaiting(), has been added to the workbox-window module to simplify the process of telling the "waiting" service worker to activate.

This offers some improvements over alternatives

  • It calls postMessage() with the de facto standard message body, {type: 'SKIP_WAITING'}, that a service worker generated by Workbox checks for to trigger skipWaiting().

  • It chooses the correct "waiting" service worker to post this message to, even if it's not the same service worker that workbox-window was registered with.

See #2394.

Removal of "external" events in favor of an isExternal property

Many developers were confused by the concept of "external" events in workbox-window, and in practice, they did not end up being a net-positive.

All "external" events are now represented as "normal" events with an isExternal property set to true. This allows developers who care about the distinction to still detect it, and developers who don't need to know can ignore the property.

See #2031.

Cleaner "Offer a page reload for users" recipe

Taken together, these two changes make the "Offer a page reload for users" recipe cleaner:

<script type="module">
import {Workbox} from 'https://storage.googleapis.com/workbox-cdn/releases/6.0.0-alpha.1/workbox-window.prod.mjs';

if ('serviceWorker' in navigator) {
  const wb = new Workbox('/sw.js');

  const showSkipWaitingPrompt = () => {
    // This assumes a hypothetical createUIPrompt() method with
    // onAccept and onReject callbacks:
    const prompt = createUIPrompt({
      onAccept: () => {
        wb.addEventListener('controlling', () => {
          window.location.reload();
        });

        // This will postMessage() to the waiting service worker.
        wb.messageSkipWaiting();
      },

      onReject: () => {
        prompt.dismiss();
      }
    });
  };

  // Listening for externalwaiting is no longer needed.
  wb.addEventListener('waiting', showSkipWaitingPrompt);
  await wb.register();
}
</script>

sameOrigin parameter in matchCallback functions

A new boolean parameter, sameOrigin, is passed to the matchCallback function used in workbox-routing. It's set to true if the request is for a same-origin URL, and false otherwise.

This simplifies some common boilerplate:

// In v5:
registerRoute(
  ({url}) => url.origin === self.location.origin &&
             url.pathname.endsWith('.png'),
  new StaleWhileRevalidate({cacheName: 'local-png'}),
);

// In v6:
registerRoute(
  ({sameOrigin, url}) => sameOrigin &&
                         url.pathname.endsWith('.png'),
  new StaleWhileRevalidate({cacheName: 'local-png'}),
);

See #2487.

matchOptions are supported in workbox-expiration

You can now set matchOptions in workbox-expiration, which will then be passed through as the CacheQueryOptions to the underlying cache.delete() call. (Most developers won't need to do this.)

See #2206.

Precaching now processes entries one by one, not in bulk

workbox-precaching has been updated so that only one entry in the precache manifest is requested and cached at a time, instead of attempting to request and cache all of them at once (leaving it to the browser to figure out how to throttle).

This should reduce the likelihood of net::ERR_INSUFFICIENT_RESOURCES errors while precaching, and also should reduce the bandwidth contention between precaching and simultaneous requests made by the web app.

See #2528.

PrecacheFallbackPlugin allows for easier offline fallback

workbox-precaching now includes a PrecacheFallbackPlugin, which implements the new handlerDidError lifecycle method added in v6.

This makes it easy to specify a precached URL as a "fallback" for a given strategy when a response otherwise wouldn't be available. The plugin will take care of properly constructing the correct cache key for the precached URL, including any revision parameter that's needed.

Here's a sample of using it to respond with a precached /offline.html when the NetworkOnly strategy can't generate a response for a navigation request—in other words, displaying a custom offline HTML page:

import {PrecacheFallbackPlugin, precacheAndRoute} from 'workbox-precaching';
import {registerRoute} from 'workbox-routing';
import {NetworkOnly} from 'workbox-strategies';

// Ensure that /offline.html is part of your precache manifest!
precacheAndRoute(self.__WB_MANIFEST);

registerRoute(
  ({request}) => request.mode === 'navigate',
  new NetworkOnly({
    plugins: [
      new PrecacheFallbackPlugin({
        fallbackURL: '/offline.html',
      }),
    ],
  }),
);

precacheFallback in runtime caching

If you're using generateSW to create a service worker for you instead of writing your service worker by hand, you can use the new precacheFallback configuration option in runtimeCaching to accomplish the same thing:

{
  // ... other generateSW config options...
  runtimeCaching: [{
    urlPattern: ({request}) => request.mode === 'navigate',
    handler: 'NetworkOnly',
    options: {
      precacheFallback: {
        // This URL needs to be included in your precache manifest.
        fallbackURL: '/offline.html',
      },
    },
  }],
}

⚠️ Breaking Changes

Build Tools

  • The minimum required version of node has been increased to v10.0.0. This applies to workbox-build, workbox-cli, and workbox-webpack-plugin. [#2462]

  • mode was not intended to be a supported parameter for the injectManifest and getManifest modes of workbox-build and workbox-cli. It's been removed from the documentation and attempting to use it outside of generateSW will now trigger a build error. This does not apply to workbox-webpack-plugin, which does support mode in its InjectManifest plugin. [#2464]

workbox-core

  • The skipWaiting() method in workbox-core wrapped the underlying call to self.skipWaiting() in an install handler. In practice, this caused undue confusion and offered little value, as it's valid to call self.skipWaiting() outside of an install event. As of v6, Workbox's skipWaiting() will no longer add in an install handler, and is equivalent to just calling self.skipWaiting(). Because of this, developers should migrate to calling self.skipWaiting() directly, and Workbox's skipWaiting() will likely be removed in v7. [#2547]

workbox-precaching

  • While this scenario is uncommon, if you precache a URL that corresponds to an HTTP redirect to an HTML document on a different origin, that cross-origin HTML document can no longer be used to satisfy a navigation request. [#2484]

  • By default, the fbclid URL query parameter is now ignored when looking up a precached response for a given request. [#2532]

workbox-routing

  • The setDefaultHandler() method now takes an optional second parameter corresponding to the HTTP method that it applies to, defaulting to 'GET'. It no longer applies to requests with any HTTP method. If you were using setDefaultHandler() and all of your web app's requests are 'GET', then no changes need to be made. [#2463]

Installation of the latest pre-release version

We are using the next tag in npm for the current pre-release version. To install a given module use, e.g., npm install --save-dev workbox-webpack-plugin@next.

latest releases: 6.3.0, 6.2.4, 6.2.3...
15 months ago