An experiment to allow different versions of the same web component on one page

WARNING

This is an experimental feature, use it at your own risk and be sure to understand its. No big applications are using scoped-elements yet - so there is no proof yet if it works in production. This page focuses on in depth explanations as it should help foster a discussion on scoping .

Complex Web Component applications are often developed by several teams across organizations. In that scenario it is common that shared component libraries are used by teams to create an homogeneous look and feel or just to avoid creating the same components multiple times, but as those libraries evolve problems between different versions of the same library may appear, as teams may not be able to evolve and update their code at the same velocity. This causes bottlenecks in software delivery that should be managed by the teams and complex build systems, to try and alleviate the problem.

Scoped Custom Element Registries is a proposal that will solve this problem, but until it is ready or a polyfill becomes available, we have to scope custom element tag names if we want to use different versions of those custom elements in our code. This package allows you to forget about how custom elements are defined, registered and scopes their tag names if it is necessary, and avoids the name collision problem.

npm i --save @open-wc/scoped-elements

Use case and demos

Consider the following setup

• Team Blue owns Page A
• Team Green owns Page B
• Team Black owns Feature A & B

1. Everything is good andyour app is live [ code ] with both pages.
2. Team Black releases a new version ( 2.x ) of Feature B which unfortunately needs to be breaking in order to support new use-cases.
3. Team Blue (on Page A ) does not use any of those new use cases and they have a tight deadline to meet so they can not update right now.
4. Team Green (on Page B ) has to deliver an important functionality to your end users but they need to upgrade to Feature B 2.x since it can only be solved with this new version.
5. Since Feature A 1.x & 2.x are both used in the same app, this will lead to nested dependencies, which then will lead to catastrophic failure, and errors [ code ].

Two possible solutions come to mind:

1. Temporarily (!) allow to ship similar source code (most breaking releases are not a total rewrite) and scope them via @open-wc/scoped-elements ; see the "fixed" examplewith-scope [ code ] running with nested dependencies.
2. Synchronizing updates of shared dependencies - e.g. make sure Team Blue & Team Green always use the same version when releasing. This can be a viable solution however it comes with a high organizational overhead and is hard to scale up (for 10+ teams)

Technical explanation of scenario

The simplified app has the following dependencies

• app
  • page-a
    • feature-a 1.x
    • feature-b 1.x
  • page-b
    • feature-a 2.x
    • feature-b 1.x

which leads to the following node_modules tree

├── node_modules
│   ├── feature-a
│   ├── feature-b
│   ├── page-a
│   └── page-b
│       └── node_modules
│           └── feature-a
├── demo-app.js
└── index.html

To demonstrate, we made three demos:

1. before-nesting [ code ] In this demo, everything works fine as Page A and B both are using the same version of Feature A

2. no-scope [ code ] Feature A version 1.x and 2.x are imported via self registering entry points which leads to the following error message, because the feature-a component tries to register multiple times:

   Uncaught DOMException: Failed to execute 'define' on 'CustomElementRegistry': the name "feature-a" has already been used with this registry
     at [...]/node_modules/page-b/node_modules/feature-a/feature-a.js:3:16

3. with-scope [ code ] This example successfully fixes the problem by using createScopedHtml on both Page A and Page B .

createScopedHtml defines your custom element classes and creates a unique scoped tag for each of them if you need to use multiple versions of an element. In case those classes were already defined, it will return the previously defined tag.

Then, the html function provided by the createScopedHtml method transforms the template literal into another one replacing the tags used by the user by the ones defined by the custom elements. Finally, the transformed template literal is going to be processed by lit-html .

<my-button>${this.text}</my-button>

becomes: <my-button-3>${this.text}</my-button-3>

1. Import createScopedHtml from @open-wc/scoped-elements .

   import { createScopedHtml } from '@open-wc/scoped-elements';

2. Import the classes of the components you want to use.

   import MyButton from './MyButton.js';
   import MyPanel from './MyPanel.js';

3. Create the html function and define the tags you want to use for your components.

   const html = createScopedHtml({
     'my-button': MyButton,
     'my-panel': MyPanel,
   });

4. Use your components in your html.

   render() {
     return html`
       <my-panel class="panel">
         <my-button>${this.text}</my-button>
       </my-panel}>
     `;
   }

import { css, LitElement } from 'lit-element';
import { createScopedHtml } from '@open-wc/scoped-elements';
import MyButton from './MyButton.js';
import MyPanel from './MyPanel.js';

const html = createScopedHtml({
  'my-button': MyButton, // <-- binds the element to a tag in your html
  'my-panel': MyPanel,
});

export default class MyElement extends LitElement {
  static get styles() {
    return css`
      .panel {
        padding: 10px;
        background-color: grey;
      }
    `;
  }

  static get properties() {
    return {
      text: String,
    };
  }

  render() {
    return html`
      <my-panel class="panel">
        <my-button>${this.text}</my-button>
      </my-panel}>
    `;
  }
}

1. Components imported via npm MUST NOT be self registering components. If a shared component (installed from npm) does not offer an export to the class alone, without the registration side effect, then this component can not be used. E.g. every component that calls customElement.define .

   export class MyEl { ... }
   customElement.define('my-el', MyEl);

   Or uses the customElement typescript decorator

   @customElement('my-el')
   export class MyEl { ... }

   Only side effects free class exports may be used

   export class MyEl { ... }

2. Every (!!!) component that uses sub components need to use scoped-elements . Any import to a self registering component can result in a browser exception - completely breaking the whole application.

3. Imported elements need to be fully side effect free (not only element registration)

4. Currently, only lit-html is supported (though other rendering engines could be incorporated in the future).

5. You can not use tag selectors in css, but you could use an id, a class name or even a property instead.

   :no_entry_sign: my-panel {
     width: 300px;
   }
   :white_check_mark: .panel {
     width: 300px;
   }

6. You can not use tag names using javascript querySelectors, but you could use an id, a class name or even a property instead.

   :no_entry_sign: this.shadowRoot.querySelector('my-panel');
   :white_check_mark: this.shadowRoot.querySelector('.panel');

7. Using scoped-elements may result in a performance degradation of up to 8%.

8. Loading of duplicate/similar source code (most breaking releases are not a total rewrite) should always be a temporary solution.

9. Often, temporary solutions tend to become more permanent. Be sure to focus on keeping the lifecycle of nested dependencies short.

We are using Tachometer to measure the performance penalty of using the scoped elements feature. The chosen test application is a slight variation of the Polymer Shop Application .

This is an example of the results obtained running the performance test.