Association and control

If you've worked with form elements before, you may be familiar with the way that labels can be associated with an input element using the "for" attribute:

<input type="checkbox" id="controlled">
<label for="controlled">Label goes here</label>

Labels associated with a form element in this way have several useful side effects. Clicking on the label will (in this case) toggle the checkbox. It also tells screen readers how to describe the input. There's a similar pattern with a lot of ARIA attributes, like "aria-labelledby" or "aria-activedescendant".

Being able to link two elements together in a declarative way turns out to be a surprisingly useful pattern, especially if they're not particularly close to each other in the page. We might want to create a floating play button for a media file, for example, and be able to change which audio or video file it controls on the fly.

Unfortunately, browsers don't offer a great way to automatically associate two elements together, and there are lots of scenarios where a naive implementation (say, running a query from the attributeChangedCallback()) will run into problems:

Not all of these scenarios matter for every purpose — if the initiating action usually comes from the controlling element, it may be sufficient to search for the target on each call — but if your goal is a more comprehensive relationship between the two, you need something more robust. In lieu of a built-in watchSelector(), we'll have to build our own.

Keeping an eye on IDs

To create a reliable link between an element based on ID, we'll use a Mutation Observer to track when nodes and attributes change. We can create one observer for the top of the document, and then allow elements (or other script functions) to register for updates when a matching element appears or is removed.

Since multiple elements might watch the same ID, and we want to reduce the number of lookups that we perform, we'll store callback registrations in a map indexed by ID when someone calls watchID(). A matching unwatchID() function removes those registrations, and clears the entry entirely if no watchers are still registered.

var watchList = new Map();

var watchID = function(id, callback) {
  var watch = watchList.get(id) || { id, callbacks: [] };
  // no duplicate callbacks allowed
  if (watch.callbacks.includes(callback)) return;
  watch.callbacks.push(callback);
  try {
    glance(watch);
    watchList.set(id, watch);
  } catch (err) {
    console.error(err);
  }
};

var unwatchID = function(id, callback) {
  var watching = watchList.get(id);
  if (!watching) return;
  watching.callbacks = watching.callbacks.filter(c => c != callback);
  if (!watching.callbacks.length) {
    watchList.delete(id);
  }
};

The glance() function called by watch() checks each ID in the map, and notifies the callback function if the value has changed. Since it's called whenever a new callback is added, we cache each ID's previous value on the callback itself, so that we won't trigger extra notifications.

var glance = function(watch) {
  var result = document.getElementById(watch.id);
  watch.callbacks.forEach(function(c) {
    if (c.previous == result) return;
    c(result);
    c.previous = result;
  })
};

Finally, the piece that ties it all together is the Mutation Observer itself, which tracks additions, removals, and changes to the document. On any changes, we check each ID and notify any callbacks if the located element has changed.

var observer = new MutationObserver(function(mutations) {
  watchList.forEach(glance);
});

observer.observe(document.body, {
  subtree: true,
  childList: true,
  attributeFilter: ["id"]
});

Creating our element assocation

For demonstration purposes, we'll recreate the click functionality of the label element. To correctly associate a control with the custom element, we'll divide the work into two parts: registering a watch in attributeChangedCallback(), and a second function to be notified if the controlled element changes.

class ClickLabel extends HTMLElement {
  constructor() {
    super();
    // reference to the controlled element, if any
    this.control = null;
    // bind the assocation callback to this instance
    this.associate = this.associate.bind(this);
    this.addEventListener("click", () => this.onClick());
  }

  static get observedAttributes() {
    return ["for"]
  }

  attributeChangedCallback(attr, was, value) {
    switch (attr) {
      case "for":
        // remove any existing registration
        if (was) unwatchID(was, this.associate);
        this.control = null;
        // register new assocation with an ID
        watchID(value, this.associate);
      break;
    }
  }

  associate(target) {
    // this callback will receive any matching element for our ID
    this.control = target;
  }

  onClick() {
    // if we have an association, click the controlled element
    if (this.control) {
      this.control.click();
    }
  }
}

With the watch functionality in place, our component should be able to handle all different permutations and orderings of assocation.

It's worth asking whether the effort required to make this pattern work is worth it. Why not simply provide a getter/setter for the control property directly, instead of wandering through the DOM via an ID string? Which is certainly a valid question.

The more advanced we are as developers, the more we often think about JavaScript deployment in terms of single-page apps and bundled code. But part of the advantage of web components is that they are self-sufficient: once defined, a custom element handles its own initialization and lifecycle. In document-oriented web development, being able to generate a relationship between two elements solely from an HTML template adds versatility.

I also think there's also something to be said for designing libraries that can be used not just by relatively experienced JavaScript developers, but also by people who have only learned their way around HTML. A setter property requires someone to understand how to find an element in the DOM, how to find our element, and pass a reference from one to the other, whereas a string ID is something much easier to understand.

When we bemoan how complicated web development has become, I believe part of what we mourn is the simple pleasure we felt when we could hook HTML together by hand and watch it work by magic. Declarative interfaces like this undoubtedly require more time and energy on our end, but there is value there nonetheless.