insertAdjacentHTML() Enables Faster HTML Snippet Injection

The following is a guest post by Henri Sivonen:

In Firefox 8, we’ve added support for insertAdjacentHTML(). It’s an ancient feature of Internet Explorer that has recently been formalized in HTML5 and then spun out into the DOM Parsing specification. The bad news is that Firefox is the last major browser to implement this feature. The good news is that since other major browsers implement it already, you can start using it unconditionally as soon as the Firefox 8 update has been rolled out to users.

Basic Usage

insertAdjacentHTML(position, markup) is a method on HTMLElement DOM nodes. It takes two string arguments. The first argument is one of "beforebegin", "afterbegin", "beforeend", or "afterend" and gives the insertion point relative to the node that insertAdjacentHTML() is invoked on. The second argument is a string containing HTML markup that gets parsed as an HTML fragment (similar to a string assigned to innerHTML) and inserted to the position given by the first argument.

If the node that insertAdjacentHTML() is invoked on is a p element
with the text content “foo”, the insertion points would be where the comments are in the following snippet:

<!-- beforebegin --><p><!-- afterbegin -->foo<!--
beforeend --></p><!-- afterend -->

The "beforebegin" and "afterend" positions work only if
the node is in a tree and has an element parent.

For example, consider the following code:

<div id=container><p id=para>foo</p></div>

This code produces this log output:

<p id="para">foo<span>bar</span></p>

Well, that does not look particularly special. In fact, it looks like something that could have been done using plain old innerHTML. So why bother with insertAdjacentHTML() when element.innerHTML += "markup"; already works?

There are two reasons.

  • insertAdjacentHTML() doesn’t corrupt what’s already in the DOM.
  • insertAdjacentHTML() is faster.

Avoiding DOM corruption

Let’s consider the DOM corruption issue first. When you do element.innerHTML += "markup";, the browser does the following:

  1. It gets the value of innerHTML by serializing the descendants of element.
  2. It appends the right hand side of += to the string.
  3. It removes the children of element.
  4. It parses the new string that contains the serialization of the old descendants followed by some new markup.

The old descendants might have been script-inserted to form a subtree that doesn’t round-trip when serialized as HTML and reparsed. In that case, after the operation, the tree would have a different shape even for the “old” parts. (For example, if element had a p child which in turn had a div child, the subtree wouldn’t round-trip.) Furthermore, even if serializing and reparsing resulted in a same-looking tree, the nodes created by the parser would be different nodes than the nodes that were children of element at first. Thus, if other parts of the JavaScript program were holding references to descendants of element, after element.innerHTML += "markup"; had been executed, those references would point to detached nodes and element would have new similar but different descendants.

When additional content is inserted using insertAdjacentHTML(), the existing nodes stay in place.

Better performance

Serializing and reparsing is also what leads to performance problems with the element.innerHTML += "markup"; pattern. Each time some more content is appended, all the existing content in element gets serialized and reparsed. This means that appending gets slower and slower, because each time there more and more previous content to serialize and reparse.

Using insertAdjacentHTML() can make a big difference. For testing purposes, I started with an empty div and ran a loop that tried to append as many tweets as possible to the div in five seconds. A tweet is actually rather large when you count all the mark-up that implements @mention linkification, the name of the tweeter, retweet and favoriting UI, etc. It weighs about 1600 characters of HTML source—most of it mark-up.

On the computer that I used for testing, the innerHTML way of appending managed to append only slightly over 200 tweets in five full seconds. In contrast, the insertAdjacentHTML("beforeend", ...) way of appending managed to append almost 30,000 tweets in 5 seconds. (Yes, that’s hundreds versus tens of thousands.) Obviously, real Web apps should never block the event loop for five seconds—this is just for benchmark purposes. However, this illustrates how the innerHTML way of appending becomes notably slower as more and more content accumulates to be serialized and reparsed each time.

At this point, some readers might wonder if insertAdjacentHTML() offers any benefit over createContextualFragment(). After all, conceptually insertAdjacentHTML() creates a fragment and inserts it.

Using createContextualFragment(), my test manages only slightly over 25,000 tweets in five seconds, while using insertAdjacentHTML() manages slightly under 30,000. This is because Gecko accelerates insertAdjacentHTML() when the insertion point has no next sibling (only for HTML though—not for XML so far). The "beforeend" insertion point never has a next sibling and is always accelerated (for HTML). The "beforebegin" insertion point always has a next sibling (the node that insertAdjacentHTML() was invoked on) and is never accelerated. For "afterbegin" and "afterend", whether the operation is accelerated depends on the situation.

In conclusion, you can make your Web app perform better by using element.insertAdjacentHTML("beforeend", "markup"); where you currently use element.innerHTML += "markup";.

View full post on Mozilla Hacks – the Web developer blog

Leave a Reply