lit-html vs hyperHTML vs lighterhtml

Photo by Pietro Mattia on Unsplash

When lit-html released at the end of July 2017 as an experimental, not production ready, library, I’ve created few days after a huge gist comparing it with what was, at that time, the already production ready hyperHTML.

However, over the weekend, I’ve decided to remove such gist, and its comparison, for various reasons, including:

  • gists suck at being tracked. No notifications, no PR/MR possible, gists are just a quick “trash bin” for Markdown or snippets
  • I sucked at remembering to update that complex Markdown table, so that everything written in it, or to be fair most of pain points regarding lit-html at that time, would be wrong today
  • lit-html released its first stable 1.0
  • lighterhtml was also recently born as hyperHTML drop-in simplified alternative, making the comparison actually fair

One does not simply compare lit against hyper

There is quite a lot happening in hyperHTML, that it makes not much sense to compare it against lit-html.

To start with, hyperHTML has a viperHTML node server side alter ego, with API features parity, asynchronous streamed renders, and components.

Speaking of which, lightweight components are part of hyperHTML too. These are yet another out-of-the-box feature that comes with hyper/viperHTML.

hyperHTML, as well as viper, also offer a way to extend their syntax, asynchronous renders through any “thenable”, and last, but not least, it offers an explicit way to relate DOM nodes with any kind of object, through an API that many found initially somehow difficult to grasp.

Removing all these features, plus simplifying the API, was the goal of lighterhtml, which in its 0.9 version is as fast as lit-html, but also always smaller, in size, than hyperHTML (about ~1K once gzipped but … still)

So … is hyperHTML slow?

Hell No. You can check by yourself this benchmark and realize it’s faster than Preact, React, Vue, Angular, you name it … but hyperHTML doesn’t fit into the lit-vs-light context ’cause it cannot, and it won’t, compete against them.

hyperHTML wins in perf compared to common frameworks

… and what about HyperHTMLElement ?

Yes, HyperHTMLElement is based on top of hyperHTML, and its part of its ecosystem, but using lighterhtml instead would be a no brainer.

Yet, in ABP, we use hyperHTML due its battle-tested Desktop to Mobile compatibility and stability.

However, since HyperHTMLElement is something a part, as a part is also the LitElement class, I’ve left these “easy to implement” abstractions, or wrappers, out of this context, comparing instead only what’s meaningful for daily tasks.

lit-html VS lighterhtml

This is the only comparison that makes sense these days, ’cause both libraries offer pretty much same functionalities, API, and performance.

It is true that lit-html might be slightly richer in features, if you use its directives, but what I consider important here, is what we can do with just their core, summarized as such:

  • be as standard based as possible
  • create, and use, arbitrary DOM content
  • be able to quickly change content on the page (aka: DOM diffing)

These points are what lit-html core offers as a whole, while lighterhtml includes some extra feature we’ll see, and compare, later on.

How much standard ?

Both lighterhtml and lit-html are based on “just standards”, meaning you can use JS without the need to transpile it, and you declare HTML, or SVG, through template literals.

// lighterhtml and lit-html: the basics
html`
<div>
This will be the content of the div,
including this ${Math.random()} value.
</div>`;

While above example will be equivalent for both libraries, lit-html would accept some sort of specialized, non standard, HTML syntax, to provide some feature such boolean attributes, events, or properties.

// lit-html specialized, non standard, syntax
html`<input
.value=${value}
@click=${handler}
?disabled=${disabled}
>`;

Hopefully needless to explain what those fancy attributes would do once rendered, I’d like to provide anyway a link that explains more about.

The equivalent in lighterhtml, based on the same hyperHTML syntax engine, would be the following one:

// lighterhtml standard syntax
html`<input
value=${value}
onclick=${handler}
disabled=${disabled}
>`;

As developer, you don’t need to do anything different than writing HTML as you know, the library will figure out at runtime, accordingly with the kind of node it’s dealing with, what should result as input.value, as input.disabled, and as input.addEventListener("click", handler) .

The only real non standard syntax available in lighterhtml, but already widely accepted by the community, is the self closing tag.

// lighterhtml simplifies closing tags
html`<div><span/><span/></div>`;
html`<custom-element attr=${value}/>`;
html`<textarea value=${text}/>`;

The <span/><span/> case would produce a span inside another one in lit-html, while it will simply create two adjacent spans in lighterhtml.

I let you judge how confusing, or expected, this behavior is in either library, all I know is that I’ve never explicitly closed any tag if I wanted the next one to be part of its content, and I wish HTML would’ve done the same.

What does `html` return ?

In the case of lit-html, the html function tag returns some instance of some internal class used by the library to understand its own content.

// FAILS in lit-html
document.body.appendChild(html`<hr>`);
..
TypeError: Argument 1 ('node') to Node.appendChild must be an instance of Node

In lighterhtml though, you can use right away html or svg to create content:

// OK in lighterhtml
document.body.appendChild(html`<hr>`);

The only special case, is when you have hybrid containers, those represented by more than a node, without being inside a single outer container:

// lighterhtml hybrid containers
// (i.e. no <div> around)
html`
<button>first</button>
<button>second</button>`;

In this case the library also uses an internal representation of a DOM fragment, but that will be returned through the standard method valueOf() , providing the ability to use any template as one off, without caring much about the content.

document.body.appendChild(html`...`.valueOf());

Bear in mind valueOf() works even if the returned content is a real node 😉

How to create arbitrary DOM content, then?

The render function is there to accomplish the task.

import {render, html} from '//unpkg.com/lit-html?module';
render(html`<div>Hello lit-html</div>`, document.body);

The same could be done via lighterhtml, but instead of render(what, where), lighterhtml has a render(where, what()) signature.

import {render, html} from '//unpkg.com/lighterhtml?module';
render(document.body, () => html`<div>Hello lighterhtml</div>`);

The what() callback is handy when it comes to render components, as in render(where, App) case, but also to distinguish between internal renders VS one-off renders, as we’ve seen before.

… and the simplicity trophy goes to … 🥁🥁🥁

Objectively esier to start with, through zero specialized syntax, if not the self closing tag, which is used by JSX and XML first since ever, and without any lock-in for rendering some one-off content, I’d say lighterhtml wins the simplicity context 🏆.

There is arguably the partial attributes “gotcha that could make lit-html simpler to use in some case, but that’s one thing VS standard syntax, self closing least surprise, and one-off capability, keeping the trophy where it is.

What about performance ?

There are two sides to this metric: one is the bundled size, where smaller would mean faster time to be interpreted, and one is the raw rendering performance of the library itself.

The bundle size

Using brotli as compressor, and executing the following in any of the keyed or non-keyed folders of the js-framework-benchmark:

brotli -c9 dist/index.js | wc -c

We can observe that lit-html is 3880 non keyed, and 4441 keyed, while lighterhtml is steady at 5794.

Since both benchmarks use pretty much the same test code, this makes lit-html surely winning the final production code size 🏆.

However, the benchmark itself shows that when the difference in size is around 1.5K, when it comes to keyed results, the final performance score might not be impacted.

Currently, lighterhtml wins in the keyed benchmark 🏆 while lit-html wins in the non keyed one 🏆, but I’d be curious to see new results once lighterhtml 0.9 goes in.

In few words, when you have slightly more horsepower, it doesn’t matter much if your car is a couple of kilos heavier, but if you’d like to know why lighterhtml is heavier, it’s mostly because it inherits the same compatibility granted by hyperHTML, and it patches internally, at runtime, a lot of little gotchas found in variuous browsers or JavaScript transpilers.

lit-html also fixes through little global runtime shims some browser, but it’s not as backward compatible as lighterhtml is.


Update, the 0.9 got merged and the score is now 1.01 vs 1.02 🎉


What does keyed VS non-keyed mean?

In few words, keyed results are the one you’d expect the most, while non keyed results mean that any created node could represent any sort of data.

If it makes any sense, you can think of non-keyed as the result of any server side rendered page, where nodes that relate to specific articles, items, database results, or user preferences, are just on the layout with some info exposed through some attribute.

<ul id="user-list">
<li data-reason="wish" data-item-id="1234">Sony PS5</li>
<li data-reason="checkout" data-item-id="456">LG Mouse</li>
</ul>

When content arrives from the server, items are usually identified by unique IDs, as in above data-item-id case, and rarely by objects or, even worse case, DOM nodes.

The relationship is between informations exposed through the rendered page, and its nodes with their attributes, plus some action to perform, on the client or the server, when some DOM node related to each item is added, modified, or removed.

This is somehow an example of non keyed results, where a single item ID could be the same in many nodes, or many panels, without binding it to any specific JavaScript object.

This is also a very common, non client side driven, scenario, but when part of the business logic is delegated to the client, keyed driven views relates nodes to specific items/objects, abstracting away the contract between an item, as JS entity itself, and one of its multiple representations on any part of the page.

The improvement is given by the fact whenever such JS entity is removed, updated, or modified, any related view would follow up the change, without needing to query every single time the whole document, or ask the server to refresh the entire page.


Quick note: non keyed results visually represent the same data, but there’s no guarantee that a node is directly related to any object held in the JS world.

With keyed results, html.for(entity) will always return the node associated to entity object, giving us the ability to eventually retrieve, or update, directly, any node associated to a single entity without even knowing if such node has been already rendered or not.


A classic TODOs could be a good example:

// items (used as references)
const list = [
{id: 1, desc: 'some'},
{id: 2, desc: 'item'}
];
// shopping chart ...
const ul = document.querySelector('#shopping-chart');
render(
ul,
() => html.for(ul)`<ul>${
list.map(item => html.for(item, ul.id)`
<li>${item.desc}</li>`)
}</ul>`
);
// ... and a summary
const summary = document.querySelector('#summary');
render(
summary,
() => html.for(summary)`<p>${
list.map((item, i) => html.for(item, summary.id)`
${i ? ', ' : ''}
<span>${item.desc}</span>`)
}</p>`
);

Since all items are referenced, as soon as one gets updated, its related view state would be reflected in every place.

This is somehow an inverted control for the rendering path of the data you hold via JavaScript, and it’s basically representing server side renders on the fly, whenever any item gets modified.

Imagine asking the whole web page each time you need to update a single item in the chart, or even a button inside a runtime calculator … right?

Instead, only what needs to get updated will actually get updated, as you can see in the console.


Quick note about having keyed results in lighterhtml, all you need to do is to pass a reference with, or without, an id, while in lit-html you’d need a repeater from its directives, as shown in this code pen.


… wait a minute, what are those comments ?

The hyperHTML technique to address anchors in the DOM to render partial updates has been used in both lighterhtml and somehow simulated as well in lit-html, but in latter library you’ll see duplicated comments in the wild, as boundaries of every dynamic portion of the DOM you will eventually update through its logic.

In both hyper and lighterhtml case, updates are performed throught the domdiff helper which uses just a node and any sort of algorithm to quickly compute the amount of changes needed between one DOM state, and another.

However, lit-html does also surprisingly well in common use cases for diffing, and I’ve created extra tests to be sure what I was testing made sense.

Non Keyed DOM diffing

Beside constant faster time in the initial creation, I wouldn’t say lighterhtml is any significantly faster than lit-html in these Code Pens.

Keyed DOM diffing

As already shown in the previous js-framework-benchmark, lighterhtml is consistently faster, and yet I wouldn’t say it’s any relevantly faster than lit.

… so, who’s the winner ?

I think both lighterhtml and lit-html win on the Web, ’cause both provide a relatively straight forward and simple mechanism to create, and update, any sort of content.

In a Web populated by 50K minimum frameworks size on average, it makes literally no sense to compare 5K VS 3.5K, and what I’d rather compre, is what each framework provides for users.

lit-html pros

  • built-in directives to simplified common tasks and easily extends the lib
  • smallest core when directives, repeat, style-map, and guard functionalities are not needed
  • blazing fast compared to any bigger framework 🏆

lighterhtml pros

  • React like hooks out of the box through its lighterhtml.hook(useRef) feature and any hooks oriented project such as augmentor, dom-augmentor, or TNG-Hooks, so that hooks can provide a way to extend renders as you need, just like directives would do
  • simple to start, simpler to use when it comes to keyed items, with no need of a repeater and style maps out of the box
  • widely backward compatible and transpilers resistant
  • blazing fast compared to any bigger framework 🏆
  • it always wins in terms of memory consumption