Social Media Photo by JJ Ying on Unsplash
LinkeDOM is a triple-linked list based DOM-like namespace, for DOM-less environments, with the following goals:
- avoid maximum callstack/recursion or crashes, even under heaviest conditions.
- guarantee linear performance from small to big documents.
- be close to the current DOM standard, but not too close.
import {DOMParser, parseHTML} from 'linkedom';
// Standard way: text/html, text/xml, image/svg+xml, etc...
// const document = (new DOMParser).parseFromString(html, 'text/html');
// Simplified way for HTML
const {
// note, these are *not* globals
window, document, customElements,
HTMLElement,
Event, CustomEvent
// other exports ..
} = parseHTML(`
<!doctype html>
<html lang="en">
<head>
<title>Hello SSR</title>
</head>
<body>
<form>
<input name="user">
<button>
Submit
</button>
</form>
</body>
</html>
`);
// builtin extends compatible too π
customElements.define('custom-element', class extends HTMLElement {
connectedCallback() {
console.log('it works π₯³');
}
});
document.body.appendChild(
document.createElement('custom-element')
);
document.toString();
// the SSR ready document
document.querySelectorAll('form, input[name], button');
// the NodeList of elements
// CSS Selector via CSSselect
- in
v0.11
a newlinkedom/worker
export has been added. This works with deno, Web, and Service Workers, and it's not strictly coupled with NodeJS. Please note, this export does not includecanvas
module, and theperformance
is retrieved from theglobalThis
context.
LinkeDOM uses a blazing fast JSDON serializer, and nodes, as well as whole documents, can be retrieved back via parseJSON(value)
.
// any node can be serialized
const array = document.toJSON();
// somewhere else ...
import {parseJSON} from 'linkedom';
const document = parseJSON(array);
Please note that Custom Elements won't be upgraded, unless the resulting nodes are imported via document.importNode(nodeOrFragment, true)
.
Alternatively, JSDON.fromJSON(array, document)
is able to initialize right away Custom Elements associated with the passed document
.
This module is based on DOMParser API, hence it creates a new document
each time new DOMParser().parseFromString(...)
is invoked.
As there's no global pollution whatsoever, to retrieve classes and features associated to the document
returned by parseFromString
, you need to access its defaultView
property, which is a special proxy that lets you get pseudo-global-but-not-global properties and classes.
Alternatively, you can use the parseHTML
utility which returns a pseudo window object with all the public references you need.
// facade to a generic JSDOM bootstrap
import {parseHTML} from 'linkedom';
function JSDOM(html) { return parseHTML(html); }
// now you can do the same as you would with JSDOM
const {document, window} = new JSDOM('<h1>Hello LinkeDOM π</h1>');
The triple-linked list data structure is explained below in How does it work?, the Deep Dive, and the presentation on Speakeasy JS.
Why "not too close"?
LinkeDOM has zero intention to:
- implement all things JSDOM already implemented. If you need a library which goal is to be 100% standard compliant, please use JSDOM because LinkeDOM doesn't want to be neirly as bloated nor as slow as JSDOM is
- implement features not interesting for Server Side Rendering. If you need to pretend your NodeJS, Worker, or any other environment, is a browser, please use JSDOM
- other points listed, or not, in the followung F.A.Q.s: this project will always prefer the minimal/fast approach over 100% compliant behavior. Again, if you are looking for 100% compliant behavior and you are not willing to have any compromise in the DOM, this is not the project you are looking for
That's it, the rule of thumb is: do I want to be able to render anything, and as fast as possible, in a DOM-less env? LinkeDOM is great!
Do I need a 100% spec compliant env that simulate a browser? I rather use cypress or JSDOM then, as LinkeDOM is not meant to be a replacement for neither projects.
Are live collections supported?
The TL;DR answer is no. Live collections are considered legacy, are slower, have side effects, and it's not intention of LinkeDOM to support these, including:
getElementsByTagName
does not update when nodes are added or removedgetElementsByClassName
does not update when nodes are added or removedchildNodes
, if trapped once, does not update when nodes are added or removedchildren
, if trapped once, does not update when nodes are added or removedattributes
, if trapped once, does not update when attributes are added or removeddocument.all
, if trapped once, does not update when attributes are added or removed
If any code you are dealing with does something like this:
const {children} = element;
while (children.length)
target.appendChild(children[0]);
it will cause an infinite loop, as the children
reference won't side-effect when nodes are moved.
You can solve this in various ways though:
// the modern approach (suggested)
target.append(...element.children);
// the check for firstElement/Child approach (good enough)
while (element.firstChild)
target.appendChild(element.firstChild);
// the convert to array approach (slow but OK)
const list = [].slice.call(element.children);
while (list.length)
target.appendChild(list.shift());
// the zero trap approach (inefficient)
while (element.childNodes.length)
target.appendChild(element.childNodes[0]);
Are childNodes and children always same?
Nope, these are discovered each time, so when heavy usage of these lists is needed, but no mutation is meant, just trap these once and use these like a frozen array.
function eachChildNode({childNodes}, callback) {
for (const child of childNodes) {
callback(child);
if (child.nodeType === child.ELEMENT_NODE)
eachChildNode(child, callback);
}
}
eachChildNode(document, console.log);
All nodes are linked on both sides, and all elements consist of 2 nodes, also linked in between.
Attributes are always at the beginning of an element, while zero or more extra nodes can be found before the end.
A fragment is a special element without boundaries, or parent node.
Node: β node β
Attr<Node>: β attr β β ownerElement?
Text<Node>: β text β β parentNode?
Comment<Node>: β comment β β parentNode?
Element<Node>: β start β end β β parentNode?
Fragment<Element>: start β end
Element example:
parentNode? (as shortcut for a linked list of previous nodes)
β
ββββββββββββββββββββββββββββββββββββββββββββββ
β β
node? β start β attr* β text* β comment* β element* β end β node?
β β
ββββββββββββββββββββββββββββββββββββββββββββββ
Fragment example:
ββββββββββββββββββββββββββββββββββββββββββββββ
β β
start β attr* β text* β comment* β element* β end
β β
ββββββββββββββββββββββββββββββββββββββββββββββ
If this is not clear, feel free to read more in the deep dive page.
Moving N nodes from a container, being it either an Element or a Fragment, requires the following steps:
- update the first left link of the moved segment
- update the last right link of the moved segment
- connect the left side, if any, of the moved node at the beginning of the segment, with the right side, if any, of the node at the end of such segment
- update the parentNode of the segment to either null, or the new parentNode
As result, there are no array operations, and no memory operations, and everything is kept in sync by updating a few properties, so that removing 3714
sparse <div>
elements in a 12M document, as example, takes as little as 3ms, while appending a whole fragment takes close to 0ms.
Try npm run benchmark:html
to see it yourself.
This structure also allows programs to avoid issues such as "Maximum call stack size exceeded" (basicHTML), or "JavaScript heap out of memory" crashes (JSDOM), thanks to its reduced usage of memory and zero stacks involved, hence scaling better from small to very big documents.
As everything is a while(...)
loop away, by default this module does not cache anything, specially because caching requires state invalidation for each container, returned queries, and so on. However, you can import linkedom/cached
instead, as long as you understand its constraints.
This module parses, and works, only with the following nodeType
:
ELEMENT_NODE
ATTRIBUTE_NODE
TEXT_NODE
COMMENT_NODE
DOCUMENT_NODE
DOCUMENT_FRAGMENT_NODE
DOCUMENT_TYPE_NODE
Everything else, at least for the time being, is considered YAGNI, and it won't likely ever land in this project, as there's no goal to replicate deprecated features of this aged Web.
This module exports both linkedom
and linkedom/cached
, which are basically the exact same thing, except the cached version outperforms linkedom
in these scenarios:
- the document, or any of its elements, are rarely changed, as opposite of frequently mutated or manipulated
- the use-case needs many repeated CSS selectors, over a sporadically mutated "tree"
- the generic DOM mutation time is not a concern (each, removal or change requires a whole document cache invalidation)
- the RAM is not a concern (all cached results are held into NodeList arrays until changes happen)
On the other hand, the basic, non-cached, module, grants the following:
- minimal amount of RAM needed, given any task to perform, as nothing is ever retained on RAM
- linear fast performance for any every-time-new structure, such as those created via
importNode
orcloneNode
(i.e. template literals based libraries) - much faster DOM manipulation, without side effect caused by cache invalidation
To run the benchmark locally, please follow these commands:
git clone https://github.com/WebReflection/linkedom.git
cd linkedom/test
npm i
cd ..
npm i
npm run benchmark