mutationdiff

This package allows you to do incremental DOM diffing on a live DOM tree. It is designed to take async, batched mutation records from MutationObserver as input for diff calculations; but you can also report individual mutations manually if desired. In addition to getting the diff, you can also patch a DOM given a list of differences, revert the DOM to its original state, or query the range where mutations occurred. It can also be used just to summarize and simplify the log of mutations given by MutationObserver.

How does this package differ from virtual DOM diffing or other diffing libraries?

  • it gives the diff of a DOM tree at two points in time, rather than the delta between two different DOM trees
  • it doesn't use a virtual DOM, only operating on a live DOM tree
  • it is computationally efficient since it only does calculations on mutated nodes (as reported by MutationObserver or other method)
  • it is memory efficient since it doesn't make a copy of the DOM tree to diff against (only the differing nodes and their corresponding diff metadata needs to be stored)

Benchmarks in Chrome v108 for a typical use case:

  • Processing a MutationRecord: 261k per/sec
  • Retrieving full diff results: 107k per/sec
  • Check if DOM has changed: 493k per/sec
  • Get the extent/range of DOM changes: 52.3k per/sec
  • Revert DOM changes: 12.6k per/sec

Originally, this library was written to revert browser-made edits to a contenteditable element. A naive approach to reversion would be to rewind a log of MutationRecord, as suggested in the MDN documentation. Performing DOM diffing is more work than simply keeping a log, but in exchange we get exact bounds for the range of mutations, memory usage is optimal, and reversion can directly place nodes back in their original positions; it will have better worst-case behavior.

API documentation | npm package | GitHub source code

Installation

npm i mutationdiff

This project uses ES 2015+ class features. A Babel transpiled and minified version is provided as mutationdiff.compat.min.js, with exports under MutationDiff; though I highly recommend building a bundle yourself to customize the target and reduce code size. A plain minified version is provided as mutationdiff.min.js.

Usage

// an alias is recommended for MutationDiffFlags
import { MutationDiff, MutationDiffObserver, MutationDiffFlags as F } from "mutationdiff";

If not using a bundler, you'll need to import from the minified version, which is pre-bundled.

Quickstart Method Reference

  • record, data, attribute, custom, children: report a DOM mutation
  • mutated: check if there are any differences
  • range: get the extent of any differences
  • diff: get diff results
  • diff_grouped_children: group node movement diffs
  • patch_grouped_children: apply grouped node movements
  • revert: undo any diff
  • clear: reset diff tracking

Quickstart Example

Make sure to check the API documentation for full details.

// a small DOM for this example
const dom = init_dom();
// create our diff tracking object
const tracker = new MutationDiff();

Report any mutations of interest to MutationDiff by calling data, attribute, custom, or children methods. Most likely, you'll want to get mutations using a MutationObserver, but you could also have some controller logic that all DOM modifications go through.

If you're using MutationObserver, there is a helper class MutationDiffObserver which can be used for simple cases. It will initialize a MutationObserver and setup all the hooks to report to MutationDiff for you. That would look like this:

// setup our helper object
const observer = new MutationDiffObserver(tracker, dom.root);
// DOM is mutated...
dom.mutate();
// custom, user-defined properties always need to be recorded manually
tracker.custom(dom.B, "child_count", dom.B.childNodes.length, dom.B_old_child_count);
// flush results and synchronize tracker
observer.flush();
// stop observing (optional)
observer.stop();

If you have more advanced use cases and want to reuse the MutationObserver, or filter/process the MutationRecords before passing to MutationDiff, then you might consider doing the setup yourself. That would look like this:

function handle_records(records){
	for (const r of records){
		// custom, user-defined properties always need to be recorded manually
		if (r.target === dom.B && r.type === "childList")
			tracker.custom(dom.B, "child_count", dom.B.childNodes.length, dom.B_old_child_count);
		tracker.record(r);
		// you could also process records manually and call data, attribute, custom, children
	}
}
// setup the MutationObserver
const observer = new MutationObserver(handle_records);
observer.observe(dom.root, {
	subtree: true,
	childList: true,
	attributes: true,
	attributeOldValue: true,
	characterData: true,
	characterDataOldValue: true
});
// DOM is mutated...
dom.mutate();
// flush MutationObserver
handle_records(observer.takeRecords());
// signals to MutationDiff that all mutations have been recorded
tracker.synchronize();
// stop observing (optional)
observer.disconnect();

Here we just watch a single node, root, but in practice you can watch any number of separate DOM trees using a single MutationDiff. We're also observing all DOM changes, but you could specify a filter to ignore attributes or character data changes, for example.

The record method is a convenient helper which calls data, attribute, and children methods appropriately for a MutationRecord.

Since MutationObserver is async and batched, we need to flush any pending MutationRecords prior to reading the results. Note also the call to synchronize after flushing; the call to synchronize is only needed if:

  1. You want to accurately revert or patch DOM trees besides root (or the set of root nodes you're observing).
  2. You want to reduce memory usage for property (data, attribute, custom) diff information

For our example, if all we wanted to do was revert root, the synchronize call would not be necessary. A more detailed explanation of what synchronize does is described in the Diffing Caveat #2 and Diffing Caveat #3 sections. When in doubt, you can always call synchronize, with just a minor increase in computation.

The init_dom function is defined as:

function init_dom(){
	// Original DOM: <div><span></span></div>, <span id=B></span>, #old text
	const root = document.createElement("div");
	const A = document.createElement("span");
	root.appendChild(A);
	const B = document.createElement("span");
	B.id = "B";
	const B_old_child_count = B.childNodes.length;
	const txt = document.createTextNode("old text");

	function mutate(){
		// Mutated DOM: <div><span id=B_modified></span>#new text<span></span></div>
		root.appendChild(B);
		root.appendChild(txt);
		A.remove();
		B.id = "B_modified";
		txt.after(A);
		txt.data = "new text";
	}

	return {root, A, B, txt, B_old_child_count, mutate};
}

Now we can read the diffing results:

console.log(tracker.mutated());
// output: true
console.log(tracker.range());
// output: BoundaryRange[(root, AFTER_OPEN), (A, BEFORE_OPEN)]
console.log(tracker.diff());
/* output:
	Map {
		dom.B => {
			attribute: {
				id: {
					original: "B",
					mutated: "B_modified"
				}
			},
			custom: Map {
				child_count => {
					original: 0,
					mutated: 1
				}
			}
			children: {
				mutated: {
					parent: dom.root,
					prev: null,
					next: dom.txt
				}
			}
		},
		dom.txt => {
			data: {
				original: "old text",
				mutated: "new text
			},
			children: {
				mutated: {
					parent: dom.root,
					prev: dom.B,
					next: dom.A
				}
			}
		}
	}
*/

Use mutated to check if there are any differences and range to get the extent of those differences. Both take a root node argument to optionally confine the results, but in this example it was not needed; if we were tracking multiple disconnected DOM trees (e.g. Node.getRootNode is different), it would be necessary.

The main diff results are given by diff. Here we're returning all results, but you can pass a filter to the method to limit what is returned (for example, excluding the "original" values). You'll notice several things from this example:

  • The original node position (inside children) for B and txt is missing; this is because the nodes were originally orphaned, having no parent node.
  • A does not appear in the diff, since even though it was moved, its relative position with respect to root was unchanged. A node is considered "unchanged" when it is next to one of its original siblings (ignoring any newly inserted siblings in-between), and that sibling is itself unchanged.
  • Full text diffing is not performed for data changes. Only string equality is checked. If full text diffing is needed, you can perform it as a post-processing step using a separate text diffing library (e.g. diff-match-patch)

As a final note about synchronize, if you did not call it, any ill affected nodes will have either a missing mutated children value, or the original next/prev siblings may be missing.

For children, there is another method, diff_grouped_children, which will group adjacent node movements. In many cases this can be more useful:

const grouped = Array.from(tracker.diff_grouped_children(F.MUTATED));
console.log(grouped);
/* output:
	[{
		nodes: [B, txt],
		parent: root,
		prev: null,
		next: A
	}]
*/

To revert the DOM back to its original state, simply call revert:

tracker.revert();
console.log(tracker.mutated());
// output: false

You can apply the output of diff_grouped_children, or a similary formed iterable, to patch a DOM tree's node positions:

MutationDiff.patch_grouped_children(grouped);

While you can't patch an unrelated DOM tree out-of-the-box, you can easily remap the nodes of diff_grouped_children to work on the unrelated tree. See patch_grouped_children method documentation for more details.

Diffing Caveat #1

The first caveat arises when you have a sequence of sibling nodes that have been rearranged. Consider the following:

[A, B, C, D] -> rearranged to -> [B, C, D, A]

We could interpret this as a single move of A to the back of the list, a movement of [B, C, D] to the front, or a bulk movement of all the nodes. While the first option gives the minimal number of node movements, this is not necessarily what MutationDiff will return.

MutationDiff does incremental diffing, meaning that every call to children updates the diff. Each call to children is treated like a single, atomic operation: we log all node removals, then node additions, and then finally perform diffing to check if any of the newly added nodes have reverted to their original position.

How the rearrangement example above will be interpreted depends on what mutations were reported to children, and in what order. The diff given by MutationDiff is thus a true representation of the actual mutations that occurred. If the rearrangement was due to a bulk movement of all nodes (e.g. via Node.replaceChildren), then it will be reflected as such in the diff; currently, there is no post-processing step to "reinterpret" an operation to minimize the number of node movements.

You could implement the Myers' diff algorithm on top of MutationDiff results if minimizing node movements is necessary, assuming the extra computational cost is worth it.

Diffing Caveat #2

The second caveat arises specifically when using MutationObserver with multiple disconnected DOM trees. If only one of the trees is being observed, its possible a node insertion could go untracked. For example, using the same starting DOM as in the Quickstart Example:

// A is removed, but since B is not currently being observed, its insertion
// into B will go untracked
B.appendChild(A);
// B has been added to root, so it is now being observed
root.appendChild(B);
// txt's previousSibling is A, revealing that A was not removed after all
A.after(txt);

Diffing will only work properly if all the node insertions and removals are accurately reported. The diffing algorithm uses the point-in-time nextSibling and previousSibling to calculate the diff.

Unfortunately, MutationObserver is designed to observe changes to a node's childList, not movements of the node itself. In this particular instance, we miss the insertion of A, and so its siblings are unknown. Since MutationObserver is async and batched, we can't look back in time to see what the siblings were when A was inserted.

One solution is to observe B from the start, so that the insertion of A is sure to be tracked:

observer.observe(B);

But in some cases, the fact that B will be inserted into root at some future time cannot be known ahead of time.

Since this scenario can arise so easily when using MutationObserver, the diffing algorithm includes code to specifically address this case. When it detects a node whose insertion was not tracked, it creates a Promise object to later perform diffing calculations when the node's siblings become known. In subsequent calls to children, those siblings may be revealed.

In the event the siblings are not revealed, a call to diff may output undefined for the mutated or original siblings of a node. This can prevent DOM reversion or patching from working properly. Importantly, this only affects diffing in the disconnected DOM tree; in our example, B could not be reverted correctly, but root would be. For some use cases, you may be fine with this.

To allow correct reversion of the disconnected tree, there is a method synchronize, which signals to MutationDiff that all mutations have been reported. MutationDiff is then free to consult the DOM to resolve any unknown siblings or parents by checking their current previousSibling, nextSibling, or parentNode values. It can then finalize all pending diff calculations and give a complete diff result.

// call this after MutationObserver.takeRecords()
tracker.synchronize();

Diffing Caveat #3

MutationObserver does not give you point-in-time mutated values for attributes or character data. Only the old point-in-time value is included in the MutationRecord. Unfortunately given the async batched nature of MutationObserver, this means we need to cache the original values even when they are unchanged. Calling synchronize however will signal to MutationDiff that all mutations have been recorded, and so it will finally be able to discard any unchanged property values.

// call this after MutationObserver.takeRecords()
tracker.synchronize();

If you will immediately call revert or clear, then synchronizing will not help in this case, as the cached values will be discarded following those methods anyways.