#minify-html #minify #html #nodejs #minification

hyperbuild

Fast allocation-less HTML minifier with smart whitespace handling

60 releases

0.2.4 Jul 11, 2020
0.2.3 Jul 10, 2020
0.1.12 Jul 10, 2020
0.1.3 Jun 19, 2020
0.0.4 Dec 30, 2019

#761 in Compression

Download history 212/week @ 2024-07-28 114/week @ 2024-09-22 1/week @ 2024-09-29

156 downloads per month
Used in pagong

MIT license

2MB
47K SLoC

hyperbuild

A fast one-pass in-place HTML minifier written in Rust with context-aware whitespace handling.

Also supports JS minification by plugging into esbuild.

Available as:

  • CLI for macOS and Linux.
  • Rust library.
  • Native library for Node.js, Python, Java, and Ruby.

Features

  • Minification is done in one pass with no backtracking or DOM/AST building.
  • No extra heap memory is allocated during processing, which increases performance.
  • Context-aware whitespace handling allows maximum minification while retaining desired spaces.
  • Well tested with a large test suite and extensive fuzzing.

Performance

Speed and effectiveness of Node.js version compared to html-minfier and minimize, run on popular already-minified web pages. See bench folder for more details.

Chart showing speed of HTML minifiers Chart showing effectiveness of HTML minifiers

Usage

CLI

Precompiled binaries are available for x86-64 macOS and Linux.

Get

macOS | Linux

Use

Use the --help argument for more details.

hyperbuild --src /path/to/src.html --out /path/to/output.min.html

API

Rust
Get
[dependencies]
hyperbuild = { version = "0.2.4", features = ["js-esbuild"] }

Building with the js-esbuild feature requires the Go compiler to be installed as well, to build the JS minifier.

If the js-esbuild feature is not enabled, cfg.minify_js will have no effect.

Use
use hyperbuild::{Cfg, FriendlyError, hyperbuild, hyperbuild_copy, hyperbuild_friendly_error, hyperbuild_truncate};

fn main() {
    let mut code = b"<p>  Hello, world!  </p>".to_vec();
    let cfg = &Cfg {
        minify_js: false,
    };

    // Minifies a slice in-place and returns the new minified length,
    // but leaves any original code after the minified code intact.
    match hyperbuild(&mut code, cfg) {
        Ok(minified_len) => {}
        Err((error_type, error_position)) => {}
    };

    // Creates a vector copy containing only minified code
    // instead of minifying in-place.
    match hyperbuild_copy(&code, cfg) {
        Ok(minified) => {}
        Err((error_type, error_position)) => {}
    };

    // Minifies a vector in-place, and then truncates the
    // vector to the new minified length.
    match hyperbuild_truncate(&mut code, cfg) {
        Ok(()) => {}
        Err((error_type, error_position)) => {}
    };

    // Identical to `hyperbuild` except with FriendlyError instead.
    // `code_context` is a string of a visual representation of the source,
    // with line numbers and position markers to aid in debugging syntax.
    match hyperbuild_friendly_error(&mut code, cfg) {
        Ok(minified_len) => {}
        Err(FriendlyError { position, message, code_context }) => {
            eprintln!("Failed at character {}:", position);
            eprintln!("{}", message);
            eprintln!("{}", code_context);
        }
    };
}
Node.js

hyperbuild is on npm, available as a Node.js native module, and supports Node.js versions 8 and higher.

Get

Using npm:

npm i hyperbuild

Using Yarn:

yarn add hyperbuild
Use
const hyperbuild = require("hyperbuild");

const cfg = { minifyJs: false };
const minified = hyperbuild.minify("<p>  Hello, world!  </p>", cfg);

// Alternatively, minify in place to avoid copying.
const source = Buffer.from("<p>  Hello, world!  </p>", cfg);
hyperbuild.minifyInPlace(source);

hyperbuild is also available for TypeScript:

import * as hyperbuild from "hyperbuild";
import * as fs from "fs";

const cfg = { minifyJs: false };
const minified = hyperbuild.minify("<p>  Hello, world!  </p>", cfg);
hyperbuild.minifyInPlace(fs.readFileSync("source.html"), cfg);
Java

hyperbuild is available via JNI, and supports Java versions 7 and higher.

Get

Add as a Maven dependency:

<dependency>
  <groupId>in.wilsonl.hyperbuild</groupId>
  <artifactId>hyperbuild</artifactId>
  <version>0.2.4</version>
</dependency>
Use
import in.wilsonl.hyperbuild.Hyperbuild;

Hyperbuild.Configuration cfg = new Hyperbuild.Configuration.Builder()
    .setMinifyJs(false)
    .build();
try {
    String minified = Hyperbuild.minify("<p>  Hello, world!  </p>", cfg);
} catch (Hyperbuild.SyntaxException e) {
    System.err.println(e.getMessage());
}

// Alternatively, minify in place:
assert source instanceof ByteBuffer && source.isDirect();
Hyperbuild.minifyInPlace(source, cfg);
Python

hyperbuild is on PyPI, available as a native module, and supports CPython (the default Python interpreter) versions 3.5 and higher.

Get

Add the PyPI project as a dependency and install it using pip or pipenv.

Use
import hyperbuild

try:
    minified = hyperbuild.minify("<p>  Hello, world!  </p>", minify_js=False)
except SyntaxError as e:
    print(e)
Ruby

hyperbuild is published on RubyGems, available as a native module for macOS and Linux, and supports Ruby versions 2.5 and higher.

Get

Add the library as a dependency to Gemfile or *.gemspec.

Use
require 'hyperbuild'

print Hyperbuild.minify("<p>  Hello, world!  </p>", { :minify_js => false })

Minification

Whitespace

hyperbuild has advanced context-aware whitespace minification that does things such as:

  • Leave whitespace untouched in pre and code, which are whitespace sensitive.
  • Trim and collapse whitespace in content tags, as whitespace is collapsed anyway when rendered.
  • Remove whitespace in layout tags, which allows the use of inline layouts while keeping formatted code.

Methods

There are three whitespace minification methods. When processing text content, hyperbuild chooses which ones to use depending on the containing element.

Collapse whitespace

Applies to: any element except whitespace sensitive elements.

Reduce a sequence of whitespace characters in text nodes to a single space (U+0020).

BeforeAfter
<p>↵
··The·quick·brown·fox↵
··jumps·over·the·lazy↵
··dog.↵
</p>
<p>·The·quick·brown·fox·jumps·over·the·lazy·dog.·</p>
Destroy whole whitespace

Applies to: any element except whitespace sensitive, content, content-first, and formatting elements.

Remove any text nodes between tags that only consist of whitespace characters.

BeforeAfter
<ul>↵
··<li>A</li>↵
··<li>B</li>↵
··<li>C</li></ul>
<ul>↵
··<li>A</li><li>B</li><li>C</li></ul>
Trim whitespace

Applies to: any element except whitespace sensitive and formatting elements.

Remove any leading/trailing whitespace from any leading/trailing text nodes of a tag.

BeforeAfter
<p>↵
··Hey,·I·<em>just</em>·found↵
··out·about·this·<strong>cool</strong>·website!↵
··<sup>[1]</sup></p>
<p>Hey,·I·<em>just</em>·found↵
··out·about·this·<strong>cool</strong>·website!↵
··<sup>[1]</sup></p>

Element types

hyperbuild recognises elements based on one of a few ways it assumes they are used. By making these assumptions, it can apply optimal whitespace minification strategies.

Group Elements Expected children
Formatting a, strong, and others Formatting elements, text.
Content h1, p, and others Formatting elements, text.
Layout div, ul, and others Layout elements, content elements.
Content-first label, li, and others Like content but could be layout with only one child.
Formatting elements

Whitespace is collapsed.

Formatting elements are usually inline elements that wrap around part of some text in a content element, so its whitespace isn't trimmed as they're probably part of the content.

Content elements

Whitespace is trimmed and collapsed.

Content elements usually represent a contiguous and complete unit of content such as a paragraph. As such, whitespace is significant but sequences of them are most likely due to formatting.

Before
<p>↵
··Hey,·I·<em>just</em>·found↵
··out·about·this·<strong>cool</strong>·website!↵
··<sup>[1]</sup></p>
After
<p>Hey,·I·<em>just</em>·found·out·about·this·<strong>cool</strong>·website!·<sup>[1]</sup></p>
Layout elements

Whitespace is trimmed and collapsed. Whole whitespace is removed.

These elements should only contain other elements and no text. This makes it possible to remove whole whitespace, which is useful when using display: inline-block so that whitespace between elements (e.g. indentation) does not alter layout and styling.

Before
<ul>↵
··<li>A</li>↵
··<li>B</li>↵
··<li>C</li></ul>
After
<ul><li>A</li><li>B</li><li>C</li></ul>
Content-first elements

Whitespace is trimmed and collapsed.

These elements are usually like content elements but are occasionally used like a layout element with one child. Whole whitespace is not removed as it might contain content, but this is OK for using as layout as there is only one child and whitespace is trimmed.

Before
<li>↵
··<article>↵
····<section></section>↵
····<section></section>↵
··</article></li>
After
<li><article><section></section><section></section></article></li>

Tags

Optional closing tags are removed.

Attributes

Any entities in attribute values are decoded, and then the shortest representation of the value is calculated and used:

  • Double quoted, with any " encoded.
  • Single quoted, with any ' encoded.
  • Unquoted, with "/' first character (if applicable), > last character (if applicable), and any whitespace encoded.

class and d attributes have their whitespace (after any decoding) trimmed and collapsed.

Boolean attribute values are removed. Some other attributes are completely removed if their value is empty or the default value after any processing.

type attributes on script tags with a value equaling a JavaScript MIME type are removed.

If an attribute value is empty after any processing, everything but the name is completely removed (i.e. no =), as an empty attribute is implicitly the same as an attribute with an empty string value.

Spaces are removed between attributes if possible.

Entities

Entities are decoded if valid (see relevant parsing section) and their decoded characters as UTF-8 is shorter or equal in length.

Numeric entities that do not refer to a valid Unicode Scalar Value are replaced with the replacement character.

If an entity is unintentionally formed after decoding, the leading ampersand is encoded, e.g. &&#97;&#109;&#112;; becomes &ampamp;. This is done as &amp is equal to or shorter than all other entity representations of characters part of an entity ([&#a-zA-Z0-9;]), and there is no other conflicting entity name that starts with amp.

It's possible to get an unintentional entity after removing comments, e.g. &am<!-- -->p.

Left chevrons after any decoding in text are encoded to &LT if possible or &LT; otherwise.

Comments

Comments are removed.

Ignored

Bangs, processing instructions, and empty elements are not removed as it is assumed there is a special reason for their declaration.

Parsing

Only UTF-8/ASCII-encoded HTML code is supported.

hyperbuild does no syntax checking or standards enforcement for performance and code complexity reasons.

For example, this means that it's not an error to have self-closing tags, declare multiple <body> elements, use incorrect attribute names and values, or write something like <br>alert('');</br>

However, there are some syntax requirements for speed and sanity.

Tags

Tag names are case sensitive. For example, this means that P won't be recognised as a content element, bR won't be considered as a void tag, and the contents of Script won't be parsed as JavaScript.

Tags must not be omitted. Void tags must not have a separate closing tag e.g. </input>.

Entities

Well-formed entities are decoded, including in attribute values.

They are interpreted as characters representing their decoded value. This means that &#9; is considered a whitespace character and could be minified.

Malformed entities are interpreted literally as a sequence of characters.

If a named entity is an invalid reference as per the specification, it is considered malformed.

Numeric character references that do not reference a valid Unicode Scalar Value are considered malformed.

Attributes

Backticks (`) are not valid quote marks and not interpreted as such. However, backticks are valid attribute value quotes in Internet Explorer.

Special handling of some attributes require case sensitive names and values. For example, CLASS won't be recognised as an attribute to minify, and type="Text/JavaScript" on a <script> will not be removed.

Script and style

script and style tags must be closed with </script> and </style> respectively (case sensitive).

hyperbuild does not handle escaped and double-escaped script content.

Issues and contributions

Pull requests and any contributions welcome!

If hyperbuild did something unexpected, misunderstood some syntax, or incorrectly kept/removed some code, raise an issue with some relevant code that can be used to reproduce and investigate the issue.

Dependencies

~110–720KB
~15K SLoC