README

html-min

A fast HTML5 compressor and minifier for PHP. Strips redundant whitespace, comments, optional tags, and default attributes, then sorts what's left so your gzip layer has less work to do.

Built on native \DOMDocument — no third-party DOM dependencies.

Requirements

PHP 8.3, 8.4, or 8.5
ext-dom, ext-libxml, ext-mbstring

Installation

composer require akankov/html-min

Usage

use Akankov\HtmlMin\HtmlMin;

$html = <<<HTML
<html>
  <body>
    <ul style="">
      <li style="display: inline;" class="foo">One</li>
      <li class="foo" style="display: inline;">Two</li>
    </ul>
  </body>
</html>
HTML;

echo (new HtmlMin())->minify($html);
// <html><body><ul><li class=foo style="display: inline;">One<li class=foo style="display: inline;">Two</ul>

Wrap any block in <nocompress>…</nocompress> to keep its whitespace intact.

Configuration

Every option is a chainable setter. All defaults are shown — the example below reproduces the default configuration.

$htmlMin = (new HtmlMin())
    // Core
    ->doOptimizeViaHtmlDomParser(true)   // run the DOM-based pass (required for most of the flags below)
    ->doRemoveComments(true)             // drop HTML comments (conditional comments are preserved)
    ->doSumUpWhitespace(true)            // collapse runs of whitespace in text nodes
    ->doRemoveWhitespaceAroundTags(false)// aggressive: also trim whitespace adjacent to block tags
    ->doRemoveSpacesBetweenTags(false)   // aggressive: remove whitespace-only text nodes between elements

    // Attribute optimization
    ->doOptimizeAttributes(true)
    ->doSortHtmlAttributes(true)         // canonical attribute order → better gzip
    ->doSortCssClassNames(true)          // canonical class order → better gzip
    ->doRemoveOmittedQuotes(true)        // class="foo" → class=foo when safe
    ->doRemoveOmittedHtmlTags(true)      // <p>x</p> → <p>x where the closing tag is optional
    ->doRemoveEmptyAttributes(true)
    ->doRemoveValueFromEmptyInput(true)
    ->doRemoveDefaultAttributes(false)   // opt-in: drop defaults like form method=get

    // URL attribute trimming
    ->doRemoveHttpPrefixFromAttributes(false)
    ->doRemoveHttpsPrefixFromAttributes(false)
    ->doKeepHttpAndHttpsPrefixOnExternalAttributes(false)
    ->doMakeSameDomainsLinksRelative([])       // e.g. ['example.com'] → strip host from same-site links

    // Deprecated attribute cleanup
    ->doRemoveDeprecatedAnchorName(true)
    ->doRemoveDeprecatedScriptCharsetAttribute(true)
    ->doRemoveDeprecatedTypeFromScriptTag(true)
    ->doRemoveDeprecatedTypeFromStylesheetLink(true)
    ->doRemoveDeprecatedTypeFromStyleAndLinkTag(true)
    ->doRemoveDefaultMediaTypeFromStyleAndLinkTag(true)
    ->doRemoveDefaultTypeFromButton(false)

    // Inline CSS / JS minification (opt-in, off by default)
    ->doMinifyInlineCss(false)           // minify the contents of inline <style> blocks
    ->doMinifyInlineJs(false);           // minify the contents of inline <script> blocks

echo $htmlMin->minify($html);

Each setter returns $this, so you can configure and call minify() in one chain.

Inline CSS and JS minification

By default the contents of <style> and <script> blocks round-trip untouched. Enable the two opt-in toggles to minify them:

$htmlMin = (new HtmlMin())
    ->doMinifyInlineCss(true)
    ->doMinifyInlineJs(true);

echo $htmlMin->minify($html);

The bundled minifiers are zero-dependency and conservative:

CSS — strips /* … */ comments and collapses whitespace; the contents of strings and url(…) are preserved.
JS — removes comments and collapses horizontal whitespace while preserving newlines (so Automatic Semicolon Insertion is unaffected), strings, regex literals, and template literals. Identifiers are never renamed.

Scripts that are not JavaScript are left alone automatically: a <script> whose type is, for example, application/ld+json or text/x-template, and any <script src="…">, passes through unminified.

Using a different minifier

For aggressive minification (identifier renaming, dead-code removal), plug in a third-party tool with setInlineCssMinifier() / setInlineJsMinifier(). Each takes any callable(string): string; pass null to restore the bundled default.

use MatthiasMullie\Minify; // composer require matthiasmullie/minify

$htmlMin = (new HtmlMin())
    ->doMinifyInlineCss(true)
    ->doMinifyInlineJs(true)
    ->setInlineCssMinifier(static fn (string $css): string => (new Minify\CSS($css))->minify())
    ->setInlineJsMinifier(static fn (string $js): string => (new Minify\JS($js))->minify());

echo $htmlMin->minify($html);

If a bundled minifier throws, the original source is kept and a warning is sent to the PSR-3 logger (when one is set via setLogger()), so a minifier bug can never corrupt the page. User-supplied callables let their exceptions propagate.

Extending

To run your own pass over every element during minification, implement Akankov\HtmlMin\Contract\DomObserver and register it:

use Akankov\HtmlMin\Contract\DomObserver;
use Akankov\HtmlMin\Contract\HtmlMinInterface;
use Akankov\HtmlMin\HtmlMin;

final class StripDataTestIds implements DomObserver
{
    public function domElementBeforeMinification(\DOMElement $element, HtmlMinInterface $htmlMin): void
    {
    }

    public function domElementAfterMinification(\DOMElement $element, HtmlMinInterface $htmlMin): void
    {
        if ($element->hasAttribute('data-testid')) {
            $element->removeAttribute('data-testid');
        }
    }
}

$htmlMin = new HtmlMin();
$htmlMin->attachObserverToTheDomLoop(new StripDataTestIds());
echo $htmlMin->minify($html);

Benchmarks

Measured against voku/html-min, wyrihaximus/html-compress, zaininnari/html-minifier, and abordage/html-min on a corpus of real-world HTML pages.

adapter	median ms/op	geomean ms/op	parse failures	avg gzipped ratio
akankov/html-min	1.9	1.9	0 / 15	90.7%
akankov/html-min (inline)	2.1	2.5	0 / 15	87.6%
voku/html-min	3.3	3.6	0 / 15	90.7%
wyrihaximus/html-compress	6.0	7.5	0 / 15	87.0%
zaininnari/html-minifier	9.5	8.3	0 / 15	94.8%
abordage/html-min †	0.2	0.2	0 / 15	90.2%

The table above is regenerated by make bench from the latest run. See latest.md for the per-fixture detail (speed, peak memory, gzipped compression ratio, methodology, and non-claims). Reproduce with make bench-install && make bench (requires Docker).

Development

composer install
make md-check                 # markdown formatting (Docker)
vendor/bin/phpunit            # tests
vendor/bin/phpstan analyse    # static analysis (level max)
vendor/bin/php-cs-fixer fix   # code style

CI runs the full matrix (PHP 8.3 / 8.4 / 8.5) on every push and pull request.

License

MIT — see LICENSE.

Originally authored by Lars Moelleken; maintained in this fork by Alex Kankov.

akankov / html-min

Maintainers

Package info

Fund package maintenance!

Statistics

Security