A basic, PHP 5.2+ compatible, HTML parser.

Brain_HtmlTokenizer.php

<?php
/**
 * Brain_HtmlTokenizer class file.
 *
 * (c) Giuseppe Mazzapica
 *
 * @license http://opensource.org/licenses/MIT MIT
 * @author  Giuseppe Mazzapica <giuseppe.mazzapica@gmail.com>
 */

/**
 * A basic, PHP 5.2+ compatible, HTML parser.
 */
class Brain_HtmlTokenizer {


	/**
	 * Parsed HTML tokens.
	 *
	 * @var array[]
	 */
	private $tokens = array();

	/**
	 * Named constructor, creates an instance from given HTML string.
	 *
	 * @param string $html HTML string to created instance from.
	 *
	 * @return Brain_HtmlTokenizer New Instance.
	 */
	public static function load_html( $html ) {

		if ( ! is_string( $html ) || ! $html ) {
			return new static( array() );
		}

		$instance         = new static();
		$instance->tokens = $instance->tokenize( $html );

		return $instance;
	}

	/**
	 * Private on purpose, use named constructor;
	 *
	 * @param array[] $tokens HTML tokens.
	 */
	private function __construct( array $tokens = array() ) {

		$this->tokens = $tokens;
	}

	/**
	 * Return (maybe) processed HTML string.
	 *
	 * @return string processed HTML string.
	 */
	public function __toString() {

		return implode( '', array_map( array( $this, 'serialize_token' ), $this->tokens ) );
	}

	/**
	 * Strips given HTML tags.
	 *
	 * @param string[] $tags HTML tags to remove.
	 *
	 * @return Brain_HtmlTokenizer Instance without given tags.
	 */
	public function strip_tags( array $tags ) {

		if ( ! $tags ) {
			return $this;
		}

		$tokens = array();
		foreach ( $this->tokens as $i => $token ) {

			list($tag) = $token['tag'];
			if (
				( ! $tag && ! array_intersect( $token['parents'], $tags ) )
				|| ( $tag && ! in_array( $tag, $tags, true )
					&& ! in_array( $tag, $token['parents'], true ) )
			) {
				$tokens[ $i ] = $token;
			}
		}

		$this->tokens = $tokens;

		return $this;
	}

	/**
	 * Keep only given HTML tags and remove all the others.
	 *
	 * @param string[] $tags HTML tags to keep.
	 *
	 * @return Brain_HtmlTokenizer Instance with only given tags.
	 */
	public function keep_tags( array $tags ) {

		$tokens = array();
		foreach ( $this->tokens as $i => $token ) {

			if ( array_diff( $token['parents'], $tags ) ) {
				continue;
			}

			list($tag) = $token['tag'];
			if ( ! $tag && $token['parents'] ) {
				$tag = end( $token['parents'] );
			}

			if ( ! $tag || in_array( $tag, $tags, true ) ) {
				$tokens[ $i ] = $token;
			}
		}

		$this->tokens = $tokens;

		return $this;
	}

	/**
	 * Strips given attributes from given HTML tags (or all of them).
	 *
	 * @param array      $attributes Attributes to remove.
	 * @param array|null $tags       HTML tags to remove attributes from. When null (default),
	 *                               means "all of them".
	 *
	 * @return Brain_HtmlTokenizer Instance without given attributes in given tags.
	 */
	public function strip_attributes( array $attributes, array $tags = null ) {

		if ( array() === $tags || array() === $attributes ) {
			return $this;
		}

		$attributes = array_filter( array_filter( array_values( $attributes ), 'is_string' ) );
		if ( ! $attributes || array() === $tags ) {
			return $this;
		}

		$tokens = array();
		foreach ( $this->tokens as $i => $token ) {

			list($tag) = $token['tag'];
			if (
				! $tag
				|| ! $token['attributes']
				|| ( is_array( $tags ) && ! in_array( $tag, $tags, true ) )
			) {
				$tokens[ $i ] = $token;
				continue;
			}

			$attributes_blacklist = array_fill_keys( $attributes, '' );
			$token['attributes']  = array_diff_key( $token['attributes'], $attributes_blacklist );

			$tokens[ $i ] = $token;
		}

		$this->tokens = $tokens;

		return $this;
	}

	/**
	 * Keep only given attributes in given HTML tags and remove all the others.
	 *
	 * @param array      $attributes Attributes to keep.
	 * @param array|null $tags       HTML tags to keep attributes for. When null (default), means
	 *                               "all of them".
	 *
	 * @return Brain_HtmlTokenizer Instance with only given attributes in given tags.
	 */
	public function keep_attributes( array $attributes, array $tags = null ) {

		if ( array() === $tags ) {
			return $this;
		}

		$attributes = array_filter( array_filter( array_values( $attributes ), 'is_string' ) );
		$tokens     = array();
		foreach ( $this->tokens as $i => $token ) {

			list($tag) = $token['tag'];
			if (
				! $tag
				|| ! $token['attributes']
				|| ( is_array( $tags ) && ! in_array( $tag, $tags, true ) )
			) {
				$tokens[ $i ] = $token;
				continue;
			}

			if ( ! $attributes ) {
				$token['attributes'] = array();
				$tokens[ $i ]        = $token;
				continue;
			}

			$attributes_whitelist = array_fill_keys( $attributes, '' );
			$token['attributes']  = array_intersect_key(
				$token['attributes'],
				$attributes_whitelist
			);

			$tokens[ $i ] = $token;
		}

		$this->tokens = $tokens;

		return $this;
	}

	/**
	 * Process the given HTML string and build a stack of token.
	 *
	 * @param string $html HTML string to process.
	 *
	 * @return array[] Array of tokens, each is an array with keys: 'tag', 'content', and
	 *                 'attributes'.
	 */
	private function tokenize( $html ) {

		$first_pass = array();
		$tokens     = array();
		$parents    = array();
		$token      = strtok( $html, '<' );
		while ( false !== $token ) {
			$first_pass[] = $token;
			$token        = strtok( '<' );
		}

		foreach ( $first_pass as $i => $part ) {

			$token_parts = explode( '>', $part, 2 );
			if ( array( $part ) === $token_parts ) {
				$tokens[] = $this->no_tag_token( $part, $parents );
				continue;
			}

			list($tag_name, $tag_type, $attributes) = $this->tokenize_tag( $token_parts[0] );

			if ( $parents && 'close' === $tag_type && end( $parents ) === $tag_name ) {
				array_pop( $parents );
			}

			$tokens[] = array(
				'tag'        => array( $tag_name, $tag_type ),
				'attributes' => 'close' !== $tag_type ? $this->tokenize_attributes( $attributes )
					: array(),
				'content'    => '',
				'parents'    => $parents,
			);

			if ( 'open' === $tag_type ) {
				$parents[] = $tag_name;
			}

			$content = $token_parts[1];
			if ( $content ) {
				$tokens[] = $this->no_tag_token( $content, $parents );
			}
		}

		return $tokens;
	}

	/**
	 * Creates a token for a string that does not belong to any tag.
	 *
	 * @param string $content HTML chunk content.
	 * @param array  $parents HTML tags stack HTML chunk content belongs to.
	 *
	 * @return array
	 */
	private function no_tag_token( $content, array $parents ) {

		return array(
			'tag'        => array( null, null ),
			'attributes' => null,
			'content'    => $content,
			'parents'    => $parents,
		);
	}

	/**
	 * Process given HTML tag string and return tag name, tag type (open, close, self-close) and
	 * the "raw" attributes.
	 *
	 * @param string $tag Piece of HTML that contains, one HTML tag.
	 *
	 * @return string[] A 3 items array:
	 *                  - Tag name, e.g. "div"
	 *                  - Tag type, can be "open", "close", and "self-close"
	 *                  - Not parsed attributes string
	 */
	private function tokenize_tag( $tag ) {

		$self_close = substr( $tag, -1, 1 ) === '/';
		$tag_type   = '';
		if ( $self_close ) {
			$tag_type = 'self-close';
			$tag      = substr( $tag, 0, -1 );
		}

		$split      = preg_split( '~\s+~', trim( $tag ), 2 );
		$attributes = count( $split ) === 2 ? end( $split ) : '';
		$tag_name   = reset( $split );
		if ( ! $self_close ) {
			$tag_type = substr( $tag_name, 0, 1 ) === '/' ? 'close' : 'open';
		}

		return array( trim( $tag_name, '/' ), $tag_type, $attributes );
	}

	/**
	 * Process given HTML tag attributes string and return an array of processed attributes in
	 * array form.
	 *
	 * @param string $attr   HTML tag attributes string to process.
	 * @param array  $tokens Array of parsed tag attributes, keys are attributes names, values
	 *                       attributes values.
	 *
	 * @return array
	 */
	private function tokenize_attributes( $attr, array $tokens = null ) {

		if ( is_null( $tokens ) ) {
			$tokens = array();
			$attr   = preg_replace( '~\s+~', ' ', $attr );
		}

		$attr = trim( $attr );
		if ( ! $attr ) {
			return $tokens;
		}

		$search_eq    = strpos( $attr, '=' );
		$search_space = strpos( $attr, ' ' );

		// Flag and last attribute.
		if ( false === $search_space && false === $search_eq ) {
			$tokens[ $attr ] = true;

			return $tokens;
		}

		// Flag.
		if ( false !== $search_space && ( false === $search_eq || ( $search_space < $search_eq ) ) ) {
			$attr_parts               = explode( ' ', $attr, 2 );
			$tokens[ $attr_parts[0] ] = true;
			if ( empty( $attr_parts[1] ) ) {
				return $tokens;
			}

			return $this->tokenize_attributes( $attr_parts[1], $tokens );
		}

		$attr_key    = strtok( $attr, '=' );
		$token_split = $this->attr_value_token_split( $attr );
		$attr_value  = strtok( $token_split );

		$tokens[ $attr_key ] = $this->strip_quote( $attr_value );

		$cursor   = strlen( $attr_key ) + strlen( $attr_value );
		$cursor  += ' ' === $token_split ? 2 : 3;
		$to_parse = substr( $attr, $cursor );

		if ( $to_parse ) {
			return $this->tokenize_attributes( $to_parse, $tokens );
		}

		return $tokens;
	}

	/**
	 * Calculate token value to use with `strtok` to split attribute key from value.
	 *
	 * It is necessary because attributes might or might not have wrapping quotes around values.
	 *
	 * @param string $attributes_string Raw attributes string.
	 *
	 * @return string The token value to use with `strtok` to split attribute key from value
	 */
	private function attr_value_token_split( $attributes_string ) {

		$double_quote_index = strpos( $attributes_string, '"' );
		$single_quote_index = strpos( $attributes_string, '\'' );

		$quotes = array_filter( compact( 'double_quote_index', 'single_quote_index' ), 'is_int' );
		if ( ! $quotes ) {
			return ' ';
		}

		$first_quote = min( $quotes );
		if ( $first_quote > strpos( $attributes_string, ' ' ) ) {
			return ' ';
		}

		return array_search( $first_quote, $quotes, true ) === 'double_quote_index' ? '"' : "'";

	}

	/**
	 * Remove wrapping quotes from attributes value, if any.
	 *
	 * Also ensure the value is suitable to be wrapped in double quotes ie. does not contain
	 * non-slashed double quotes.
	 *
	 * @param string $attribute_value Attribute value.
	 *
	 * @return string Attribute value without wrapping quotes.
	 */
	private function strip_quote( $attribute_value ) {

		$first = substr( $attribute_value, 0, 1 );
		$last  = substr( $attribute_value, -1, 1 );
		if ( $first === $last && in_array( $first, array( '"', '\'' ), true ) ) {
			$attribute_value = substr( $attribute_value, 1, -1 );
		}

		if ( substr_count( $attribute_value, '"' ) !== substr_count( $attribute_value, '\"' ) ) {
			$attribute_value = str_replace( '\"', '"', $attribute_value );
			$attribute_value = str_replace( '"', '\"', $attribute_value );
		}

		return $attribute_value;
	}

	/**
	 * Take a processed token in array form and serialize back to string.
	 *
	 * @param array $token HTML token.
	 *
	 * @return string
	 */
	private function serialize_token( array $token ) {

		list($tag, $type) = $token['tag'];
		$content          = $token['content'];

		if ( ! $tag ) {
			return $content;
		}

		if ( 'close' === $type ) {
			return "</{$tag}>{$content}";
		}

		$attributes = $token['attributes'];
		$string     = "<{$tag}";
		foreach ( $attributes as $name => $value ) {
			if ( true === $value ) {
				$string .= " {$name}";
				continue;
			}
			$string .= " {$name}=\"{$value}\"";
		}

		return 'self-close' === $type ? "{$string}/>{$content}" : "{$string}>{$content}";

	}
}

The class takes an HTML string and tokenize it.

Few methods are there to process generated token and either whitelist or blacklist tags and attributes.
The __toString() transform can be called to obtain the (maybe) processed HTML string.

For example, assuming following HTML:

<div class="wrapper" id="0">
    <p id="1">Paragraph <strong>bold</strong> normal.</p>
    <p id="2"><strong>bold</strong><span id="3" class="foo">A span!</span></p>
    <input id="4" type="hidden" value="x" />
    <em id="5" class="italic">Italic</em>
    <script src="http://example.com/example.js" />
</div>

And the following code:

$processed = Brain_HtmlTokenizer::load_html( $html )  // HTML from above
    ->keep_tags( array( 'div', 'p', 'strong', 'em' ) )
    ->strip_attributes( array( 'class' ), array( 'div' ) )
    ->strip_attributes( array( 'id' ), array( 'p' ) )
    ->keep_attributes( array( 'class' ), array( 'em' ) )
    ->__toString();

$processed variable will be equal to:

<div id="0">
    <p>Paragraph <strong>bold</strong> normal.</p>
    <p><strong>bold</strong></p>
    <em class="italic">Italic</em>
</div>

Another example could be the implementation of genesis_strip_attr function from Genesis framework. It would be:

function genesis_strip_attr( $text, $tags, $attributes ) {

    return (string) Brain_HtmlTokenizer::load_html( $text )->strip_attributes( (array) $tags, (array) $attributes ) );
}

Could this be amended to also include a way to add attributes to tags?

@GaryJones yes, sure not hard.

It would be something like this:

/**
 * @param  array $attributes Associative array, where keys are attributes keys and values attributes values.
 *                           For flags, use `true` as value.
 */
public function add_attributes( array $attributes, array $tags = array()) {

    if ( ! $tags || array() === $attributes ) {
        return $this;
     }

     $tokens = array();
     foreach ( $this->tokens as $i => $token ) {
         list($tag) = $token['tag'];
         if ( ! $tag || ! in_array( $tag, $tags, true ) ) {
             $tokens[ $i ] = $token;
             continue;
          }

          foreach ( $attributes as $name => $value ) {
              if ( is_string( $value ) || true === $value ) {
                  $tag['attributes'][$name] = true === $value ? $value : esc_attr( $value );
              }
          }

          $tokens[ $i ] = $token;
     }
     
     $this->tokens = $tokens;
     return $this;
}

Note that is completely untested, typed directly in this comment, and because of esc_attr, coupled to WordPress (unlike the rest of the class).