` (with restrictions [2]).
 *  - The empty end tag `</>` which is ignored in the browser and DOM.
 *
 * [1]: There are no CDATA sections in HTML. When encountering `<![CDATA[`, everything
 *      until the next `>` becomes a bogus HTML comment, meaning there can be no CDATA
 *      section in an HTML document containing `>`. The Tag Processor will first find
 *      all valid and bogus HTML comments, and then if the comment _would_ have been a
 *      CDATA section _were they to exist_, it will indicate this as the type of comment.
 *
 * [2]: XML allows a broader range of characters in a processing instruction's target name
 *      and disallows "xml" as a name, since it's special. The Tag Processor only recognizes
 *      target names with an ASCII-representable subset of characters. It also exhibits the
 *      same constraint as with CDATA sections, in that `>` cannot exist within the token
 *      since Processing Instructions do no exist within HTML and their syntax transforms
 *      into a bogus comment in the DOM.
 *
 * ## Design and limitations
 *
 * The Tag Processor is designed to linearly scan HTML documents and tokenize
 * HTML tags and their attributes. It's designed to do this as efficiently as
 * possible without compromising parsing integrity. Therefore it will be
 * slower than some methods of modifying HTML, such as those incorporating
 * over-simplified PCRE patterns, but will not introduce the defects and
 * failures that those methods bring in, which lead to broken page renders
 * and often to security vulnerabilities. On the other hand, it will be faster
 * than full-blown HTML parsers such as DOMDocument and use considerably
 * less memory. It requires a negligible memory overhead, enough to consider
 * it a zero-overhead system.
 *
 * The performance characteristics are maintained by avoiding tree construction
 * and semantic cleanups which are specified in HTML5. Because of this, for
 * example, it's not possible for the Tag Processor to associate any given
 * opening tag with its corresponding closing tag, or to return the inner markup
 * inside an element. Systems may be built on top of the Tag Processor to do
 * this, but the Tag Processor is and should be constrained so it can remain an
 * efficient, low-level, and reliable HTML scanner.
 *
 * The Tag Processor's design incorporates a "garbage-in-garbage-out" philosophy.
 * HTML5 specifies that certain invalid content be transformed into different forms
 * for display, such as removing null bytes from an input document and replacing
 * invalid characters with the Unicode replacement character `U+FFFD` (visually "�").
 * Where errors or transformations exist within the HTML5 specification, the Tag Processor
 * leaves those invalid inputs untouched, passing them through to the final browser
 * to handle. While this implies that certain operations will be non-spec-compliant,
 * such as reading the value of an attribute with invalid content, it also preserves a
 * simplicity and efficiency for handling those error cases.
 *
 * Most operations within the Tag Processor are designed to minimize the difference
 * between an input and output document for any given change. For example, the
 * `add_class` and `remove_class` methods preserve whitespace and the class ordering
 * within the `class` attribute; and when encountering tags with duplicated attributes,
 * the Tag Processor will leave those invalid duplicate attributes where they are but
 * update the proper attribute which the browser will read for parsing its value. An
 * exception to this rule is that all attribute updates store their values as
 * double-quoted strings, meaning that attributes on input with single-quoted or
 * unquoted values will appear in the output with double-quotes.
 *
 * ### Scripting Flag
 *
 * The Tag Processor parses HTML with the "scripting flag" disabled. This means
 * that it doesn't run any scripts while parsing the page. In a browser with
 * JavaScript enabled, for example, the script can change the parse of the
 * document as it loads. On the server, however, evaluating JavaScript is not
 * only impractical, but also unwanted.
 *
 * Practically this means that the Tag Processor will descend into NOSCRIPT
 * elements and process its child tags. Were the scripting flag enabled, such
 * as in a typical browser, the contents of NOSCRIPT are skipped entirely.
 *
 * This allows the HTML API to process the content that will be presented in
 * a browser when scripting is disabled, but it offers a different view of a
 * page than most browser sessions will experience. E.g. the tags inside the
 * NOSCRIPT disappear.
 *
 * ### Text Encoding
 *
 * The Tag Processor assumes that the input HTML document is encoded with a
 * text encoding compatible with 7-bit ASCII's '<', '>', '&', ';', '/', '=',
 * "'", '"', 'a' - 'z', 'A' - 'Z', and the whitespace characters ' ', tab,
 * carriage-return, newline, and form-feed.
 *
 * In practice, this includes almost every single-byte encoding as well as
 * UTF-8. Notably, however, it does not include UTF-16. If providing input
 * that's incompatible, then convert the encoding beforehand.
 *
 * @since 6.2.0
 * @since 6.2.1 Fix: Support for various invalid comments; attribute updates are case-insensitive.
 * @since 6.3.2 Fix: Skip HTML-like content inside rawtext elements such as STYLE.
 * @since 6.5.0 Pauses processor when input ends in an incomplete syntax token.
 *              Introduces "special" elements which act like void elements, e.g. TITLE, STYLE.
 *              Allows scanning through all tokens and processing modifiable text, where applicable.
 */
class WP_HTML_Tag_Processor {
	
	const MAX_BOOKMARKS = 10;

	
	const MAX_SEEK_OPS = 1000;

	
	protected $html;

	
	private $last_query;

	
	private $sought_tag_name;

	
	private $sought_class_name;

	
	private $sought_match_offset;

	
	private $stop_on_tag_closers;

	
	protected $parser_state = self::STATE_READY;

	
	protected $compat_mode = self::NO_QUIRKS_MODE;

	
	private $parsing_namespace = 'html';

	
	protected $comment_type = null;

	
	protected $text_node_classification = self::TEXT_IS_GENERIC;

	
	private $bytes_already_parsed = 0;

	
	private $token_starts_at;

	
	private $token_length;

	
	private $tag_name_starts_at;

	
	private $tag_name_length;

	
	private $text_starts_at;

	
	private $text_length;

	
	private $is_closing_tag;

	
	private $attributes = array();

	
	private $duplicate_attributes = null;

	
	private $classname_updates = array();

	
	protected $bookmarks = array();

	const ADD_CLASS    = true;
	const REMOVE_CLASS = false;
	const SKIP_CLASS   = null;

	
	protected $lexical_updates = array();

	
	protected $seek_count = 0;

	
	private $skip_newline_at = null;

	
	public function __construct( $html ) {
		$this->html = $html;
	}

	
	public function change_parsing_namespace( string $new_namespace ): bool {
		if ( ! in_array( $new_namespace, array( 'html', 'math', 'svg' ), true ) ) {
			return false;
		}

		$this->parsing_namespace = $new_namespace;
		return true;
	}

	
	public function next_tag( $query = null ): bool {
		$this->parse_query( $query );
		$already_found = 0;

		do {
			if ( false === $this->next_token() ) {
				return false;
			}

			if ( self::STATE_MATCHED_TAG !== $this->parser_state ) {
				continue;
			}

			if ( $this->matches() ) {
				++$already_found;
			}
		} while ( $already_found < $this->sought_match_offset );

		return true;
	}

	
	public function next_token(): bool {
		return $this->base_class_next_token();
	}

	
	private function base_class_next_token(): bool {
		$was_at = $this->bytes_already_parsed;
		$this->after_tag();

		// Don't proceed if there's nothing more to scan.
		if (
			self::STATE_COMPLETE === $this->parser_state ||
			self::STATE_INCOMPLETE_INPUT === $this->parser_state
		) {
			return false;
		}

		/*
		 * The next step in the parsing loop determines the parsing state;
		 * clear it so that state doesn't linger from the previous step.
		 */
		$this->parser_state = self::STATE_READY;

		if ( $this->bytes_already_parsed >= strlen( $this->html ) ) {
			$this->parser_state = self::STATE_COMPLETE;
			return false;
		}

		// Find the next tag if it exists.
		if ( false === $this->parse_next_tag() ) {
			if ( self::STATE_INCOMPLETE_INPUT === $this->parser_state ) {
				$this->bytes_already_parsed = $was_at;
			}

			return false;
		}

		/*
		 * For legacy reasons the rest of this function handles tags and their
		 * attributes. If the processor has reached the end of the document
		 * or if it matched any other token then it should return here to avoid
		 * attempting to process tag-specific syntax.
		 */
		if (
			self::STATE_INCOMPLETE_INPUT !== $this->parser_state &&
			self::STATE_COMPLETE !== $this->parser_state &&
			self::STATE_MATCHED_TAG !== $this->parser_state
		) {
			return true;
		}

		// Parse all of its attributes.
		while ( $this->parse_next_attribute() ) {
			continue;
		}

		// Ensure that the tag closes before the end of the document.
		if (
			self::STATE_INCOMPLETE_INPUT === $this->parser_state ||
			$this->bytes_already_parsed >= strlen( $this->html )
		) {
			// Does this appropriately clear state (parsed attributes)?
			$this->parser_state         = self::STATE_INCOMPLETE_INPUT;
			$this->bytes_already_parsed = $was_at;

			return false;
		}

		$tag_ends_at = strpos( $this->html, '>', $this->bytes_already_parsed );
		if ( false === $tag_ends_at ) {
			$this->parser_state         = self::STATE_INCOMPLETE_INPUT;
			$this->bytes_already_parsed = $was_at;

			return false;
		}
		$this->parser_state         = self::STATE_MATCHED_TAG;
		$this->bytes_already_parsed = $tag_ends_at + 1;
		$this->token_length         = $this->bytes_already_parsed - $this->token_starts_at;

		/*
		 * Certain tags require additional processing. The first-letter pre-check
		 * avoids unnecessary string allocation when comparing the tag names.
		 *
		 *  - IFRAME
		 *  - LISTING (deprecated)
		 *  - NOEMBED (deprecated)
		 *  - NOFRAMES (deprecated)
		 *  - PRE
		 *  - SCRIPT
		 *  - STYLE
		 *  - TEXTAREA
		 *  - TITLE
		 *  - XMP (deprecated)
		 */
		if (
			$this->is_closing_tag ||
			'html' !== $this->parsing_namespace ||
			1 !== strspn( $this->html, 'iIlLnNpPsStTxX', $this->tag_name_starts_at, 1 )
		) {
			return true;
		}

		$tag_name = $this->get_tag();

		/*
		 * For LISTING, PRE, and TEXTAREA, the first linefeed of an immediately-following
		 * text node is ignored as an authoring convenience.
		 *
		 * @see static::skip_newline_at
		 */
		if ( 'LISTING' === $tag_name || 'PRE' === $tag_name ) {
			$this->skip_newline_at = $this->bytes_already_parsed;
			return true;
		}

		/*
		 * There are certain elements whose children are not DATA but are instead
		 * RCDATA or RAWTEXT. These cannot contain other elements, and the contents
		 * are parsed as plaintext, with character references decoded in RCDATA but
		 * not in RAWTEXT.
		 *
		 * These elements are described here as "self-contained" or special atomic
		 * elements whose end tag is consumed with the opening tag, and they will
		 * contain modifiable text inside of them.
		 *
		 * Preserve the opening tag pointers, as these will be overwritten
		 * when finding the closing tag. They will be reset after finding
		 * the closing to tag to point to the opening of the special atomic
		 * tag sequence.
		 */
		$tag_name_starts_at   = $this->tag_name_starts_at;
		$tag_name_length      = $this->tag_name_length;
		$tag_ends_at          = $this->token_starts_at + $this->token_length;
		$attributes           = $this->attributes;
		$duplicate_attributes = $this->duplicate_attributes;

		// Find the closing tag if necessary.
		switch ( $tag_name ) {
			case 'SCRIPT':
				$found_closer = $this->skip_script_data();
				break;

			case 'TEXTAREA':
			case 'TITLE':
				$found_closer = $this->skip_rcdata( $tag_name );
				break;

			/*
			 * In the browser this list would include the NOSCRIPT element,
			 * but the Tag Processor is an environment with the scripting
			 * flag disabled, meaning that it needs to descend into the
			 * NOSCRIPT element to be able to properly process what will be
			 * sent to a browser.
			 *
			 * Note that this rule makes HTML5 syntax incompatible with XML,
			 * because the parsing of this token depends on client application.
			 * The NOSCRIPT element cannot be represented in the XHTML syntax.
			 */
			case 'IFRAME':
			case 'NOEMBED':
			case 'NOFRAMES':
			case 'STYLE':
			case 'XMP':
				$found_closer = $this->skip_rawtext( $tag_name );
				break;

			// No other tags should be treated in their entirety here.
			default:
				return true;
		}

		if ( ! $found_closer ) {
			$this->parser_state         = self::STATE_INCOMPLETE_INPUT;
			$this->bytes_already_parsed = $was_at;
			return false;
		}

		/*
		 * The values here look like they reference the opening tag but they reference
		 * the closing tag instead. This is why the opening tag values were stored
		 * above in a variable. It reads confusingly here, but that's because the
		 * functions that skip the contents have moved all the internal cursors past
		 * the inner content of the tag.
		 */
		$this->token_starts_at      = $was_at;
		$this->token_length         = $this->bytes_already_parsed - $this->token_starts_at;
		$this->text_starts_at       = $tag_ends_at;
		$this->text_length          = $this->tag_name_starts_at - $this->text_starts_at;
		$this->tag_name_starts_at   = $tag_name_starts_at;
		$this->tag_name_length      = $tag_name_length;
		$this->attributes           = $attributes;
		$this->duplicate_attributes = $duplicate_attributes;

		return true;
	}

	
	public function paused_at_incomplete_token(): bool {
		return self::STATE_INCOMPLETE_INPUT === $this->parser_state;
	}

	
	public function class_list() {
		if ( self::STATE_MATCHED_TAG !== $this->parser_state ) {
			return;
		}

		
		$class = $this->get_attribute( 'class' );

		if ( ! is_string( $class ) ) {
			return;
		}

		$seen = array();

		$is_quirks = self::QUIRKS_MODE === $this->compat_mode;

		$at = 0;
		while ( $at < strlen( $class ) ) {
			// Skip past any initial boundary characters.
			$at += strspn( $class, " \t\f\r\n", $at );
			if ( $at >= strlen( $class ) ) {
				return;
			}

			// Find the byte length until the next boundary.
			$length = strcspn( $class, " \t\f\r\n", $at );
			if ( 0 === $length ) {
				return;
			}

			$name = str_replace( "\x00", "\u{FFFD}", substr( $class, $at, $length ) );
			if ( $is_quirks ) {
				$name = strtolower( $name );
			}
			$at += $length;

			/*
			 * It's expected that the number of class names for a given tag is relatively small.
			 * Given this, it is probably faster overall to scan an array for a value rather
			 * than to use the class name as a key and check if it's a key of $seen.
			 */
			if ( in_array( $name, $seen, true ) ) {
				continue;
			}

			$seen[] = $name;
			yield $name;
		}
	}


	
	public function has_class( $wanted_class ): ?bool {
		if ( self::STATE_MATCHED_TAG !== $this->parser_state ) {
			return null;
		}

		$case_insensitive = self::QUIRKS_MODE === $this->compat_mode;

		$wanted_length = strlen( $wanted_class );
		foreach ( $this->class_list() as $class_name ) {
			if (
				strlen( $class_name ) === $wanted_length &&
				0 === substr_compare( $class_name, $wanted_class, 0, strlen( $wanted_class ), $case_insensitive )
			) {
				return true;
			}
		}

		return false;
	}


	
	public function set_bookmark( $name ): bool {
		// It only makes sense to set a bookmark if the parser has paused on a concrete token.
		if (
			self::STATE_COMPLETE === $this->parser_state ||
			self::STATE_INCOMPLETE_INPUT === $this->parser_state
		) {
			return false;
		}

		if ( ! array_key_exists( $name, $this->bookmarks ) && count( $this->bookmarks ) >= static::MAX_BOOKMARKS ) {
			_doing_it_wrong(
				__METHOD__,
				__( 'Too many bookmarks: cannot create any more.' ),
				'6.2.0'
			);
			return false;
		}

		$this->bookmarks[ $name ] = new WP_HTML_Span( $this->token_starts_at, $this->token_length );

		return true;
	}


	
	public function release_bookmark( $name ): bool {
		if ( ! array_key_exists( $name, $this->bookmarks ) ) {
			return false;
		}

		unset( $this->bookmarks[ $name ] );

		return true;
	}

	
	private function skip_rawtext( string $tag_name ): bool {
		/*
		 * These two functions distinguish themselves on whether character references are
		 * decoded, and since functionality to read the inner markup isn't supported, it's
		 * not necessary to implement these two functions separately.
		 */
		return $this->skip_rcdata( $tag_name );
	}

	
	private function skip_rcdata( string $tag_name ): bool {
		$html       = $this->html;
		$doc_length = strlen( $html );
		$tag_length = strlen( $tag_name );

		$at = $this->bytes_already_parsed;

		while ( false !== $at && $at < $doc_length ) {
			$at                       = strpos( $this->html, '</', $at );
			$this->tag_name_starts_at = $at;

			// Fail if there is no possible tag closer.
			if ( false === $at || ( $at + $tag_length ) >= $doc_length ) {
				return false;
			}

			$at += 2;

			/*
			 * Find a case-insensitive match to the tag name.
			 *
			 * Because tag names are limited to US-ASCII there is no
			 * need to perform any kind of Unicode normalization when
			 * comparing; any character which could be impacted by such
			 * normalization could not be part of a tag name.
			 */
			for ( $i = 0; $i < $tag_length; $i++ ) {
				$tag_char  = $tag_name[ $i ];
				$html_char = $html[ $at + $i ];

				if ( $html_char !== $tag_char && strtoupper( $html_char ) !== $tag_char ) {
					$at += $i;
					continue 2;
				}
			}

			$at                        += $tag_length;
			$this->bytes_already_parsed = $at;

			if ( $at >= strlen( $html ) ) {
				return false;
			}

			/*
			 * Ensure that the tag name terminates to avoid matching on
			 * substrings of a longer tag name. For example, the sequence
			 * "</textarearug" should not match for "</textarea" even
			 * though "textarea" is found within the text.
			 */
			$c = $html[ $at ];
			if ( ' ' !== $c && "\t" !== $c && "\r" !== $c && "\n" !== $c && '/' !== $c && '>' !== $c ) {
				continue;
			}

			while ( $this->parse_next_attribute() ) {
				continue;
			}

			$at = $this->bytes_already_parsed;
			if ( $at >= strlen( $this->html ) ) {
				return false;
			}

			if ( '>' === $html[ $at ] ) {
				$this->bytes_already_parsed = $at + 1;
				return true;
			}

			if ( $at + 1 >= strlen( $this->html ) ) {
				return false;
			}

			if ( '/' === $html[ $at ] && '>' === $html[ $at + 1 ] ) {
				$this->bytes_already_parsed = $at + 2;
				return true;
			}
		}

		return false;
	}

	
	private function skip_script_data(): bool {
		$state      = 'unescaped';
		$html       = $this->html;
		$doc_length = strlen( $html );
		$at         = $this->bytes_already_parsed;

		while ( false !== $at && $at < $doc_length ) {
			$at += strcspn( $html, '-<', $at );

			/*
			 * For all script states a "-->"  transitions
			 * back into the normal unescaped script mode,
			 * even if that's the current state.
			 */
			if (
				$at + 2 < $doc_length &&
				'-' === $html[ $at ] &&
				'-' === $html[ $at + 1 ] &&
				'>' === $html[ $at + 2 ]
			) {
				$at   += 3;
				$state = 'unescaped';
				continue;
			}

			if ( $at + 1 >= $doc_length ) {
				return false;
			}

			/*
			 * Everything of interest past here starts with "<".
			 * Check this character and advance position regardless.
			 */
			if ( '<' !== $html[ $at++ ] ) {
				continue;
			}

			/*
			 * Unlike with "-->", the "<!--" only transitions
			 * into the escaped mode if not already there.
			 *
			 * Inside the escaped modes it will be ignored; and
			 * should never break out of the double-escaped
			 * mode and back into the escaped mode.
			 *
			 * While this requires a mode change, it does not
			 * impact the parsing otherwise, so continue
			 * parsing after updating the state.
			 */
			if (
				$at + 2 < $doc_length &&
				'!' === $html[ $at ] &&
				'-' === $html[ $at + 1 ] &&
				'-' === $html[ $at + 2 ]
			) {
				$at   += 3;
				$state = 'unescaped' === $state ? 'escaped' : $state;
				continue;
			}

			if ( '/' === $html[ $at ] ) {
				$closer_potentially_starts_at = $at - 1;
				$is_closing                   = true;
				++$at;
			} else {
				$is_closing = false;
			}

			/*
			 * At this point the only remaining state-changes occur with the
			 * <script> and </script> tags; unless one of these appears next,
			 * proceed scanning to the next potential token in the text.
			 */
			if ( ! (
				$at + 6 < $doc_length &&
				( 's' === $html[ $at ] || 'S' === $html[ $at ] ) &&
				( 'c' === $html[ $at + 1 ] || 'C' === $html[ $at + 1 ] ) &&
				( 'r' === $html[ $at + 2 ] || 'R' === $html[ $at + 2 ] ) &&
				( 'i' === $html[ $at + 3 ] || 'I' === $html[ $at + 3 ] ) &&
				( 'p' === $html[ $at + 4 ] || 'P' === $html[ $at + 4 ] ) &&
				( 't' === $html[ $at + 5 ] || 'T' === $html[ $at + 5 ] )
			) ) {
				++$at;
				continue;
			}

			/*
			 * Ensure that the script tag terminates to avoid matching on
			 * substrings of a non-match. For example, the sequence
			 * "<script123" should not end a script region even though
			 * "<script" is found within the text.
			 */
			if ( $at + 6 >= $doc_length ) {
				continue;
			}
			$at += 6;
			$c   = $html[ $at ];
			if ( ' ' !== $c && "\t" !== $c && "\r" !== $c && "\n" !== $c && '/' !== $c && '>' !== $c ) {
				++$at;
				continue;
			}

			if ( 'escaped' === $state && ! $is_closing ) {
				$state = 'double-escaped';
				continue;
			}

			if ( 'double-escaped' === $state && $is_closing ) {
				$state = 'escaped';
				continue;
			}

			if ( $is_closing ) {
				$this->bytes_already_parsed = $closer_potentially_starts_at;
				$this->tag_name_starts_at   = $closer_potentially_starts_at;
				if ( $this->bytes_already_parsed >= $doc_length ) {
					return false;
				}

				while ( $this->parse_next_attribute() ) {
					continue;
				}

				if ( $this->bytes_already_parsed >= $doc_length ) {
					$this->parser_state = self::STATE_INCOMPLETE_INPUT;

					return false;
				}

				if ( '>' === $html[ $this->bytes_already_parsed ] ) {
					++$this->bytes_already_parsed;
					return true;
				}
			}

			++$at;
		}

		return false;
	}

	
	private function parse_next_tag(): bool {
		$this->after_tag();

		$html       = $this->html;
		$doc_length = strlen( $html );
		$was_at     = $this->bytes_already_parsed;
		$at         = $was_at;

		while ( $at < $doc_length ) {
			$at = strpos( $html, '<', $at );
			if ( false === $at ) {
				break;
			}

			if ( $at > $was_at ) {
				/*
				 * A "<" normally starts a new HTML tag or syntax token, but in cases where the
				 * following character can't produce a valid token, the "<" is instead treated
				 * as plaintext and the parser should skip over it. This avoids a problem when
				 * following earlier practices of typing emoji with text, e.g. "<3". This
				 * should be a heart, not a tag. It's supposed to be rendered, not hidden.
				 *
				 * At this point the parser checks if this is one of those cases and if it is
				 * will continue searching for the next "<" in search of a token boundary.
				 *
				 * @see https://html.spec.whatwg.org/#tag-open-state
				 */
				if ( 1 !== strspn( $html, '!/?abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ', $at + 1, 1 ) ) {
					++$at;
					continue;
				}

				$this->parser_state         = self::STATE_TEXT_NODE;
				$this->token_starts_at      = $was_at;
				$this->token_length         = $at - $was_at;
				$this->text_starts_at       = $was_at;
				$this->text_length          = $this->token_length;
				$this->bytes_already_parsed = $at;
				return true;
			}

			$this->token_starts_at = $at;

			if ( $at + 1 < $doc_length && '/' === $this->html[ $at + 1 ] ) {
				$this->is_closing_tag = true;
				++$at;
			} else {
				$this->is_closing_tag = false;
			}

			/*
			 * HTML tag names must start with [a-zA-Z] otherwise they are not tags.
			 * For example, "<3" is rendered as text, not a tag opener. If at least
			 * one letter follows the "<" then _it is_ a tag, but if the following
			 * character is anything else it _is not a tag_.
			 *
			 * It's not uncommon to find non-tags starting with `<` in an HTML
			 * document, so it's good for performance to make this pre-check before
			 * continuing to attempt to parse a tag name.
			 *
			 * Reference:
			 * * https://html.spec.whatwg.org/multipage/parsing.html#data-state
			 * * https://html.spec.whatwg.org/multipage/parsing.html#tag-open-state
			 */
			$tag_name_prefix_length = strspn( $html, 'abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ', $at + 1 );
			if ( $tag_name_prefix_length > 0 ) {
				++$at;
				$this->parser_state         = self::STATE_MATCHED_TAG;
				$this->tag_name_starts_at   = $at;
				$this->tag_name_length      = $tag_name_prefix_length + strcspn( $html, " \t\f\r\n/>", $at + $tag_name_prefix_length );
				$this->bytes_already_parsed = $at + $this->tag_name_length;
				return true;
			}

			/*
			 * Abort if no tag is found before the end of
			 * the document. There is nothing left to parse.
			 */
			if ( $at + 1 >= $doc_length ) {
				$this->parser_state = self::STATE_INCOMPLETE_INPUT;

				return false;
			}

			/*
			 * `<!` transitions to markup declaration open state
			 * https://html.spec.whatwg.org/multipage/parsing.html#markup-declaration-open-state
			 */
			if ( ! $this->is_closing_tag && '!' === $html[ $at + 1 ] ) {
				/*
				 * `<!--` transitions to a comment state – apply further comment rules.
				 * https://html.spec.whatwg.org/multipage/parsing.html#tag-open-state
				 */
				if ( 0 === substr_compare( $html, '--', $at + 2, 2 ) ) {
					$closer_at = $at + 4;
					// If it's not possible to close the comment then there is nothing more to scan.
					if ( $doc_length <= $closer_at ) {
						$this->parser_state = self::STATE_INCOMPLETE_INPUT;

						return false;
					}

					// Abruptly-closed empty comments are a sequence of dashes followed by `>`.
					$span_of_dashes = strspn( $html, '-', $closer_at );
					if ( '>' === $html[ $closer_at + $span_of_dashes ] ) {
						/*
						 * @todo When implementing `set_modifiable_text()` ensure that updates to this token
						 *       don't break the syntax for short comments, e.g. `<!--->`. Unlike other comment
						 *       and bogus comment syntax, these leave no clear insertion point for text and
						 *       they need to be modified specially in order to contain text. E.g. to store
						 *       `?` as the modifiable text, the `<!--->` needs to become `<!--?-->`, which
						 *       involves inserting an additional `-` into the token after the modifiable text.
						 */
						$this->parser_state = self::STATE_COMMENT;
						$this->comment_type = self::COMMENT_AS_ABRUPTLY_CLOSED_COMMENT;
						$this->token_length = $closer_at + $span_of_dashes + 1 - $this->token_starts_at;

						// Only provide modifiable text if the token is long enough to contain it.
						if ( $span_of_dashes >= 2 ) {
							$this->comment_type   = self::COMMENT_AS_HTML_COMMENT;
							$this->text_starts_at = $this->token_starts_at + 4;
							$this->text_length    = $span_of_dashes - 2;
						}

						$this->bytes_already_parsed = $closer_at + $span_of_dashes + 1;
						return true;
					}

					/*
					 * Comments may be closed by either a --> or an invalid --!>.
					 * The first occurrence closes the comment.
					 *
					 * See https://html.spec.whatwg.org/#parse-error-incorrectly-closed-comment
					 */
					--$closer_at; // Pre-increment inside condition below reduces risk of accidental infinite looping.
					while ( ++$closer_at < $doc_length ) {
						$closer_at = strpos( $html, '--', $closer_at );
						if ( false === $closer_at ) {
							$this->parser_state = self::STATE_INCOMPLETE_INPUT;

							return false;
						}

						if ( $closer_at + 2 < $doc_length && '>' === $html[ $closer_at + 2 ] ) {
							$this->parser_state         = self::STATE_COMMENT;
							$this->comment_type         = self::COMMENT_AS_HTML_COMMENT;
							$this->token_length         = $closer_at + 3 - $this->token_starts_at;
							$this->text_starts_at       = $this->token_starts_at + 4;
							$this->text_length          = $closer_at - $this->text_starts_at;
							$this->bytes_already_parsed = $closer_at + 3;
							return true;
						}

						if (
							$closer_at + 3 < $doc_length &&
							'!' === $html[ $closer_at + 2 ] &&
							'>' === $html[ $closer_at + 3 ]
						) {
							$this->parser_state         = self::STATE_COMMENT;
							$this->comment_type         = self::COMMENT_AS_HTML_COMMENT;
							$this->token_length         = $closer_at + 4 - $this->token_starts_at;
							$this->text_starts_at       = $this->token_starts_at + 4;
							$this->text_length          = $closer_at - $this->text_starts_at;
							$this->bytes_already_parsed = $closer_at + 4;
							return true;
						}
					}
				}

				/*
				 * `<!DOCTYPE` transitions to DOCTYPE state – skip to the nearest >
				 * These are ASCII-case-insensitive.
				 * https://html.spec.whatwg.org/multipage/parsing.html#tag-open-state
				 */
				if (
					$doc_length > $at + 8 &&
					( 'D' === $html[ $at + 2 ] || 'd' === $html[ $at + 2 ] ) &&
					( 'O' === $html[ $at + 3 ] || 'o' === $html[ $at + 3 ] ) &&
					( 'C' === $html[ $at + 4 ] || 'c' === $html[ $at + 4 ] ) &&
					( 'T' === $html[ $at + 5 ] || 't' === $html[ $at + 5 ] ) &&
					( 'Y' === $html[ $at + 6 ] || 'y' === $html[ $at + 6 ] ) &&
					( 'P' === $html[ $at + 7 ] || 'p' === $html[ $at + 7 ] ) &&
					( 'E' === $html[ $at + 8 ] || 'e' === $html[ $at + 8 ] )
				) {
					$closer_at = strpos( $html, '>', $at + 9 );
					if ( false === $closer_at ) {
						$this->parser_state = self::STATE_INCOMPLETE_INPUT;

						return false;
					}

					$this->parser_state         = self::STATE_DOCTYPE;
					$this->token_length         = $closer_at + 1 - $this->token_starts_at;
					$this->text_starts_at       = $this->token_starts_at + 9;
					$this->text_length          = $closer_at - $this->text_starts_at;
					$this->bytes_already_parsed = $closer_at + 1;
					return true;
				}

				if (
					'html' !== $this->parsing_namespace &&
					strlen( $html ) > $at + 8 &&
					'[' === $html[ $at + 2 ] &&
					'C' === $html[ $at + 3 ] &&
					'D' === $html[ $at + 4 ] &&
					'A' === $html[ $at + 5 ] &&
					'T' === $html[ $at + 6 ] &&
					'A' === $html[ $at + 7 ] &&
					'[' === $html[ $at + 8 ]
				) {
					$closer_at = strpos( $html, ']]>', $at + 9 );
					if ( false === $closer_at ) {
						$this->parser_state = self::STATE_INCOMPLETE_INPUT;

						return false;
					}

					$this->parser_state         = self::STATE_CDATA_NODE;
					$this->text_starts_at       = $at + 9;
					$this->text_length          = $closer_at - $this->text_starts_at;
					$this->token_length         = $closer_at + 3 - $this->token_starts_at;
					$this->bytes_already_parsed = $closer_at + 3;
					return true;
				}

				/*
				 * Anything else here is an incorrectly-opened comment and transitions
				 * to the bogus comment state - skip to the nearest >. If no closer is
				 * found then the HTML was truncated inside the markup declaration.
				 */
				$closer_at = strpos( $html, '>', $at + 1 );
				if ( false === $closer_at ) {
					$this->parser_state = self::STATE_INCOMPLETE_INPUT;

					return false;
				}

				$this->parser_state         = self::STATE_COMMENT;
				$this->comment_type         = self::COMMENT_AS_INVALID_HTML;
				$this->token_length         = $closer_at + 1 - $this->token_starts_at;
				$this->text_starts_at       = $this->token_starts_at + 2;
				$this->text_length          = $closer_at - $this->text_starts_at;
				$this->bytes_already_parsed = $closer_at + 1;

				/*
				 * Identify nodes that would be CDATA if HTML had CDATA sections.
				 *
				 * This section must occur after identifying the bogus comment end
				 * because in an HTML parser it will span to the nearest `>`, even
				 * if there's no `]]>` as would be required in an XML document. It
				 * is therefore not possible to parse a CDATA section containing
				 * a `>` in the HTML syntax.
				 *
				 * Inside foreign elements there is a discrepancy between browsers
				 * and the specification on this.
				 *
				 * @todo Track whether the Tag Processor is inside a foreign element
				 *       and require the proper closing `]]>` in those cases.
				 */
				if (
					$this->token_length >= 10 &&
					'[' === $html[ $this->token_starts_at + 2 ] &&
					'C' === $html[ $this->token_starts_at + 3 ] &&
					'D' === $html[ $this->token_starts_at + 4 ] &&
					'A' === $html[ $this->token_starts_at + 5 ] &&
					'T' === $html[ $this->token_starts_at + 6 ] &&
					'A' === $html[ $this->token_starts_at + 7 ] &&
					'[' === $html[ $this->token_starts_at + 8 ] &&
					']' === $html[ $closer_at - 1 ] &&
					']' === $html[ $closer_at - 2 ]
				) {
					$this->parser_state    = self::STATE_COMMENT;
					$this->comment_type    = self::COMMENT_AS_CDATA_LOOKALIKE;
					$this->text_starts_at += 7;
					$this->text_length    -= 9;
				}

				return true;
			}

			/*
			 * </> is a missing end tag name, which is ignored.
			 *
			 * This was also known as the "presumptuous empty tag"
			 * in early discussions as it was proposed to close
			 * the nearest previous opening tag.
			 *
			 * See https://html.spec.whatwg.org/#parse-error-missing-end-tag-name
			 */
			if ( '>' === $html[ $at + 1 ] ) {
				// `<>` is interpreted as plaintext.
				if ( ! $this->is_closing_tag ) {
					++$at;
					continue;
				}

				$this->parser_state         = self::STATE_PRESUMPTUOUS_TAG;
				$this->token_length         = $at + 2 - $this->token_starts_at;
				$this->bytes_already_parsed = $at + 2;
				return true;
			}

			/*
			 * `<?` transitions to a bogus comment state – skip to the nearest >
			 * See https://html.spec.whatwg.org/multipage/parsing.html#tag-open-state
			 */
			if ( ! $this->is_closing_tag && '?' === $html[ $at + 1 ] ) {
				$closer_at = strpos( $html, '>', $at + 2 );
				if ( false === $closer_at ) {
					$this->parser_state = self::STATE_INCOMPLETE_INPUT;

					return false;
				}

				$this->parser_state         = self::STATE_COMMENT;
				$this->comment_type         = self::COMMENT_AS_INVALID_HTML;
				$this->token_length         = $closer_at + 1 - $this->token_starts_at;
				$this->text_starts_at       = $this->token_starts_at + 2;
				$this->text_length          = $closer_at - $this->text_starts_at;
				$this->bytes_already_parsed = $closer_at + 1;

				/*
				 * Identify a Processing Instruction node were HTML to have them.
				 *
				 * This section must occur after identifying the bogus comment end
				 * because in an HTML parser it will span to the nearest `>`, even
				 * if there's no `?>` as would be required in an XML document. It
				 * is therefore not possible to parse a Processing Instruction node
				 * containing a `>` in the HTML syntax.
				 *
				 * XML allows for more target names, but this code only identifies
				 * those with ASCII-representable target names. This means that it
				 * may identify some Processing Instruction nodes as bogus comments,
				 * but it will not misinterpret the HTML structure. By limiting the
				 * identification to these target names the Tag Processor can avoid
				 * the need to start parsing UTF-8 sequences.
				 *
				 * > NameStartChar ::= ":" | [A-Z] | "_" | [a-z] | [#xC0-#xD6] | [#xD8-#xF6] | [#xF8-#x2FF] |
				 *                     [#x370-#x37D] | [#x37F-#x1FFF] | [#x200C-#x200D] | [#x2070-#x218F] |
				 *                     [#x2C00-#x2FEF] | [#x3001-#xD7FF] | [#xF900-#xFDCF] | [#xFDF0-#xFFFD] |
				 *                     [#x10000-#xEFFFF]
				 * > NameChar      ::= NameStartChar | "-" | "." | [0-9] | #xB7 | [#x0300-#x036F] | [#x203F-#x2040]
				 *
				 * @todo Processing instruction nodes in SGML may contain any kind of markup. XML defines a
				 *       special case with `<?xml ... ?>` syntax, but the `?` is part of the bogus comment.
				 *
				 * @see https://www.w3.org/TR/2006/REC-xml11-20060816/#NT-PITarget
				 */
				if ( $this->token_length >= 5 && '?' === $html[ $closer_at - 1 ] ) {
					$comment_text     = substr( $html, $this->token_starts_at + 2, $this->token_length - 4 );
					$pi_target_length = strspn( $comment_text, 'abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ:_' );

					if ( 0 < $pi_target_length ) {
						$pi_target_length += strspn( $comment_text, 'abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ0123456789:_-.', $pi_target_length );

						$this->comment_type       = self::COMMENT_AS_PI_NODE_LOOKALIKE;
						$this->tag_name_starts_at = $this->token_starts_at + 2;
						$this->tag_name_length    = $pi_target_length;
						$this->text_starts_at    += $pi_target_length;
						$this->text_length       -= $pi_target_length + 1;
					}
				}

				return true;
			}

			/*
			 * If a non-alpha starts the tag name in a tag closer it's a comment.
			 * Find the first `>`, which closes the comment.
			 *
			 * This parser classifies these particular comments as special "funky comments"
			 * which are made available for further processing.
			 *
			 * See https://html.spec.whatwg.org/#parse-error-invalid-first-character-of-tag-name
			 */
			if ( $this->is_closing_tag ) {
				// No chance of finding a closer.
				if ( $at + 3 > $doc_length ) {
					$this->parser_state = self::STATE_INCOMPLETE_INPUT;

					return false;
				}

				$closer_at = strpos( $html, '>', $at + 2 );
				if ( false === $closer_at ) {
					$this->parser_state = self::STATE_INCOMPLETE_INPUT;

					return false;
				}

				$this->parser_state         = self::STATE_FUNKY_COMMENT;
				$this->token_length         = $closer_at + 1 - $this->token_starts_at;
				$this->text_starts_at       = $this->token_starts_at + 2;
				$this->text_length          = $closer_at - $this->text_starts_at;
				$this->bytes_already_parsed = $closer_at + 1;
				return true;
			}

			++$at;
		}

		/*
		 * This does not imply an incomplete parse; it indicates that there
		 * can be nothing left in the document other than a #text node.
		 */
		$this->parser_state         = self::STATE_TEXT_NODE;
		$this->token_starts_at      = $was_at;
		$this->token_length         = $doc_length - $was_at;
		$this->text_starts_at       = $was_at;
		$this->text_length          = $this->token_length;
		$this->bytes_already_parsed = $doc_length;
		return true;
	}

	
	private function parse_next_attribute(): bool {
		$doc_length = strlen( $this->html );

		// Skip whitespace and slashes.
		$this->bytes_already_parsed += strspn( $this->html, " \t\f\r\n/", $this->bytes_already_parsed );
		if ( $this->bytes_already_parsed >= $doc_length ) {
			$this->parser_state = self::STATE_INCOMPLETE_INPUT;

			return false;
		}

		/*
		 * Treat the equal sign as a part of the attribute
		 * name if it is the first encountered byte.
		 *
		 * @see https://html.spec.whatwg.org/multipage/parsing.html#before-attribute-name-state
		 */
		$name_length = '=' === $this->html[ $this->bytes_already_parsed ]
			? 1 + strcspn( $this->html, "=/> \t\f\r\n", $this->bytes_already_parsed + 1 )
			: strcspn( $this->html, "=/> \t\f\r\n", $this->bytes_already_parsed );

		// No attribute, just tag closer.
		if ( 0 === $name_length || $this->bytes_already_parsed + $name_length >= $doc_length ) {
			return false;
		}

		$attribute_start             = $this->bytes_already_parsed;
		$attribute_name              = substr( $this->html, $attribute_start, $name_length );
		$this->bytes_already_parsed += $name_length;
		if ( $this->bytes_already_parsed >= $doc_length ) {
			$this->parser_state = self::STATE_INCOMPLETE_INPUT;

			return false;
		}

		$this->skip_whitespace();
		if ( $this->bytes_already_parsed >= $doc_length ) {
			$this->parser_state = self::STATE_INCOMPLETE_INPUT;

			return false;
		}

		$has_value = '=' === $this->html[ $this->bytes_already_parsed ];
		if ( $has_value ) {
			++$this->bytes_already_parsed;
			$this->skip_whitespace();
			if ( $this->bytes_already_parsed >= $doc_length ) {
				$this->parser_state = self::STATE_INCOMPLETE_INPUT;

				return false;
			}

			switch ( $this->html[ $this->bytes_already_parsed ] ) {
				case "'":
				case '"':
					$quote                      = $this->html[ $this->bytes_already_parsed ];
					$value_start                = $this->bytes_already_parsed + 1;
					$end_quote_at               = strpos( $this->html, $quote, $value_start );
					$end_quote_at               = false === $end_quote_at ? $doc_length : $end_quote_at;
					$value_length               = $end_quote_at - $value_start;
					$attribute_end              = $end_quote_at + 1;
					$this->bytes_already_parsed = $attribute_end;
					break;

				default:
					$value_start                = $this->bytes_already_parsed;
					$value_length               = strcspn( $this->html, "> \t\f\r\n", $value_start );
					$attribute_end              = $value_start + $value_length;
					$this->bytes_already_parsed = $attribute_end;
			}
		} else {
			$value_start   = $this->bytes_already_parsed;
			$value_length  = 0;
			$attribute_end = $attribute_start + $name_length;
		}

		if ( $attribute_end >= $doc_length ) {
			$this->parser_state = self::STATE_INCOMPLETE_INPUT;

			return false;
		}

		if ( $this->is_closing_tag ) {
			return true;
		}

		/*
		 * > There must never be two or more attributes on
		 * > the same start tag whose names are an ASCII
		 * > case-insensitive match for each other.
		 *     - HTML 5 spec
		 *
		 * @see https://html.spec.whatwg.org/multipage/syntax.html#attributes-2:ascii-case-insensitive
		 */
		$comparable_name = strtolower( $attribute_name );

		// If an attribute is listed many times, only use the first declaration and ignore the rest.
		if ( ! isset( $this->attributes[ $comparable_name ] ) ) {
			$this->attributes[ $comparable_name ] = new WP_HTML_Attribute_Token(
				$attribute_name,
				$value_start,
				$value_length,
				$attribute_start,
				$attribute_end - $attribute_start,
				! $has_value
			);

			return true;
		}

		/*
		 * Track the duplicate attributes so if we remove it, all disappear together.
		 *
		 * While `$this->duplicated_attributes` could always be stored as an `array()`,
		 * which would simplify the logic here, storing a `null` and only allocating
		 * an array when encountering duplicates avoids needless allocations in the
		 * normative case of parsing tags with no duplicate attributes.
		 */
		$duplicate_span = new WP_HTML_Span( $attribute_start, $attribute_end - $attribute_start );
		if ( null === $this->duplicate_attributes ) {
			$this->duplicate_attributes = array( $comparable_name => array( $duplicate_span ) );
		} elseif ( ! isset( $this->duplicate_attributes[ $comparable_name ] ) ) {
			$this->duplicate_attributes[ $comparable_name ] = array( $duplicate_span );
		} else {
			$this->duplicate_attributes[ $comparable_name ][] = $duplicate_span;
		}

		return true;
	}

	
	private function skip_whitespace(): void {
		$this->bytes_already_parsed += strspn( $this->html, " \t\f\r\n", $this->bytes_already_parsed );
	}

	
	private function after_tag(): void {
		/*
		 * There could be lexical updates enqueued for an attribute that
		 * also exists on the next tag. In order to avoid conflating the
		 * attributes across the two tags, lexical updates with names
		 * need to be flushed to raw lexical updates.
		 */
		$this->class_name_updates_to_attributes_updates();

		/*
		 * Purge updates if there are too many. The actual count isn't
		 * scientific, but a few values from 100 to a few thousand were
		 * tests to find a practically-useful limit.
		 *
		 * If the update queue grows too big, then the Tag Processor
		 * will spend more time iterating through them and lose the
		 * efficiency gains of deferring applying them.
		 */
		if ( 1000 < count( $this->lexical_updates ) ) {
			$this->get_updated_html();
		}

		foreach ( $this->lexical_updates as $name => $update ) {
			/*
			 * Any updates appearing after the cursor should be applied
			 * before proceeding, otherwise they may be overlooked.
			 */
			if ( $update->start >= $this->bytes_already_parsed ) {
				$this->get_updated_html();
				break;
			}

			if ( is_int( $name ) ) {
				continue;
			}

			$this->lexical_updates[] = $update;
			unset( $this->lexical_updates[ $name ] );
		}

		$this->token_starts_at          = null;
		$this->token_length             = null;
		$this->tag_name_starts_at       = null;
		$this->tag_name_length          = null;
		$this->text_starts_at           = 0;
		$this->text_length              = 0;
		$this->is_closing_tag           = null;
		$this->attributes               = array();
		$this->comment_type             = null;
		$this->text_node_classification = self::TEXT_IS_GENERIC;
		$this->duplicate_attributes     = null;
	}

	
	private function class_name_updates_to_attributes_updates(): void {
		if ( count( $this->classname_updates ) === 0 ) {
			return;
		}

		$existing_class = $this->get_enqueued_attribute_value( 'class' );
		if ( null === $existing_class || true === $existing_class ) {
			$existing_class = '';
		}

		if ( false === $existing_class && isset( $this->attributes['class'] ) ) {
			$existing_class = substr(
				$this->html,
				$this->attributes['class']->value_starts_at,
				$this->attributes['class']->value_length
			);
		}

		if ( false === $existing_class ) {
			$existing_class = '';
		}

		
		$class = '';

		
		$at = 0;

		
		$modified = false;

		$seen      = array();
		$to_remove = array();
		$is_quirks = self::QUIRKS_MODE === $this->compat_mode;
		if ( $is_quirks ) {
			foreach ( $this->classname_updates as $updated_name => $action ) {
				if ( self::REMOVE_CLASS === $action ) {
					$to_remove[] = strtolower( $updated_name );
				}
			}
		} else {
			foreach ( $this->classname_updates as $updated_name => $action ) {
				if ( self::REMOVE_CLASS === $action ) {
					$to_remove[] = $updated_name;
				}
			}
		}

		// Remove unwanted classes by only copying the new ones.
		$existing_class_length = strlen( $existing_class );
		while ( $at < $existing_class_length ) {
			// Skip to the first non-whitespace character.
			$ws_at     = $at;
			$ws_length = strspn( $existing_class, " \t\f\r\n", $ws_at );
			$at       += $ws_length;

			// Capture the class name – it's everything until the next whitespace.
			$name_length = strcspn( $existing_class, " \t\f\r\n", $at );
			if ( 0 === $name_length ) {
				// If no more class names are found then that's the end.
				break;
			}

			$name                  = substr( $existing_class, $at, $name_length );
			$comparable_class_name = $is_quirks ? strtolower( $name ) : $name;
			$at                   += $name_length;

			// If this class is marked for removal, remove it and move on to the next one.
			if ( in_array( $comparable_class_name, $to_remove, true ) ) {
				$modified = true;
				continue;
			}

			// If a class has already been seen then skip it; it should not be added twice.
			if ( in_array( $comparable_class_name, $seen, true ) ) {
				continue;
			}

			$seen[] = $comparable_class_name;

			/*
			 * Otherwise, append it to the new "class" attribute value.
			 *
			 * There are options for handling whitespace between tags.
			 * Preserving the existing whitespace produces fewer changes
			 * to the HTML content and should clarify the before/after
			 * content when debugging the modified output.
			 *
			 * This approach contrasts normalizing the inter-class
			 * whitespace to a single space, which might appear cleaner
			 * in the output HTML but produce a noisier change.
			 */
			if ( '' !== $class ) {
				$class .= substr( $existing_class, $ws_at, $ws_length );
			}
			$class .= $name;
		}

		// Add new classes by appending those which haven't already been seen.
		foreach ( $this->classname_updates as $name => $operation ) {
			$comparable_name = $is_quirks ? strtolower( $name ) : $name;
			if ( self::ADD_CLASS === $operation && ! in_array( $comparable_name, $seen, true ) ) {
				$modified = true;

				$class .= strlen( $class ) > 0 ? ' ' : '';
				$class .= $name;
			}
		}

		$this->classname_updates = array();
		if ( ! $modified ) {
			return;
		}

		if ( strlen( $class ) > 0 ) {
			$this->set_attribute( 'class', $class );
		} else {
			$this->remove_attribute( 'class' );
		}
	}

	
	private function apply_attributes_updates( int $shift_this_point ): int {
		if ( ! count( $this->lexical_updates ) ) {
			return 0;
		}

		$accumulated_shift_for_given_point = 0;

		/*
		 * Attribute updates can be enqueued in any order but updates
		 * to the document must occur in lexical order; that is, each
		 * replacement must be made before all others which follow it
		 * at later string indices in the input document.
		 *
		 * Sorting avoid making out-of-order replacements which
		 * can lead to mangled output, partially-duplicated
		 * attributes, and overwritten attributes.
		 */
		usort( $this->lexical_updates, array( self::class, 'sort_start_ascending' ) );

		$bytes_already_copied = 0;
		$output_buffer        = '';
		foreach ( $this->lexical_updates as $diff ) {
			$shift = strlen( $diff->text ) - $diff->length;

			// Adjust the cursor position by however much an update affects it.
			if ( $diff->start < $this->bytes_already_parsed ) {
				$this->bytes_already_parsed += $shift;
			}

			// Accumulate shift of the given pointer within this function call.
			if ( $diff->start < $shift_this_point ) {
				$accumulated_shift_for_given_point += $shift;
			}

			$output_buffer       .= substr( $this->html, $bytes_already_copied, $diff->start - $bytes_already_copied );
			$output_buffer       .= $diff->text;
			$bytes_already_copied = $diff->start + $diff->length;
		}

		$this->html = $output_buffer . substr( $this->html, $bytes_already_copied );

		/*
		 * Adjust bookmark locations to account for how the text
		 * replacements adjust offsets in the input document.
		 */
		foreach ( $this->bookmarks as $bookmark_name => $bookmark ) {
			$bookmark_end = $bookmark->start + $bookmark->length;

			/*
			 * Each lexical update which appears before the bookmark's endpoints
			 * might shift the offsets for those endpoints. Loop through each change
			 * and accumulate the total shift for each bookmark, then apply that
			 * shift after tallying the full delta.
			 */
			$head_delta = 0;
			$tail_delta = 0;

			foreach ( $this->lexical_updates as $diff ) {
				$diff_end = $diff->start + $diff->length;

				if ( $bookmark->start < $diff->start && $bookmark_end < $diff->start ) {
					break;
				}

				if ( $bookmark->start >= $diff->start && $bookmark_end < $diff_end ) {
					$this->release_bookmark( $bookmark_name );
					continue 2;
				}

				$delta = strlen( $diff->text ) - $diff->length;

				if ( $bookmark->start >= $diff->start ) {
					$head_delta += $delta;
				}

				if ( $bookmark_end >= $diff_end ) {
					$tail_delta += $delta;
				}
			}

			$bookmark->start  += $head_delta;
			$bookmark->length += $tail_delta - $head_delta;
		}

		$this->lexical_updates = array();

		return $accumulated_shift_for_given_point;
	}

	
	public function has_bookmark( $bookmark_name ): bool {
		return array_key_exists( $bookmark_name, $this->bookmarks );
	}

	
	public function seek( $bookmark_name ): bool {
		if ( ! array_key_exists( $bookmark_name, $this->bookmarks ) ) {
			_doing_it_wrong(
				__METHOD__,
				__( 'Unknown bookmark name.' ),
				'6.2.0'
			);
			return false;
		}

		$existing_bookmark = $this->bookmarks[ $bookmark_name ];

		if (
			$this->token_starts_at === $existing_bookmark->start &&
			$this->token_length === $existing_bookmark->length
		) {
			return true;
		}

		if ( ++$this->seek_count > static::MAX_SEEK_OPS ) {
			_doing_it_wrong(
				__METHOD__,
				__( 'Too many calls to seek() - this can lead to performance issues.' ),
				'6.2.0'
			);
			return false;
		}

		// Flush out any pending updates to the document.
		$this->get_updated_html();

		// Point this tag processor before the sought tag opener and consume it.
		$this->bytes_already_parsed = $this->bookmarks[ $bookmark_name ]->start;
		$this->parser_state         = self::STATE_READY;
		return $this->next_token();
	}

	
	private static function sort_start_ascending( WP_HTML_Text_Replacement $a, WP_HTML_Text_Replacement $b ): int {
		$by_start = $a->start - $b->start;
		if ( 0 !== $by_start ) {
			return $by_start;
		}

		$by_text = isset( $a->text, $b->text ) ? strcmp( $a->text, $b->text ) : 0;
		if ( 0 !== $by_text ) {
			return $by_text;
		}

		/*
		 * This code should be unreachable, because it implies the two replacements
		 * start at the same location and contain the same text.
		 */
		return $a->length - $b->length;
	}

	
	private function get_enqueued_attribute_value( string $comparable_name ) {
		if ( self::STATE_MATCHED_TAG !== $this->parser_state ) {
			return false;
		}

		if ( ! isset( $this->lexical_updates[ $comparable_name ] ) ) {
			return false;
		}

		$enqueued_text = $this->lexical_updates[ $comparable_name ]->text;

		// Removed attributes erase the entire span.
		if ( '' === $enqueued_text ) {
			return null;
		}

		/*
		 * Boolean attribute updates are just the attribute name without a corresponding value.
		 *
		 * This value might differ from the given comparable name in that there could be leading
		 * or trailing whitespace, and that the casing follows the name given in `set_attribute`.
		 *
		 * Example:
		 *
		 *     $p->set_attribute( 'data-TEST-id', 'update' );
		 *     'update' === $p->get_enqueued_attribute_value( 'data-test-id' );
		 *
		 * Detect this difference based on the absence of the `=`, which _must_ exist in any
		 * attribute containing a value, e.g. `<input type="text" enabled />`.
		 *                                            ¹           ²
		 *                                       1. Attribute with a string value.
		 *                                       2. Boolean attribute whose value is `true`.
		 */
		$equals_at = strpos( $enqueued_text, '=' );
		if ( false === $equals_at ) {
			return true;
		}

		/*
		 * Finally, a normal update's value will appear after the `=` and
		 * be double-quoted, as performed incidentally by `set_attribute`.
		 *
		 * e.g. `type="text"`
		 *           ¹²    ³
		 *        1. Equals is here.
		 *        2. Double-quoting starts one after the equals sign.
		 *        3. Double-quoting ends at the last character in the update.
		 */
		$enqueued_value = substr( $enqueued_text, $equals_at + 2, -1 );
		return WP_HTML_Decoder::decode_attribute( $enqueued_value );
	}

	
	public function get_attribute( $name ) {
		if ( self::STATE_MATCHED_TAG !== $this->parser_state ) {
			return null;
		}

		$comparable = strtolower( $name );

		/*
		 * For every attribute other than `class` it's possible to perform a quick check if
		 * there's an enqueued lexical update whose value takes priority over what's found in
		 * the input document.
		 *
		 * The `class` attribute is special though because of the exposed helpers `add_class`
		 * and `remove_class`. These form a builder for the `class` attribute, so an additional
		 * check for enqueued class changes is required in addition to the check for any enqueued
		 * attribute values. If any exist, those enqueued class changes must first be flushed out
		 * into an attribute value update.
		 */
		if ( 'class' === $name ) {
			$this->class_name_updates_to_attributes_updates();
		}

		// Return any enqueued attribute value updates if they exist.
		$enqueued_value = $this->get_enqueued_attribute_value( $comparable );
		if ( false !== $enqueued_value ) {
			return $enqueued_value;
		}

		if ( ! isset( $this->attributes[ $comparable ] ) ) {
			return null;
		}

		$attribute = $this->attributes[ $comparable ];

		/*
		 * This flag distinguishes an attribute with no value
		 * from an attribute with an empty string value. For
		 * unquoted attributes this could look very similar.
		 * It refers to whether an `=` follows the name.
		 *
		 * e.g. <div boolean-attribute empty-attribute=></div>
		 *           ¹                 ²
		 *        1. Attribute `boolean-attribute` is `true`.
		 *        2. Attribute `empty-attribute` is `""`.
		 */
		if ( true === $attribute->is_true ) {
			return true;
		}

		$raw_value = substr( $this->html, $attribute->value_starts_at, $attribute->value_length );

		return WP_HTML_Decoder::decode_attribute( $raw_value );
	}

	
	public function get_attribute_names_with_prefix( $prefix ): ?array {
		if (
			self::STATE_MATCHED_TAG !== $this->parser_state ||
			$this->is_closing_tag
		) {
			return null;
		}

		$comparable = strtolower( $prefix );

		$matches = array();
		foreach ( array_keys( $this->attributes ) as $attr_name ) {
			if ( str_starts_with( $attr_name, $comparable ) ) {
				$matches[] = $attr_name;
			}
		}
		return $matches;
	}

	
	public function get_namespace(): string {
		return $this->parsing_namespace;
	}

	
	public function get_tag(): ?string {
		if ( null === $this->tag_name_starts_at ) {
			return null;
		}

		$tag_name = substr( $this->html, $this->tag_name_starts_at, $this->tag_name_length );

		if ( self::STATE_MATCHED_TAG === $this->parser_state ) {
			return strtoupper( $tag_name );
		}

		if (
			self::STATE_COMMENT === $this->parser_state &&
			self::COMMENT_AS_PI_NODE_LOOKALIKE === $this->get_comment_type()
		) {
			return $tag_name;
		}

		return null;
	}

	
	public function get_qualified_tag_name(): ?string {
		$tag_name = $this->get_tag();
		if ( null === $tag_name ) {
			return null;
		}

		if ( 'html' === $this->get_namespace() ) {
			return $tag_name;
		}

		$lower_tag_name = strtolower( $tag_name );
		if ( 'math' === $this->get_namespace() ) {
			return $lower_tag_name;
		}

		if ( 'svg' === $this->get_namespace() ) {
			switch ( $lower_tag_name ) {
				case 'altglyph':
					return 'altGlyph';

				case 'altglyphdef':
					return 'altGlyphDef';

				case 'altglyphitem':
					return 'altGlyphItem';

				case 'animatecolor':
					return 'animateColor';

				case 'animatemotion':
					return 'animateMotion';

				case 'animatetransform':
					return 'animateTransform';

				case 'clippath':
					return 'clipPath';

				case 'feblend':
					return 'feBlend';

				case 'fecolormatrix':
					return 'feColorMatrix';

				case 'fecomponenttransfer':
					return 'feComponentTransfer';

				case 'fecomposite':
					return 'feComposite';

				case 'feconvolvematrix':
					return 'feConvolveMatrix';

				case 'fediffuselighting':
					return 'feDiffuseLighting';

				case 'fedisplacementmap':
					return 'feDisplacementMap';

				case 'fedistantlight':
					return 'feDistantLight';

				case 'fedropshadow':
					return 'feDropShadow';

				case 'feflood':
					return 'feFlood';

				case 'fefunca':
					return 'feFuncA';

				case 'fefuncb':
					return 'feFuncB';

				case 'fefuncg':
					return 'feFuncG';

				case 'fefuncr':
					return 'feFuncR';

				case 'fegaussianblur':
					return 'feGaussianBlur';

				case 'feimage':
					return 'feImage';

				case 'femerge':
					return 'feMerge';

				case 'femergenode':
					return 'feMergeNode';

				case 'femorphology':
					return 'feMorphology';

				case 'feoffset':
					return 'feOffset';

				case 'fepointlight':
					return 'fePointLight';

				case 'fespecularlighting':
					return 'feSpecularLighting';

				case 'fespotlight':
					return 'feSpotLight';

				case 'fetile':
					return 'feTile';

				case 'feturbulence':
					return 'feTurbulence';

				case 'foreignobject':
					return 'foreignObject';

				case 'glyphref':
					return 'glyphRef';

				case 'lineargradient':
					return 'linearGradient';

				case 'radialgradient':
					return 'radialGradient';

				case 'textpath':
					return 'textPath';

				default:
					return $lower_tag_name;
			}
		}

		// This unnecessary return prevents tools from inaccurately reporting type errors.
		return $tag_name;
	}

	
	public function get_qualified_attribute_name( $attribute_name ): ?string {
		if ( self::STATE_MATCHED_TAG !== $this->parser_state ) {
			return null;
		}

		$namespace  = $this->get_namespace();
		$lower_name = strtolower( $attribute_name );

		if ( 'math' === $namespace && 'definitionurl' === $lower_name ) {
			return 'definitionURL';
		}

		if ( 'svg' === $this->get_namespace() ) {
			switch ( $lower_name ) {
				case 'attributename':
					return 'attributeName';

				case 'attributetype':
					return 'attributeType';

				case 'basefrequency':
					return 'baseFrequency';

				case 'baseprofile':
					return 'baseProfile';

				case 'calcmode':
					return 'calcMode';

				case 'clippathunits':
					return 'clipPathUnits';

				case 'diffuseconstant':
					return 'diffuseConstant';

				case 'edgemode':
					return 'edgeMode';

				case 'filterunits':
					return 'filterUnits';

				case 'glyphref':
					return 'glyphRef';

				case 'gradienttransform':
					return 'gradientTransform';

				case 'gradientunits':
					return 'gradientUnits';

				case 'kernelmatrix':
					return 'kernelMatrix';

				case 'kernelunitlength':
					return 'kernelUnitLength';

				case 'keypoints':
					return 'keyPoints';

				case 'keysplines':
					return 'keySplines';

				case 'keytimes':
					return 'keyTimes';

				case 'lengthadjust':
					return 'lengthAdjust';

				case 'limitingconeangle':
					return 'limitingConeAngle';

				case 'markerheight':
					return 'markerHeight';

				case 'markerunits':
					return 'markerUnits';

				case 'markerwidth':
					return 'markerWidth';

				case 'maskcontentunits':
					return 'maskContentUnits';

				case 'maskunits':
					return 'maskUnits';

				case 'numoctaves':
					return 'numOctaves';

				case 'pathlength':
					return 'pathLength';

				case 'patterncontentunits':
					return 'patternContentUnits';

				case 'patterntransform':
					return 'patternTransform';

				case 'patternunits':
					return 'patternUnits';

				case 'pointsatx':
					return 'pointsAtX';

				case 'pointsaty':
					return 'pointsAtY';

				case 'pointsatz':
					return 'pointsAtZ';

				case 'preservealpha':
					return 'preserveAlpha';

				case 'preserveaspectratio':
					return 'preserveAspectRatio';

				case 'primitiveunits':
					return 'primitiveUnits';

				case 'refx':
					return 'refX';

				case 'refy':
					return 'refY';

				case 'repeatcount':
					return 'repeatCount';

				case 'repeatdur':
					return 'repeatDur';

				case 'requiredextensions':
					return 'requiredExtensions';

				case 'requiredfeatures':
					return 'requiredFeatures';

				case 'specularconstant':
					return 'specularConstant';

				case 'specularexponent':
					return 'specularExponent';

				case 'spreadmethod':
					return 'spreadMethod';

				case 'startoffset':
					return 'startOffset';

				case 'stddeviation':
					return 'stdDeviation';

				case 'stitchtiles':
					return 'stitchTiles';

				case 'surfacescale':
					return 'surfaceScale';

				case 'systemlanguage':
					return 'systemLanguage';

				case 'tablevalues':
					return 'tableValues';

				case 'targetx':
					return 'targetX';

				case 'targety':
					return 'targetY';

				case 'textlength':
					return 'textLength';

				case 'viewbox':
					return 'viewBox';

				case 'viewtarget':
					return 'viewTarget';

				case 'xchannelselector':
					return 'xChannelSelector';

				case 'ychannelselector':
					return 'yChannelSelector';

				case 'zoomandpan':
					return 'zoomAndPan';
			}
		}

		if ( 'html' !== $namespace ) {
			switch ( $lower_name ) {
				case 'xlink:actuate':
					return 'xlink actuate';

				case 'xlink:arcrole':
					return 'xlink arcrole';

				case 'xlink:href':
					return 'xlink href';

				case 'xlink:role':
					return 'xlink role';

				case 'xlink:show':
					return 'xlink show';

				case 'xlink:title':
					return 'xlink title';

				case 'xlink:type':
					return 'xlink type';

				case 'xml:lang':
					return 'xml lang';

				case 'xml:space':
					return 'xml space';

				case 'xmlns':
					return 'xmlns';

				case 'xmlns:xlink':
					return 'xmlns xlink';
			}
		}

		return $attribute_name;
	}

	
	public function has_self_closing_flag(): bool {
		if ( self::STATE_MATCHED_TAG !== $this->parser_state ) {
			return false;
		}

		/*
		 * The self-closing flag is the solidus at the _end_ of the tag, not the beginning.
		 *
		 * Example:
		 *
		 *     <figure />
		 *             ^ this appears one character before the end of the closing ">".
		 */
		return '/' === $this->html[ $this->token_starts_at + $this->token_length - 2 ];
	}

	
	public function is_tag_closer(): bool {
		return (
			self::STATE_MATCHED_TAG === $this->parser_state &&
			$this->is_closing_tag &&

			/*
			 * The BR tag can only exist as an opening tag. If something like `</br>`
			 * appears then the HTML parser will treat it as an opening tag with no
			 * attributes. The BR tag is unique in this way.
			 *
			 * @see https://html.spec.whatwg.org/#parsing-main-inbody
			 */
			'BR' !== $this->get_tag()
		);
	}

	
	public function get_token_type(): ?string {
		switch ( $this->parser_state ) {
			case self::STATE_MATCHED_TAG:
				return '#tag';

			case self::STATE_DOCTYPE:
				return '#doctype';

			default:
				return $this->get_token_name();
		}
	}

	
	public function get_token_name(): ?string {
		switch ( $this->parser_state ) {
			case self::STATE_MATCHED_TAG:
				return $this->get_tag();

			case self::STATE_TEXT_NODE:
				return '#text';

			case self::STATE_CDATA_NODE:
				return '#cdata-section';

			case self::STATE_COMMENT:
				return '#comment';

			case self::STATE_DOCTYPE:
				return 'html';

			case self::STATE_PRESUMPTUOUS_TAG:
				return '#presumptuous-tag';

			case self::STATE_FUNKY_COMMENT:
				return '#funky-comment';
		}

		return null;
	}

	
	public function get_comment_type(): ?string {
		if ( self::STATE_COMMENT !== $this->parser_state ) {
			return null;
		}

		return $this->comment_type;
	}

	
	public function get_full_comment_text(): ?string {
		if ( self::STATE_FUNKY_COMMENT === $this->parser_state ) {
			return $this->get_modifiable_text();
		}

		if ( self::STATE_COMMENT !== $this->parser_state ) {
			return null;
		}

		switch ( $this->get_comment_type() ) {
			case self::COMMENT_AS_HTML_COMMENT:
			case self::COMMENT_AS_ABRUPTLY_CLOSED_COMMENT:
				return $this->get_modifiable_text();

			case self::COMMENT_AS_CDATA_LOOKALIKE:
				return "[CDATA[{$this->get_modifiable_text()}]]";

			case self::COMMENT_AS_PI_NODE_LOOKALIKE:
				return "?{$this->get_tag()}{$this->get_modifiable_text()}?";

			/*
			 * This represents "bogus comments state" from HTML tokenization.
			 * This can be entered by `<?` or `<!`, where `?` is included in
			 * the comment text but `!` is not.
			 */
			case self::COMMENT_AS_INVALID_HTML:
				$preceding_character = $this->html[ $this->text_starts_at - 1 ];
				$comment_start       = '?' === $preceding_character ? '?' : '';
				return "{$comment_start}{$this->get_modifiable_text()}";
		}

		return null;
	}

	
	public function subdivide_text_appropriately(): bool {
		if ( self::STATE_TEXT_NODE !== $this->parser_state ) {
			return false;
		}

		$this->text_node_classification = self::TEXT_IS_GENERIC;

		/*
		 * NULL bytes are treated categorically different than numeric character
		 * references whose number is zero. `&#x00;` is not the same as `"\x00"`.
		 */
		$leading_nulls = strspn( $this->html, "\x00", $this->text_starts_at, $this->text_length );
		if ( $leading_nulls > 0 ) {
			$this->token_length             = $leading_nulls;
			$this->text_length              = $leading_nulls;
			$this->bytes_already_parsed     = $this->token_starts_at + $leading_nulls;
			$this->text_node_classification = self::TEXT_IS_NULL_SEQUENCE;
			return true;
		}

		/*
		 * Start a decoding loop to determine the point at which the
		 * text subdivides. This entails raw whitespace bytes and any
		 * character reference that decodes to the same.
		 */
		$at  = $this->text_starts_at;
		$end = $this->text_starts_at + $this->text_length;
		while ( $at < $end ) {
			$skipped = strspn( $this->html, " \t\f\r\n", $at, $end - $at );
			$at     += $skipped;

			if ( $at < $end && '&' === $this->html[ $at ] ) {
				$matched_byte_length = null;
				$replacement         = WP_HTML_Decoder::read_character_reference( 'data', $this->html, $at, $matched_byte_length );
				if ( isset( $replacement ) && 1 === strspn( $replacement, " \t\f\r\n" ) ) {
					$at += $matched_byte_length;
					continue;
				}
			}

			break;
		}

		if ( $at > $this->text_starts_at ) {
			$new_length                     = $at - $this->text_starts_at;
			$this->text_length              = $new_length;
			$this->token_length             = $new_length;
			$this->bytes_already_parsed     = $at;
			$this->text_node_classification = self::TEXT_IS_WHITESPACE;
			return true;
		}

		return false;
	}

	
	public function get_modifiable_text(): string {
		$has_enqueued_update = isset( $this->lexical_updates['modifiable text'] );

		if ( ! $has_enqueued_update && ( null === $this->text_starts_at || 0 === $this->text_length ) ) {
			return '';
		}

		$text = $has_enqueued_update
			? $this->lexical_updates['modifiable text']->text
			: substr( $this->html, $this->text_starts_at, $this->text_length );

		/*
		 * Pre-processing the input stream would normally happen before
		 * any parsing is done, but deferring it means it's possible to
		 * skip in most cases. When getting the modifiable text, however
		 * it's important to apply the pre-processing steps, which is
		 * normalizing newlines.
		 *
		 * @see https://html.spec.whatwg.org/#preprocessing-the-input-stream
		 * @see https://infra.spec.whatwg.org/#normalize-newlines
		 */
		$text = str_replace( "\r\n", "\n", $text );
		$text = str_replace( "\r", "\n", $text );

		// Comment data is not decoded.
		if (
			self::STATE_CDATA_NODE === $this->parser_state ||
			self::STATE_COMMENT === $this->parser_state ||
			self::STATE_DOCTYPE === $this->parser_state ||
			self::STATE_FUNKY_COMMENT === $this->parser_state
		) {
			return str_replace( "\x00", "\u{FFFD}", $text );
		}

		$tag_name = $this->get_token_name();
		if (
			// Script data is not decoded.
			'SCRIPT' === $tag_name ||

			// RAWTEXT data is not decoded.
			'IFRAME' === $tag_name ||
			'NOEMBED' === $tag_name ||
			'NOFRAMES' === $tag_name ||
			'STYLE' === $tag_name ||
			'XMP' === $tag_name
		) {
			return str_replace( "\x00", "\u{FFFD}", $text );
		}

		$decoded = WP_HTML_Decoder::decode_text_node( $text );

		/*
		 * Skip the first line feed after LISTING, PRE, and TEXTAREA opening tags.
		 *
		 * Note that this first newline may come in the form of a character
		 * reference, such as `&#x0a;`, and so it's important to perform
		 * this transformation only after decoding the raw text content.
		 */
		if (
			( "\n" === ( $decoded[0] ?? '' ) ) &&
			( ( $this->skip_newline_at === $this->token_starts_at && '#text' === $tag_name ) || 'TEXTAREA' === $tag_name )
		) {
			$decoded = substr( $decoded, 1 );
		}

		/*
		 * Only in normative text nodes does the NULL byte (U+0000) get removed.
		 * In all other contexts it's replaced by the replacement character (U+FFFD)
		 * for security reasons (to avoid joining together strings that were safe
		 * when separated, but not when joined).
		 *
		 * @todo Inside HTML integration points and MathML integration points, the
		 *       text is processed according to the insertion mode, not according
		 *       to the foreign content rules. This should strip the NULL bytes.
		 */
		return ( '#text' === $tag_name && 'html' === $this->get_namespace() )
			? str_replace( "\x00", '', $decoded )
			: str_replace( "\x00", "\u{FFFD}", $decoded );
	}

	
	public function set_modifiable_text( string $plaintext_content ): bool {
		if ( self::STATE_TEXT_NODE === $this->parser_state ) {
			$this->lexical_updates['modifiable text'] = new WP_HTML_Text_Replacement(
				$this->text_starts_at,
				$this->text_length,
				htmlspecialchars( $plaintext_content, ENT_QUOTES | ENT_HTML5 )
			);

			return true;
		}

		// Comment data is not encoded.
		if (
			self::STATE_COMMENT === $this->parser_state &&
			self::COMMENT_AS_HTML_COMMENT === $this->comment_type
		) {
			// Check if the text could close the comment.
			if ( 1 === preg_match( '/--!?>/', $plaintext_content ) ) {
				return false;
			}

			$this->lexical_updates['modifiable text'] = new WP_HTML_Text_Replacement(
				$this->text_starts_at,
				$this->text_length,
				$plaintext_content
			);

			return true;
		}

		if ( self::STATE_MATCHED_TAG !== $this->parser_state ) {
			return false;
		}

		switch ( $this->get_tag() ) {
			case 'SCRIPT':
				/*
				 * This is over-protective, but ensures the update doesn't break
				 * out of the SCRIPT element. A more thorough check would need to
				 * ensure that the script closing tag doesn't exist, and isn't
				 * also "hidden" inside the script double-escaped state.
				 *
				 * It may seem like replacing `</script` with `<\/script` would
				 * properly escape these things, but this could mask regex patterns
				 * that previously worked. Resolve this by not sending `</script`
				 */
				if ( false !== stripos( $plaintext_content, '</script' ) ) {
					return false;
				}

				$this->lexical_updates['modifiable text'] = new WP_HTML_Text_Replacement(
					$this->text_starts_at,
					$this->text_length,
					$plaintext_content
				);

				return true;

			case 'STYLE':
				$plaintext_content = preg_replace_callback(
					'~</(?P<TAG_NAME>style)~i',
					static function ( $tag_match ) {
						return "\\3c\\2f{$tag_match['TAG_NAME']}";
					},
					$plaintext_content
				);

				$this->lexical_updates['modifiable text'] = new WP_HTML_Text_Replacement(
					$this->text_starts_at,
					$this->text_length,
					$plaintext_content
				);

				return true;

			case 'TEXTAREA':
			case 'TITLE':
				$plaintext_content = preg_replace_callback(
					"~</(?P<TAG_NAME>{$this->get_tag()})~i",
					static function ( $tag_match ) {
						return "&lt;/{$tag_match['TAG_NAME']}";
					},
					$plaintext_content
				);

				/*
				 * These don't _need_ to be escaped, but since they are decoded it's
				 * safe to leave them escaped and this can prevent other code from
				 * naively detecting tags within the contents.
				 *
				 * @todo It would be useful to prefix a multiline replacement text
				 *       with a newline, but not necessary. This is for aesthetics.
				 */
				$this->lexical_updates['modifiable text'] = new WP_HTML_Text_Replacement(
					$this->text_starts_at,
					$this->text_length,
					$plaintext_content
				);

				return true;
		}

		return false;
	}

	
	public function set_attribute( $name, $value ): bool {
		if (
			self::STATE_MATCHED_TAG !== $this->parser_state ||
			$this->is_closing_tag
		) {
			return false;
		}

		/*
		 * WordPress rejects more characters than are strictly forbidden
		 * in HTML5. This is to prevent additional security risks deeper
		 * in the WordPress and plugin stack. Specifically the
		 * less-than (<) greater-than (>) and ampersand (&) aren't allowed.
		 *
		 * The use of a PCRE match enables looking for specific Unicode
		 * code points without writing a UTF-8 decoder. Whereas scanning
		 * for one-byte characters is trivial (with `strcspn`), scanning
		 * for the longer byte sequences would be more complicated. Given
		 * that this shouldn't be in the hot path for execution, it's a
		 * reasonable compromise in efficiency without introducing a
		 * noticeable impact on the overall system.
		 *
		 * @see https://html.spec.whatwg.org/#attributes-2
		 *
		 * @todo As the only regex pattern maybe we should take it out?
		 *       Are Unicode patterns available broadly in Core?
		 */
		if ( preg_match(
			'~[' .
				// Syntax-like characters.
				'"\'>&</ =' .
				// Control characters.
				'\x{00}-\x{1F}' .
				// HTML noncharacters.
				'\x{FDD0}-\x{FDEF}' .
				'\x{FFFE}\x{FFFF}\x{1FFFE}\x{1FFFF}\x{2FFFE}\x{2FFFF}\x{3FFFE}\x{3FFFF}' .
				'\x{4FFFE}\x{4FFFF}\x{5FFFE}\x{5FFFF}\x{6FFFE}\x{6FFFF}\x{7FFFE}\x{7FFFF}' .
				'\x{8FFFE}\x{8FFFF}\x{9FFFE}\x{9FFFF}\x{AFFFE}\x{AFFFF}\x{BFFFE}\x{BFFFF}' .
				'\x{CFFFE}\x{CFFFF}\x{DFFFE}\x{DFFFF}\x{EFFFE}\x{EFFFF}\x{FFFFE}\x{FFFFF}' .
				'\x{10FFFE}\x{10FFFF}' .
			']~Ssu',
			$name
		) ) {
			_doing_it_wrong(
				__METHOD__,
				__( 'Invalid attribute name.' ),
				'6.2.0'
			);

			return false;
		}

		/*
		 * > The values "true" and "false" are not allowed on boolean attributes.
		 * > To represent a false value, the attribute has to be omitted altogether.
		 *     - HTML5 spec, https://html.spec.whatwg.org/#boolean-attributes
		 */
		if ( false === $value ) {
			return $this->remove_attribute( $name );
		}

		if ( true === $value ) {
			$updated_attribute = $name;
		} else {
			$comparable_name = strtolower( $name );

			/*
			 * Escape URL attributes.
			 *
			 * @see https://html.spec.whatwg.org/#attributes-3
			 */
			$escaped_new_value = in_array( $comparable_name, wp_kses_uri_attributes(), true ) ? esc_url( $value ) : esc_attr( $value );

			// If the escaping functions wiped out the update, reject it and indicate it was rejected.
			if ( '' === $escaped_new_value && '' !== $value ) {
				return false;
			}

			$updated_attribute = "{$name}=\"{$escaped_new_value}\"";
		}

		/*
		 * > There must never be two or more attributes on
		 * > the same start tag whose names are an ASCII
		 * > case-insensitive match for each other.
		 *     - HTML 5 spec
		 *
		 * @see https://html.spec.whatwg.org/multipage/syntax.html#attributes-2:ascii-case-insensitive
		 */
		$comparable_name = strtolower( $name );

		if ( isset( $this->attributes[ $comparable_name ] ) ) {
			/*
			 * Update an existing attribute.
			 *
			 * Example – set attribute id to "new" in <div id="initial_id" />:
			 *
			 *     <div id="initial_id"/>
			 *          ^-------------^
			 *          start         end
			 *     replacement: `id="new"`
			 *
			 *     Result: <div id="new"/>
			 */
			$existing_attribute                        = $this->attributes[ $comparable_name ];
			$this->lexical_updates[ $comparable_name ] = new WP_HTML_Text_Replacement(
				$existing_attribute->start,
				$existing_attribute->length,
				$updated_attribute
			);
		} else {
			/*
			 * Create a new attribute at the tag's name end.
			 *
			 * Example – add attribute id="new" to <div />:
			 *
			 *     <div/>
			 *         ^
			 *         start and end
			 *     replacement: ` id="new"`
			 *
			 *     Result: <div id="new"/>
			 */
			$this->lexical_updates[ $comparable_name ] = new WP_HTML_Text_Replacement(
				$this->tag_name_starts_at + $this->tag_name_length,
				0,
				' ' . $updated_attribute
			);
		}

		/*
		 * Any calls to update the `class` attribute directly should wipe out any
		 * enqueued class changes from `add_class` and `remove_class`.
		 */
		if ( 'class' === $comparable_name && ! empty( $this->classname_updates ) ) {
			$this->classname_updates = array();
		}

		return true;
	}

	
	public function remove_attribute( $name ): bool {
		if (
			self::STATE_MATCHED_TAG !== $this->parser_state ||
			$this->is_closing_tag
		) {
			return false;
		}

		/*
		 * > There must never be two or more attributes on
		 * > the same start tag whose names are an ASCII
		 * > case-insensitive match for each other.
		 *     - HTML 5 spec
		 *
		 * @see https://html.spec.whatwg.org/multipage/syntax.html#attributes-2:ascii-case-insensitive
		 */
		$name = strtolower( $name );

		/*
		 * Any calls to update the `class` attribute directly should wipe out any
		 * enqueued class changes from `add_class` and `remove_class`.
		 */
		if ( 'class' === $name && count( $this->classname_updates ) !== 0 ) {
			$this->classname_updates = array();
		}

		/*
		 * If updating an attribute that didn't exist in the input
		 * document, then remove the enqueued update and move on.
		 *
		 * For example, this might occur when calling `remove_attribute()`
		 * after calling `set_attribute()` for the same attribute
		 * and when that attribute wasn't originally present.
		 */
		if ( ! isset( $this->attributes[ $name ] ) ) {
			if ( isset( $this->lexical_updates[ $name ] ) ) {
				unset( $this->lexical_updates[ $name ] );
			}
			return false;
		}

		/*
		 * Removes an existing tag attribute.
		 *
		 * Example – remove the attribute id from <div id="main"/>:
		 *    <div id="initial_id"/>
		 *         ^-------------^
		 *         start         end
		 *    replacement: ``
		 *
		 *    Result: <div />
		 */
		$this->lexical_updates[ $name ] = new WP_HTML_Text_Replacement(
			$this->attributes[ $name ]->start,
			$this->attributes[ $name ]->length,
			''
		);

		// Removes any duplicated attributes if they were also present.
		foreach ( $this->duplicate_attributes[ $name ] ?? array() as $attribute_token ) {
			$this->lexical_updates[] = new WP_HTML_Text_Replacement(
				$attribute_token->start,
				$attribute_token->length,
				''
			);
		}

		return true;
	}

	
	public function add_class( $class_name ): bool {
		if (
			self::STATE_MATCHED_TAG !== $this->parser_state ||
			$this->is_closing_tag
		) {
			return false;
		}

		if ( self::QUIRKS_MODE !== $this->compat_mode ) {
			$this->classname_updates[ $class_name ] = self::ADD_CLASS;
			return true;
		}

		/*
		 * Because class names are matched ASCII-case-insensitively in quirks mode,
		 * this needs to see if a case variant of the given class name is already
		 * enqueued and update that existing entry, if so. This picks the casing of
		 * the first-provided class name for all lexical variations.
		 */
		$class_name_length = strlen( $class_name );
		foreach ( $this->classname_updates as $updated_name => $action ) {
			if (
				strlen( $updated_name ) === $class_name_length &&
				0 === substr_compare( $updated_name, $class_name, 0, $class_name_length, true )
			) {
				$this->classname_updates[ $updated_name ] = self::ADD_CLASS;
				return true;
			}
		}

		$this->classname_updates[ $class_name ] = self::ADD_CLASS;
		return true;
	}

	
	public function remove_class( $class_name ): bool {
		if (
			self::STATE_MATCHED_TAG !== $this->parser_state ||
			$this->is_closing_tag
		) {
			return false;
		}

		if ( self::QUIRKS_MODE !== $this->compat_mode ) {
			$this->classname_updates[ $class_name ] = self::REMOVE_CLASS;
			return true;
		}

		/*
		 * Because class names are matched ASCII-case-insensitively in quirks mode,
		 * this needs to see if a case variant of the given class name is already
		 * enqueued and update that existing entry, if so. This picks the casing of
		 * the first-provided class name for all lexical variations.
		 */
		$class_name_length = strlen( $class_name );
		foreach ( $this->classname_updates as $updated_name => $action ) {
			if (
				strlen( $updated_name ) === $class_name_length &&
				0 === substr_compare( $updated_name, $class_name, 0, $class_name_length, true )
			) {
				$this->classname_updates[ $updated_name ] = self::REMOVE_CLASS;
				return true;
			}
		}

		$this->classname_updates[ $class_name ] = self::REMOVE_CLASS;
		return true;
	}

	
	public function __toString(): string {
		return $this->get_updated_html();
	}

	
	public function get_updated_html(): string {
		$requires_no_updating = 0 === count( $this->classname_updates ) && 0 === count( $this->lexical_updates );

		/*
		 * When there is nothing more to update and nothing has already been
		 * updated, return the original document and avoid a string copy.
		 */
		if ( $requires_no_updating ) {
			return $this->html;
		}

		/*
		 * Keep track of the position right before the current tag. This will
		 * be necessary for reparsing the current tag after updating the HTML.
		 */
		$before_current_tag = $this->token_starts_at ?? 0;

		/*
		 * 1. Apply the enqueued edits and update all the pointers to reflect those changes.
		 */
		$this->class_name_updates_to_attributes_updates();
		$before_current_tag += $this->apply_attributes_updates( $before_current_tag );

		/*
		 * 2. Rewind to before the current tag and reparse to get updated attributes.
		 *
		 * At this point the internal cursor points to the end of the tag name.
		 * Rewind before the tag name starts so that it's as if the cursor didn't
		 * move; a call to `next_tag()` will reparse the recently-updated attributes
		 * and additional calls to modify the attributes will apply at this same
		 * location, but in order to avoid issues with subclasses that might add
		 * behaviors to `next_tag()`, the internal methods should be called here
		 * instead.
		 *
		 * It's important to note that in this specific place there will be no change
		 * because the processor was already at a tag when this was called and it's
		 * rewinding only to the beginning of this very tag before reprocessing it
		 * and its attributes.
		 *
		 * <p>Previous HTML<em>More HTML</em></p>
		 *                 ↑  │ back up by the length of the tag name plus the opening <
		 *                 └←─┘ back up by strlen("em") + 1 ==> 3
		 */
		$this->bytes_already_parsed = $before_current_tag;
		$this->base_class_next_token();

		return $this->html;
	}

	
	private function parse_query( $query ) {
		if ( null !== $query && $query === $this->last_query ) {
			return;
		}

		$this->last_query          = $query;
		$this->sought_tag_name     = null;
		$this->sought_class_name   = null;
		$this->sought_match_offset = 1;
		$this->stop_on_tag_closers = false;

		// A single string value means "find the tag of this name".
		if ( is_string( $query ) ) {
			$this->sought_tag_name = $query;
			return;
		}

		// An empty query parameter applies no restrictions on the search.
		if ( null === $query ) {
			return;
		}

		// If not using the string interface, an associative array is required.
		if ( ! is_array( $query ) ) {
			_doing_it_wrong(
				__METHOD__,
				__( 'The query argument must be an array or a tag name.' ),
				'6.2.0'
			);
			return;
		}

		if ( isset( $query['tag_name'] ) && is_string( $query['tag_name'] ) ) {
			$this->sought_tag_name = $query['tag_name'];
		}

		if ( isset( $query['class_name'] ) && is_string( $query['class_name'] ) ) {
			$this->sought_class_name = $query['class_name'];
		}

		if ( isset( $query['match_offset'] ) && is_int( $query['match_offset'] ) && 0 < $query['match_offset'] ) {
			$this->sought_match_offset = $query['match_offset'];
		}

		if ( isset( $query['tag_closers'] ) ) {
			$this->stop_on_tag_closers = 'visit' === $query['tag_closers'];
		}
	}


	
	private function matches(): bool {
		if ( $this->is_closing_tag && ! $this->stop_on_tag_closers ) {
			return false;
		}

		// Does the tag name match the requested tag name in a case-insensitive manner?
		if (
			isset( $this->sought_tag_name ) &&
			(
				strlen( $this->sought_tag_name ) !== $this->tag_name_length ||
				0 !== substr_compare( $this->html, $this->sought_tag_name, $this->tag_name_starts_at, $this->tag_name_length, true )
			)
		) {
			return false;
		}

		if ( null !== $this->sought_class_name && ! $this->has_class( $this->sought_class_name ) ) {
			return false;
		}

		return true;
	}

	
	public function get_doctype_info(): ?WP_HTML_Doctype_Info {
		if ( self::STATE_DOCTYPE !== $this->parser_state ) {
			return null;
		}

		return WP_HTML_Doctype_Info::from_doctype_token( substr( $this->html, $this->token_starts_at, $this->token_length ) );
	}

	
	const STATE_READY = 'STATE_READY';

	
	const STATE_COMPLETE = 'STATE_COMPLETE';

	
	const STATE_INCOMPLETE_INPUT = 'STATE_INCOMPLETE_INPUT';

	
	const STATE_MATCHED_TAG = 'STATE_MATCHED_TAG';

	
	const STATE_TEXT_NODE = 'STATE_TEXT_NODE';

	
	const STATE_CDATA_NODE = 'STATE_CDATA_NODE';

	
	const STATE_COMMENT = 'STATE_COMMENT';

	
	const STATE_DOCTYPE = 'STATE_DOCTYPE';

	
	const STATE_PRESUMPTUOUS_TAG = 'STATE_PRESUMPTUOUS_TAG';

	
	const STATE_FUNKY_COMMENT = 'STATE_WP_FUNKY';

	
	const COMMENT_AS_ABRUPTLY_CLOSED_COMMENT = 'COMMENT_AS_ABRUPTLY_CLOSED_COMMENT';

	
	const COMMENT_AS_CDATA_LOOKALIKE = 'COMMENT_AS_CDATA_LOOKALIKE';

	
	const COMMENT_AS_HTML_COMMENT = 'COMMENT_AS_HTML_COMMENT';

	
	const COMMENT_AS_PI_NODE_LOOKALIKE = 'COMMENT_AS_PI_NODE_LOOKALIKE';

	
	const COMMENT_AS_INVALID_HTML = 'COMMENT_AS_INVALID_HTML';

	
	const NO_QUIRKS_MODE = 'no-quirks-mode';

	
	const QUIRKS_MODE = 'quirks-mode';

	
	const TEXT_IS_GENERIC = 'TEXT_IS_GENERIC';

	
	const TEXT_IS_NULL_SEQUENCE = 'TEXT_IS_NULL_SEQUENCE';

	
	const TEXT_IS_WHITESPACE = 'TEXT_IS_WHITESPACE';
}