* Class for efficiently looking up and mapping string keys to string values, with limits.
* Use this class in specific circumstances with a static set of lookup keys which map to
* a static set of transformed values. For example, this class is used to map HTML named
* character references to their equivalent UTF-8 values.
* This class works differently than code calling `in_array()` and other methods. It
* internalizes lookup logic and provides helper interfaces to optimize lookup and
* transformation. It provides a method for precomputing the lookup tables and storing
* them as PHP source code.
* All tokens and substitutions must be shorter than 256 bytes.
* $smilies = WP_Token_Map::from_array( array(
* true === $smilies->contains( ':)' );
* false === $smilies->contains( 'simile' );
* '😕' === $smilies->read_token( 'Not sure :?.', 9, $length_of_smily_syntax );
* 2 === $length_of_smily_syntax;
* ## Precomputing the Token Map.
* Creating the class involves some work sorting and organizing the tokens and their
* replacement values. In order to skip this, it's possible for the class to export
* its state and be used as actual PHP source code.
* // Export with four spaces as the indent, only for the sake of this docblock.
* // The default indent is a tab character.
* echo $smilies->precomputed_php_source_table( $indent );
* // Output, to be pasted into a PHP source file:
* WP_Token_Map::from_precomputed_table(
* "storage_version" => "6.6.0",
* "long_words" => array(),
* "small_words" => "8O\x00:)\x00:(\x00:?\x00",
* "small_mappings" => array( "😯", "🙂", "🙁", "😕" )
* ## Large vs. small words.
* This class uses a short prefix called the "key" to optimize lookup of its tokens.
* This means that some tokens may be shorter than or equal in length to that key.
* Those words that are longer than the key are called "large" while those shorter
* than or equal to the key length are called "small."
* This separation of large and small words is incidental to the way this class
* optimizes lookup, and should be considered an internal implementation detail
* of the class. It may still be important to be aware of it, however.
* ## Determining Key Length.
* The choice of the size of the key length should be based on the data being stored in
* the token map. It should divide the data as evenly as possible, but should not create
* so many groups that a large fraction of the groups only contain a single token.
* For the HTML5 named character references, a key length of 2 was found to provide a
* sufficient spread and should be a good default for relatively large sets of tokens.
* However, for some data sets this might be too long. For example, a list of smilies
* may be too small for a key length of 2. Perhaps 1 would be more appropriate. It's
* best to experiment and determine empirically which values are appropriate.
* ## Generate Pre-Computed Source Code.
* Since the `WP_Token_Map` is designed for relatively static lookups, it can be
* advantageous to precompute the values and instantiate a table that has already
* sorted and grouped the tokens and built the lookup strings.
* This can be done with `WP_Token_Map::precomputed_php_source_table()`.
* Note that if there is a leading character that all tokens need, such as `&` for
* HTML named character references, it can be beneficial to exclude this from the
* token map. Instead, find occurrences of the leading character and then use the
* token map to see if the following characters complete the token.
* $map = WP_Token_Map::from_array( array( 'simple_smile:' => '🙂', 'sob:' => '😭', 'soba:' => '🍜' ) );
* echo $map->precomputed_php_source_table();
* WP_Token_Map::from_precomputed_table(
* "storage_version" => "6.6.0",
* "groups" => "si\x00so\x00",
* "\x0bmple_smile:\x04🙂",
* "\x03ba:\x04🍜\x02b:\x04😭",
* "short_mappings" => array()
* This precomputed value can be stored directly in source code and will skip the
* startup cost of generating the lookup strings. See `$html5_named_character_entities`.
* Note that any updates to the precomputed format should update the storage version
* constant. It would also be best to provide an update function to take older known
* versions and upgrade them in place when loading into `from_precomputed_table()`.
* It may be viable to dynamically increase the length limits such that there's no need to impose them.
* The limit appears because of the packing structure, which indicates how many bytes each segment of
* text in the lookup tables spans. If, however, care were taken to track the longest word length, then
* the packing structure could change its representation to allow for that. Each additional byte storing
* length, however, increases the memory overhead and lookup runtime.
* An alternative approach could be to borrow the UTF-8 variable-length encoding and store lengths of less
* than 127 as a single byte with the high bit unset, storing longer lengths as the combination of
* Since it has not been shown during the development of this class that longer strings are required, this
* update is deferred until such a need is clear.
* Denotes the version of the code which produces pre-computed source tables.
* This version will be used not only to verify pre-computed data, but also
* to upgrade pre-computed data from older versions. Choosing a name that
* corresponds to the WordPress release will help people identify where an
* old copy of data came from.
const STORAGE_VERSION = '6.6.0-trunk';
* Maximum length for each key and each transformed value in the table (in bytes).
* How many bytes of each key are used to form a group key for lookup.
* This also determines whether a word is considered short or long.
* Stores an optimized form of the word set, where words are grouped
* by a prefix of the `$key_length` and then collapsed into a string.
* In each group, the keys and lookups form a packed data structure.
* The keys in the string are stripped of their "group key," which is
* the prefix of length `$this->key_length` shared by all of the items
* in the group. Each word in the string is prefixed by a single byte
* whose raw unsigned integer value represents how many bytes follow.
* ┌────────────────┬───────────────┬─────────────────┬────────┐
* │ Length of rest │ Rest of key │ Length of value │ Value │
* │ of key (bytes) │ │ (bytes) │ │
* ├────────────────┼───────────────┼─────────────────┼────────┤
* │ 0x08 │ nterDot; │ 0x02 │ · │
* └────────────────┴───────────────┴─────────────────┴────────┘
* In this example, the key `CenterDot;` has a group key `Ce`, leaving
* eight bytes for the rest of the key, `nterDot;`, and two bytes for
* the transformed value `·` (or U+B7 or "\xC2\xB7").
* // Stores array( 'CenterDot;' => '·', 'Cedilla;' => '¸' ).
* $large_words = array( "\x08nterDot;\x02·\x06dilla;\x02¸" )
* The prefixes appear in the `$groups` string, each followed by a null
* byte. This makes for quick lookup of where in the group string the key
* is found, and then a simple division converts that offset into the index
* in the `$large_words` array where the group string is to be found.
* This lookup data structure is designed to optimize cache locality and
* minimize indirect memory reads when matching strings in the set.
private $large_words = array();
* Stores the group keys for sequential string lookup.
* The offset into this string where the group key appears corresponds with the index
* into the group array where the rest of the group string appears. This is an optimization
* to improve cache locality while searching and minimize indirect memory accesses.
* Stores an optimized row of small words, where every entry is
* `$this->key_size + 1` bytes long and zero-extended.
* This packing allows for direct lookup of a short word followed
* by the null byte, if extended to `$this->key_size + 1`.
* // Stores array( 'GT', 'LT', 'gt', 'lt' ).
* "GT\x00LT\x00gt\x00lt\x00"
private $small_words = '';
* Replacements for the small words, in the same order they appear.
* With the position of a small word it's possible to index the translation
* directly, as its position in the `$small_words` string corresponds to
* the index of the replacement in the `$small_mapping` array.
* array( '>', '<', '>', '<' )
private $small_mappings = array();
* Create a token map using an associative array of key/value pairs as the input.
* $smilies = WP_Token_Map::from_array( array(
* @param array $mappings The keys transform into the values, both are strings.
* @param int $key_length Determines the group key length. Leave at the default value
* of 2 unless there's an empirical reason to change it.
* @return WP_Token_Map|null Token map, unless unable to create it.
public static function from_array( $mappings, $key_length = 2 ) {
$map = new WP_Token_Map();
$map->key_length = $key_length;
// Start by grouping words.
foreach ( $mappings as $word => $mapping ) {
self::MAX_LENGTH <= strlen( $word ) ||
self::MAX_LENGTH <= strlen( $mapping )
/* translators: 1: maximum byte length (a count) */
__( 'Token Map tokens and substitutions must all be shorter than %1$d bytes.' ),
$length = strlen( $word );
if ( $key_length >= $length ) {
$group = substr( $word, 0, $key_length );
if ( ! isset( $groups[ $group ] ) ) {
$groups[ $group ] = array();
$groups[ $group ][] = array( substr( $word, $key_length ), $mapping );
* Sort the words to ensure that no smaller substring of a match masks the full match.
* For example, `Cap` should not match before `CapitalDifferentialD`.
usort( $shorts, 'WP_Token_Map::longest_first_then_alphabetical' );
foreach ( $groups as $group_key => $group ) {
static function ( $a, $b ) {
return self::longest_first_then_alphabetical( $a[0], $b[0] );
// Finally construct the optimized lookups.
foreach ( $shorts as $word ) {
$map->small_words .= str_pad( $word, $key_length + 1, "\x00", STR_PAD_RIGHT );
$map->small_mappings[] = $mappings[ $word ];
$group_keys = array_keys( $groups );
foreach ( $group_keys as $group ) {
$map->groups .= "{$group}\x00";
foreach ( $groups[ $group ] as $group_word ) {
list( $word, $mapping ) = $group_word;
$word_length = pack( 'C', strlen( $word ) );
$mapping_length = pack( 'C', strlen( $mapping ) );
$group_string .= "{$word_length}{$word}{$mapping_length}{$mapping}";
$map->large_words[] = $group_string;
* Creates a token map from a pre-computed table.
* This skips the initialization cost of generating the table.
* This function should only be used to load data created with
* WP_Token_Map::precomputed_php_source_tag().
* Stores pre-computed state for directly loading into a Token Map.
* @type string $storage_version Which version of the code produced this state.
* @type int $key_length Group key length.
* @type string $groups Group lookup index.
* @type array $large_words Large word groups and packed strings.
* @type string $small_words Small words packed string.
* @type array $small_mappings Small word mappings.
* @return WP_Token_Map Map with precomputed data loaded.
public static function from_precomputed_table( $state ) {
$has_necessary_state = isset(
$state['storage_version'],
if ( ! $has_necessary_state ) {
__( 'Missing required inputs to pre-computed WP_Token_Map.' ),
if ( self::STORAGE_VERSION !== $state['storage_version'] ) {
/* translators: 1: version string, 2: version string. */
sprintf( __( 'Loaded version \'%1$s\' incompatible with expected version \'%2$s\'.' ), $state['storage_version'], self::STORAGE_VERSION ),
$map = new WP_Token_Map();
$map->key_length = $state['key_length'];
$map->groups = $state['groups'];
$map->large_words = $state['large_words'];
$map->small_words = $state['small_words'];
$map->small_mappings = $state['small_mappings'];
* Indicates if a given word is a lookup key in the map.
* true === $smilies->contains( ':)' );
* false === $smilies->contains( 'simile' );
* @param string $word Determine if this word is a lookup key in the map.
* @param string $case_sensitivity Optional. Pass 'ascii-case-insensitive' to ignore ASCII case when matching. Default 'case-sensitive'.
* @return bool Whether there's an entry for the given word in the map.
public function contains( $word, $case_sensitivity = 'case-sensitive' ) {
$ignore_case = 'ascii-case-insensitive' === $case_sensitivity;
if ( $this->key_length >= strlen( $word ) ) {
if ( 0 === strlen( $this->small_words ) ) {
$term = str_pad( $word, $this->key_length + 1, "\x00", STR_PAD_RIGHT );
$word_at = $ignore_case ? stripos( $this->small_words, $term ) : strpos( $this->small_words, $term );
if ( false === $word_at ) {
$group_key = substr( $word, 0, $this->key_length );
$group_at = $ignore_case ? stripos( $this->groups, $group_key ) : strpos( $this->groups, $group_key );
if ( false === $group_at ) {
$group = $this->large_words[ $group_at / ( $this->key_length + 1 ) ];
$group_length = strlen( $group );
$slug = substr( $word, $this->key_length );
$length = strlen( $slug );
while ( $at < $group_length ) {
$token_length = unpack( 'C', $group[ $at++ ] )[1];
$mapping_length = unpack( 'C', $group[ $at++ ] )[1];
if ( $token_length === $length && 0 === substr_compare( $group, $slug, $token_at, $token_length, $ignore_case ) ) {
$at = $mapping_at + $mapping_length;
* If the text starting at a given offset is a lookup key in the map,
* return the corresponding transformation from the map, else `false`.
* This function returns the translated string, but accepts an optional
* parameter `$matched_token_byte_length`, which communicates how many
* bytes long the lookup key was, if it found one. This can be used to
* advance a cursor in calling code if a lookup key was found.
* false === $smilies->read_token( 'Not sure :?.', 0, $token_byte_length );
* '😕' === $smilies->read_token( 'Not sure :?.', 9, $token_byte_length );
* 2 === $token_byte_length;
It is recommended that you Edit text format, this type of Fix handles quite a lot in one request
