xregexp.js

/*!
 * XRegExp 3.0.0-pre
 * <http://xregexp.com/>
 * Steven Levithan © 2007-2012 MIT License
 */

/**
 * XRegExp provides augmented, extensible regular expressions. You get new syntax, flags, and
 * methods beyond what browsers support natively. XRegExp is also a regex utility belt with tools
 * to make your client-side grepping simpler and more powerful, while freeing you from worrying
 * about pesky cross-browser inconsistencies and the dubious `lastIndex` property.
 */
var XRegExp = (function(undefined) {
    'use strict';

/* ==============================
 * Private variables
 * ============================== */

    var // ...

// Property name used for extended regex instance data
    REGEX_DATA = 'xregexp',

// Internal reference to the `XRegExp` object
    self,

// Optional features that can be installed and uninstalled
    features = {
        astral: false,
        natives: false
    },

// Store native methods to use and restore ('native' is an ES3 reserved keyword)
    nativ = {
        exec: RegExp.prototype.exec,
        test: RegExp.prototype.test,
        match: String.prototype.match,
        replace: String.prototype.replace,
        split: String.prototype.split
    },

// Storage for fixed/extended native methods
    fixed = {},

// Storage for regexes cached by `XRegExp.cache`
    cache = {},

// Storage for pattern details cached by the `XRegExp` constructor
    patternCache = {},

// Storage for regex syntax tokens added internally or by `XRegExp.addToken`
    tokens = [],

// Token scopes
    defaultScope = 'default',
    classScope = 'class',

// Regexes that match native regex syntax, including octals
    nativeTokens = {
        // Any native multicharacter token in default scope, or any single character
        'default': /\\(?:0(?:[0-3][0-7]{0,2}|[4-7][0-7]?)?|[1-9]\d*|x[\dA-Fa-f]{2}|u[\dA-Fa-f]{4}|c[A-Za-z]|[\s\S])|\(\?[:=!]|[?*+]\?|{\d+(?:,\d*)?}\??|[\s\S]/,
        // Any native multicharacter token in character class scope, or any single character
        'class': /\\(?:[0-3][0-7]{0,2}|[4-7][0-7]?|x[\dA-Fa-f]{2}|u[\dA-Fa-f]{4}|c[A-Za-z]|[\s\S])|[\s\S]/
    },

// Any backreference or dollar-prefixed character in replacement strings
    replacementToken = /\$(?:{([\w$]+)}|(\d\d?|[\s\S]))/g,

// Check for correct `exec` handling of nonparticipating capturing groups
    correctExecNpcg = nativ.exec.call(/()??/, '')[1] === undefined,

// Check for flag y support
    hasNativeY = RegExp.prototype.sticky !== undefined,

// Used to kill infinite recursion during XRegExp construction
    isInsideConstructor = false,

// Tracker for known flags, including addon flags
    registeredFlags = {
        g: true,
        i: true,
        m: true,
        y: hasNativeY
    },

// Shortcut to `XRegExp.addToken`
    add;

/* ==============================
 * Private functions
 * ============================== */

/**
 * Attaches named capture data and `XRegExp.prototype` properties to a regex object.
 * @private
 * @param {RegExp} regex Regex to augment.
 * @param {Array} captureNames Array with capture names, or `null`.
 * @param {Boolean} [addProto=false] Whether to attach `XRegExp.prototype` properties. Not
 *   attaching properties avoids a minor performance penalty.
 * @returns {RegExp} Augmented regex.
 */
    function augment(regex, captureNames, addProto) {
        var p;
        if (addProto) {
            // Can't auto-inherit these since the XRegExp constructor returns a nonprimitive value
            if (regex.__proto__) {
                regex.__proto__ = self.prototype;
            } else {
                for (p in self.prototype) {
                    // A `self.prototype.hasOwnProperty(p)` check wouldn't be worth it here, since
                    // this is performance sensitive, and enumerable `Object.prototype` or
                    // `RegExp.prototype` extensions exist on `regex.prototype` anyway
                    regex[p] = self.prototype[p];
                }
            }
        }
        regex[REGEX_DATA] = {captureNames: captureNames};
        return regex;
    }

/**
 * Removes any duplicate characters from the provided string.
 * @private
 * @param {String} str String to remove duplicate characters from.
 * @returns {String} String with any duplicate characters removed.
 */
    function clipDuplicates(str) {
        return nativ.replace.call(str, /([\s\S])(?=[\s\S]*\1)/g, '');
    }

/**
 * Copies a regex object while preserving special properties for named capture and augmenting with
 * `XRegExp.prototype` methods. The copy has a fresh `lastIndex` property (set to zero). Allows
 * adding and removing native flags while copying the regex.
 * @private
 * @param {RegExp} regex Regex to copy.
 * @param {Object} [options] Allows specifying native flags to add or remove while copying the
 *   regex, and whether to attach `XRegExp.prototype` properties.
 * @returns {RegExp} Copy of the provided regex, possibly with modified flags.
 */
    function copy(regex, options) {
        if (!self.isRegExp(regex)) {
            throw new TypeError('Type RegExp expected');
        }

        // Get native flags in use
        var flags = nativ.exec.call(/\/([a-z]*)$/i, String(regex))[1];
        options = options || {};

        if (options.add) {
            flags = clipDuplicates(flags + options.add);
        }
        if (options.remove) {
            // Would need to escape `options.remove` if this was public
            flags = nativ.replace.call(flags, new RegExp('[' + options.remove + ']+', 'g'), '');
        }

        // Augment with `XRegExp.prototype` methods, but use the native `RegExp` constructor and
        // avoid searching for special tokens. That would be wrong for regexes constructed by
        // `RegExp`, and unnecessary for regexes constructed by `XRegExp` because the regex has
        // already undergone the translation to native regex syntax
        regex = augment(
            new RegExp(regex.source, flags),
            hasNamedCapture(regex) ? regex[REGEX_DATA].captureNames.slice(0) : null,
            options.addProto
        );

        return regex;
    }

/**
 * Returns a new copy of the object used to hold extended regex instance data, tailored for a
 * native nonaugmented regex.
 * @private
 * @returns {Object} Object with base regex instance data.
 */
    function getBaseProps() {
        return {captureNames: null};
    }

/**
 * Determines whether a regex has extended instance data used to track capture names.
 * @private
 * @param {RegExp} regex Regex to check.
 * @returns {Boolean} Whether the regex uses named capture.
 */
    function hasNamedCapture(regex) {
        return !!(regex[REGEX_DATA] && regex[REGEX_DATA].captureNames);
    }

/**
 * Returns the first index at which a given value can be found in an array.
 * @private
 * @param {Array} array Array to search.
 * @param {*} value Value to locate in the array.
 * @returns {Number} Zero-based index at which the item is found, or -1.
 */
    function indexOf(array, value) {
        // Use the native array method, if available
        if (Array.prototype.indexOf) {
            return array.indexOf(value);
        }
        var len = array.length, i;
        // Not a very good shim, but good enough for XRegExp's use of it
        for (i = 0; i < len; ++i) {
            if (array[i] === value) {
                return i;
            }
        }
        return -1;
    }

/**
 * Determines whether a value is of the specified type, by resolving its internal [[Class]].
 * @private
 * @param {*} value Object to check.
 * @param {String} type Type to check for, in TitleCase.
 * @returns {Boolean} Whether the object matches the type.
 */
    function isType(value, type) {
        return Object.prototype.toString.call(value) === '[object ' + type + ']';
    }

/**
 * Checks whether the next nonignorable token after the specified position is a quantifier.
 * @private
 * @param {String} pattern Pattern to search within.
 * @param {Number} pos Index in `pattern` to search at.
 * @param {String} flags Flags used by the pattern.
 * @returns {Boolean} Whether the next token is a quantifier.
 */
    function isQuantifierNext(pattern, pos, flags) {
        return nativ.test.call(
            flags.indexOf('x') > -1 ?
                // Ignore any leading whitespace, line-comments, and inline-comments
                /^(?:\s+|#.*|\(\?#[^)]*\))*(?:[?*+]|{\d+(?:,\d*)?})/ :
                // Ignore any leading inline-comments
                /^(?:\(\?#[^)]*\))*(?:[?*+]|{\d+(?:,\d*)?})/,
            pattern.slice(pos)
        );
    }

/**
 * Prepares an options object from the given value.
 * @private
 * @param {String|Object} value Value to convert to an options object.
 * @returns {Object} Options object.
 */
    function prepareOptions(value) {
        value = value || {};
        if (isType(value, 'String')) {
            value = self.forEach(value, /[^\s,]+/, function(match) {
                this[match] = true;
            }, {});
        }
        return value;
    }

/**
 * Registers a flag so it doesn't throw an 'unknown flag' error.
 * @private
 * @param {String} flag Single-character flag to register.
 */
    function registerFlag(flag) {
        if (!/^[\w$]$/.test(flag)) {
            throw new Error('Flag must be a single character A-Za-z0-9_$');
        }
        registeredFlags[flag] = true;
    }

/**
 * Runs built-in and custom regex syntax tokens in reverse insertion order at the specified
 * position, until a match is found.
 * @private
 * @param {String} pattern Original pattern from which an XRegExp object is being built.
 * @param {String} flags Flags being used to construct the regex.
 * @param {Number} pos Position to search for tokens within `pattern`.
 * @param {Number} scope Regex scope to apply: 'default' or 'class'.
 * @param {Object} context Context object to use for token handler functions.
 * @returns {Object} Object with properties `matchLength`, `output`, and `reparse`; or `null`.
 */
    function runTokens(pattern, flags, pos, scope, context) {
        var i = tokens.length,
            result = null,
            match,
            t;
        // Protect against constructing XRegExp objects within token handler functions
        isInsideConstructor = true;
        try {
            // Run in reverse insertion order
            while (i--) {
                t = tokens[i];
                if (
                    (t.scope === scope || t.scope === 'all') &&
                    (!t.flag || flags.indexOf(t.flag) > -1)
                ) {
                    match = self.exec(pattern, t.regex, pos, 'sticky');
                    if (match) {
                        result = {
                            matchLength: match[0].length,
                            output: t.handler.call(context, match, scope, flags),
                            reparse: t.reparse
                        };
                        // Finished with token tests
                        break;
                    }
                }
            }
        } catch (err) {
            throw err;
        } finally {
            isInsideConstructor = false;
        }
        return result;
    }

/**
 * Enables or disables implicit astral mode opt-in.
 * @private
 * @param {Boolean} on `true` to enable; `false` to disable.
 */
    function setAstral(on) {
        features.astral = on;
        if (on) {
            // Reset the pattern cache used by the `XRegExp` constructor, since the same pattern
            // and flags might now produce different results
            self.cache.flush('patterns');
        }
    }

/**
 * Enables or disables native method overrides.
 * @private
 * @param {Boolean} on `true` to enable; `false` to disable.
 */
    function setNatives(on) {
        RegExp.prototype.exec = (on ? fixed : nativ).exec;
        RegExp.prototype.test = (on ? fixed : nativ).test;
        String.prototype.match = (on ? fixed : nativ).match;
        String.prototype.replace = (on ? fixed : nativ).replace;
        String.prototype.split = (on ? fixed : nativ).split;
        features.natives = on;
    }

/**
 * Returns the object, or throws an error if it is `null` or `undefined`. This is used to follow
 * the ES5 abstract operation `ToObject`.
 * @private
 * @param {*} value Object to check and return.
 * @returns {*} The provided object.
 */
    function toObject(value) {
        // This matches both `null` and `undefined`
        if (value == null) {
            throw new TypeError('Cannot convert null or undefined to object');
        }
        return value;
    }

/* ==============================
 * Constructor
 * ============================== */

/**
 * Creates an extended regular expression object for matching text with a pattern. Differs from a
 * native regular expression in that additional syntax and flags are supported. The returned object
 * is in fact a native `RegExp` and works with all native methods.
 * @class XRegExp
 * @constructor
 * @param {String|RegExp} pattern Regex pattern string, or an existing regex object to copy.
 * @param {String} [flags] Any combination of flags.
 *   Native flags:
 *     <li>`g` - global
 *     <li>`i` - ignore case
 *     <li>`m` - multiline anchors
 *     <li>`y` - sticky (Firefox 3+)
 *   Additional XRegExp flags:
 *     <li>`n` - explicit capture
 *     <li>`s` - dot matches all (aka singleline)
 *     <li>`x` - free-spacing and line comments (aka extended)
 *     <li>`A` - astral (requires the Unicode Base addon)
 *   Flags cannot be provided when constructing one `RegExp` from another.
 * @returns {RegExp} Extended regular expression object.
 * @example
 *
 * // With named capture and flag x
 * XRegExp('(?<year>  [0-9]{4} ) -?  # year  \n\
 *          (?<month> [0-9]{2} ) -?  # month \n\
 *          (?<day>   [0-9]{2} )     # day   ', 'x');
 *
 * // Providing a regex object copies it. Native regexes are recompiled using native (not XRegExp)
 * // syntax. Copies maintain special properties for named capture, are augmented with
 * // `XRegExp.prototype` methods, and have fresh `lastIndex` properties (set to zero).
 * XRegExp(/regex/);
 */
    self = function(pattern, flags) {
        var ERR_COPY_WITH_FLAGS = 'Cannot supply flags when copying a RegExp',
            ERR_CONSTRUTOR_RECURSION = 'Cannot create XRegExp objects within token handler functions',
            ERR_DUPLICATE_FLAG = 'Invalid duplicate regex flag ',
            ERR_BAD_INLINE_FLAG = 'Cannot use flag g or y in mode modifier ',
            ERR_UNKNOWN_FLAG = 'Unknown regex flag ',
            output = '',
            scope = defaultScope,
            context = {
                hasNamedCapture: false,
                captureNames: []
            },
            pos = 0,
            tokenResult,
            match,
            key,
            i;

        if (self.isRegExp(pattern)) {
            if (flags !== undefined) {
                throw new TypeError(ERR_COPY_WITH_FLAGS);
            }
            return copy(pattern, {addProto: true});
        }

        // Tokens become part of the regex construction process, so protect against infinite
        // recursion when an XRegExp is constructed within a token handler function
        if (isInsideConstructor) {
            throw new Error(ERR_CONSTRUTOR_RECURSION);
        }

        // Copy the argument behavior of `RegExp`
        pattern = pattern === undefined ? '' : String(pattern);
        flags = flags === undefined ? '' : String(flags);

        // Cache-lookup key
        key = pattern + '/' + flags;

        // If this pattern and flag combination hasn't been used since the last pattern cache reset
        if (!patternCache[key]) {
            // Recent browsers throw on duplicate flags, so copy this behavior for nonnative flags
            if (clipDuplicates(flags) !== flags) {
                throw new SyntaxError(ERR_DUPLICATE_FLAG + flags);
            }

            // Strip and apply a leading mode modifier with any combination of flags except g or y
            pattern = nativ.replace.call(pattern, /^\(\?([\w$]+)\)/, function($0, $1) {
                if (nativ.test.call(/[gy]/, $1)) {
                    throw new SyntaxError(ERR_BAD_INLINE_FLAG + $0);
                }
                // Allow duplicate flags within the mode modifier
                flags = clipDuplicates(flags + $1);
                return '';
            });

            // Throw on unknown native or nonnative flags
            for (i = 0; i < flags.length; ++i) {
                if (!registeredFlags[flags.charAt(i)]) {
                    throw new SyntaxError(ERR_UNKNOWN_FLAG + flags.charAt(i));
                }
            }

            // Use XRegExp's syntax tokens to translate the pattern to a native regex pattern...
            // `pattern.length` may change on each iteration, if tokens use the `reparse` option
            while (pos < pattern.length) {
                do {
                    // Check for custom tokens at the current position
                    tokenResult = runTokens(pattern, flags, pos, scope, context);
                    // If the matched token used the `reparse` option, splice its output into the
                    // pattern before running tokens again at the same position
                    if (tokenResult && tokenResult.reparse) {
                        pattern = pattern.slice(0, pos) +
                            tokenResult.output +
                            pattern.slice(pos + tokenResult.matchLength);
                    }
                } while (tokenResult && tokenResult.reparse);

                if (tokenResult) {
                    output += tokenResult.output;
                    pos += (tokenResult.matchLength || 1);
                } else {
                    // Get the native token at the current position. This could use the native
                    // `exec`, except that sticky processing avoids string slicing in browsers that
                    // support flag y
                    match = self.exec(pattern, nativeTokens[scope], pos, 'sticky')[0];
                    output += match;
                    pos += match.length;
                    if (match === '[' && scope === defaultScope) {
                        scope = classScope;
                    } else if (match === ']' && scope === classScope) {
                        scope = defaultScope;
                    }
                }
            }

            patternCache[key] = {
                // Cleanup token cruft: repeated `(?:)(?:)` and leading/trailing `(?:)`
                pattern: nativ.replace.call(output, /\(\?:\)(?=\(\?:\))|^\(\?:\)|\(\?:\)$/g, ''),
                // Strip all but native flags
                flags: nativ.replace.call(flags, /[^gimy]+/g, ''),
                // `context.captureNames` has an item for each capturing group, even if unnamed
                captures: context.hasNamedCapture ? context.captureNames : null
            }
        }

        key = patternCache[key];
        return augment(new RegExp(key.pattern, key.flags), key.captures, /*addProto*/ true);
    };

// Add `RegExp.prototype` to the prototype chain for XRegExp instances that have their prototype
// changed to `XRegExp.prototype` via `__proto__`
    self.prototype = new RegExp;

/* ==============================
 * Public properties
 * ============================== */

/**
 * The XRegExp version number.
 * @static
 * @memberOf XRegExp
 * @type String
 */
    self.version = '3.0.0-pre';

/* ==============================
 * Public methods
 * ============================== */

/**
 * Extends XRegExp syntax and allows custom flags. This is used internally and can be used to
 * create XRegExp addons. If more than one token can match the same string, the last added wins.
 * @memberOf XRegExp
 * @param {RegExp} regex Regex object that matches the new token.
 * @param {Function} handler Function that returns a new pattern string (using native regex syntax)
 *   to replace the matched token within all future XRegExp regexes. Has access to persistent
 *   properties of the regex being built, through `this`. Invoked with three arguments:
 *   <li>The match array, with named backreference properties.
 *   <li>The regex scope where the match was found: 'default' or 'class'.
 *   <li>The flags used by the regex, including any flags in a leading mode modifier.
 * @param {Object} [options] Options object with optional properties:
 *   <li>`scope` {String} Scope where the token applies: 'default', 'class', or 'all'.
 *   <li>`flag` {String} Single-character flag that triggers the token. This also registers the
 *     flag, which prevents XRegExp from throwing an 'unknown flag' error when the flag is used.
 *   <li>`optionalFlags` {String} Any custom flags checked for within the token `handler` that are
 *     not required to trigger the token. This registers the flags, to prevent XRegExp from
 *     throwing an 'unknown flag' error when any of the flags are used.
 *   <li>`reparse` {Boolean} Whether the `handler` function's output should not be treated as
 *     final, and instead be reparseable by other tokens (including the current token). Allows
 *     token chaining or deferring.
 * @example
 *
 * // Basic usage: Add \a for the ALERT control code
 * XRegExp.addToken(
 *   /\\a/,
 *   function() {return '\\x07';},
 *   {scope: 'all'}
 * );
 * XRegExp('\\a[\\a-\\n]+').test('\x07\n\x07'); // -> true
 *
 * // Add the U (ungreedy) flag from PCRE and RE2, which reverses greedy and lazy quantifiers
 * XRegExp.addToken(
 *   /([?*+]|{\d+(?:,\d*)?})(\??)/,
 *   function(match) {return match[1] + (match[2] ? '' : '?');},
 *   {flag: 'U'}
 * );
 * XRegExp('a+', 'U').exec('aaa')[0]; // -> 'a'
 * XRegExp('a+?', 'U').exec('aaa')[0]; // -> 'aaa'
 */
    self.addToken = function(regex, handler, options) {
        options = options || {};
        var optionalFlags = options.optionalFlags, i;

        if (options.flag) {
            registerFlag(options.flag);
        }

        if (optionalFlags) {
            optionalFlags = nativ.split.call(optionalFlags, '');
            for (i = 0; i < optionalFlags.length; ++i) {
                registerFlag(optionalFlags[i]);
            }
        }

        // Add to the private list of syntax tokens
        tokens.push({
            regex: copy(regex, {add: 'g' + (hasNativeY ? 'y' : '')}),
            handler: handler,
            scope: options.scope || defaultScope,
            flag: options.flag,
            reparse: options.reparse
        });

        // Reset the pattern cache used by the `XRegExp` constructor, since the same pattern and
        // flags might now produce different results
        self.cache.flush('patterns');
    };

/**
 * Caches and returns the result of calling `XRegExp(pattern, flags)`. On any subsequent call with
 * the same pattern and flag combination, the cached copy of the regex is returned.
 * @memberOf XRegExp
 * @param {String} pattern Regex pattern string.
 * @param {String} [flags] Any combination of XRegExp flags.
 * @returns {RegExp} Cached XRegExp object.
 * @example
 *
 * while (match = XRegExp.cache('.', 'gs').exec(str)) {
 *   // The regex is compiled once only
 * }
 */
    self.cache = function(pattern, flags) {
        var key = pattern + '/' + (flags || '');
        return cache[key] || (cache[key] = self(pattern, flags));
    };

// Intentionally undocumented
    self.cache.flush = function(cacheName) {
        if (cacheName === 'patterns') {
            // Flush the pattern cache used by the `XRegExp` constructor
            patternCache = {};
        } else {
            // Flush the regex object cache populated by `XRegExp.cache`
            cache = {};
        }
    };

/**
 * Escapes any regular expression metacharacters, for use when matching literal strings. The result
 * can safely be used at any point within a regex that uses any flags.
 * @memberOf XRegExp
 * @param {String} str String to escape.
 * @returns {String} String with regex metacharacters escaped.
 * @example
 *
 * XRegExp.escape('Escaped? <.>');
 * // -> 'Escaped\?\ <\.>'
 */
    self.escape = function(str) {
        return nativ.replace.call(toObject(str), /[-[\]{}()*+?.,\\^$|#\s]/g, '\\$&');
    };

/**
 * Executes a regex search in a specified string. Returns a match array or `null`. If the provided
 * regex uses named capture, named backreference properties are included on the match array.
 * Optional `pos` and `sticky` arguments specify the search start position, and whether the match
 * must start at the specified position only. The `lastIndex` property of the provided regex is not
 * used, but is updated for compatibility. Also fixes browser bugs compared to the native
 * `RegExp.prototype.exec` and can be used reliably cross-browser.
 * @memberOf XRegExp
 * @param {String} str String to search.
 * @param {RegExp} regex Regex to search with.
 * @param {Number} [pos=0] Zero-based index at which to start the search.
 * @param {Boolean|String} [sticky=false] Whether the match must start at the specified position
 *   only. The string `'sticky'` is accepted as an alternative to `true`.
 * @returns {Array} Match array with named backreference properties, or `null`.
 * @example
 *
 * // Basic use, with named backreference
 * var match = XRegExp.exec('U+2620', XRegExp('U\\+(?<hex>[0-9A-F]{4})'));
 * match.hex; // -> '2620'
 *
 * // With pos and sticky, in a loop
 * var pos = 2, result = [], match;
 * while (match = XRegExp.exec('<1><2><3><4>5<6>', /<(\d)>/, pos, 'sticky')) {
 *   result.push(match[1]);
 *   pos = match.index + match[0].length;
 * }
 * // result -> ['2', '3', '4']
 */
    self.exec = function(str, regex, pos, sticky) {
        var cacheFlags = 'g', match, r2;

        if (hasNativeY && (sticky || (regex.sticky && sticky !== false))) {
            cacheFlags += 'y';
        }

        regex[REGEX_DATA] = regex[REGEX_DATA] || getBaseProps();

        // Shares cached copies with `XRegExp.match`/`replace`
        r2 = regex[REGEX_DATA][cacheFlags] || (
            regex[REGEX_DATA][cacheFlags] = copy(regex, {
                add: cacheFlags,
                remove: sticky === false ? 'y' : ''
            })
        );

        r2.lastIndex = pos = pos || 0;

        // Fixed `exec` required for `lastIndex` fix, named backreferences, etc.
        match = fixed.exec.call(r2, str);

        if (sticky && match && match.index !== pos) {
            match = null;
        }

        if (regex.global) {
            regex.lastIndex = match ? r2.lastIndex : 0;
        }

        return match;
    };

/**
 * Executes a provided function once per regex match.
 * @memberOf XRegExp
 * @param {String} str String to search.
 * @param {RegExp} regex Regex to search with.
 * @param {Function} callback Function to execute for each match. Invoked with four arguments:
 *   <li>The match array, with named backreference properties.
 *   <li>The zero-based match index.
 *   <li>The string being traversed.
 *   <li>The regex object being used to traverse the string.
 * @param {*} [context] Object to use as `this` when executing `callback`.
 * @returns {*} Provided `context` object.
 * @example
 *
 * // Extracts every other digit from a string
 * XRegExp.forEach('1a2345', /\d/, function(match, i) {
 *   if (i % 2) this.push(+match[0]);
 * }, []);
 * // -> [2, 4]
 */
    self.forEach = function(str, regex, callback, context) {
        var pos = 0,
            i = -1,
            match;

        while ((match = self.exec(str, regex, pos))) {
            // Because `regex` is provided to `callback`, the function can use the deprecated/
            // nonstandard `RegExp.prototype.compile` to mutate the regex. However, since
            // `XRegExp.exec` doesn't use `lastIndex` to set the search position, this can't lead
            // to an infinite loop, at least. Actually, because of the way `XRegExp.exec` caches
            // globalized versions of regexes, mutating the regex will not have any effect on the
            // iteration or matched strings, which is a nice side effect that brings extra safety
            callback.call(context, match, ++i, str, regex);

            pos = match.index + (match[0].length || 1);
        }

        return context;
    };

/**
 * Copies a regex object and adds flag `g`. The copy maintains special properties for named
 * capture, is augmented with `XRegExp.prototype` methods, and has a fresh `lastIndex` property
 * (set to zero). Native regexes are not recompiled using XRegExp syntax.
 * @memberOf XRegExp
 * @param {RegExp} regex Regex to globalize.
 * @returns {RegExp} Copy of the provided regex with flag `g` added.
 * @example
 *
 * var globalCopy = XRegExp.globalize(/regex/);
 * globalCopy.global; // -> true
 */
    self.globalize = function(regex) {
        return copy(regex, {add: 'g', addProto: true});
    };

/**
 * Installs optional features according to the specified options. Can be undone using
 * {@link #XRegExp.uninstall}.
 * @memberOf XRegExp
 * @param {Object|String} options Options object or string.
 * @example
 *
 * // With an options object
 * XRegExp.install({
 *   // Enables support for astral code points in Unicode addons (implicitly sets flag A)
 *   astral: true,
 *
 *   // Overrides native regex methods with fixed/extended versions that support named
 *   // backreferences and fix numerous cross-browser bugs
 *   natives: true
 * });
 *
 * // With an options string
 * XRegExp.install('astral natives');
 */
    self.install = function(options) {
        options = prepareOptions(options);

        if (!features.astral && options.astral) {
            setAstral(true);
        }

        if (!features.natives && options.natives) {
            setNatives(true);
        }
    };

/**
 * Checks whether an individual optional feature is installed.
 * @memberOf XRegExp
 * @param {String} feature Name of the feature to check. One of:
 *   <li>`natives`
 *   <li>`astral`
 * @returns {Boolean} Whether the feature is installed.
 * @example
 *
 * XRegExp.isInstalled('natives');
 */
    self.isInstalled = function(feature) {
        return !!(features[feature]);
    };

/**
 * Returns `true` if an object is a regex; `false` if it isn't. This works correctly for regexes
 * created in another frame, when `instanceof` and `constructor` checks would fail.
 * @memberOf XRegExp
 * @param {*} value Object to check.
 * @returns {Boolean} Whether the object is a `RegExp` object.
 * @example
 *
 * XRegExp.isRegExp('string'); // -> false
 * XRegExp.isRegExp(/regex/i); // -> true
 * XRegExp.isRegExp(RegExp('^', 'm')); // -> true
 * XRegExp.isRegExp(XRegExp('(?s).')); // -> true
 */
    self.isRegExp = function(value) {
        return isType(value, 'RegExp');
    };

/**
 * Returns the first matched string, or in global mode, an array containing all matched strings.
 * This is essentially a more convenient re-implementation of `String.prototype.match` that gives
 * the result types you actually want (string instead of `exec`-style array in match-first mode,
 * and an empty array instead of `null` when no matches are found in match-all mode). It also lets
 * you override flag g and ignore `lastIndex`, and fixes browser bugs.
 * @memberOf XRegExp
 * @param {String} str String to search.
 * @param {RegExp} regex Regex to search with.
 * @param {String} [scope='one'] Use 'one' to return the first match as a string. Use 'all' to
 *   return an array of all matched strings. If not explicitly specified and `regex` uses flag g,
 *   `scope` is 'all'.
 * @returns {String|Array} In match-first mode: First match as a string, or `null`. In match-all
 *   mode: Array of all matched strings, or an empty array.
 * @example
 *
 * // Match first
 * XRegExp.match('abc', /\w/); // -> 'a'
 * XRegExp.match('abc', /\w/g, 'one'); // -> 'a'
 * XRegExp.match('abc', /x/g, 'one'); // -> null
 *
 * // Match all
 * XRegExp.match('abc', /\w/g); // -> ['a', 'b', 'c']
 * XRegExp.match('abc', /\w/, 'all'); // -> ['a', 'b', 'c']
 * XRegExp.match('abc', /x/, 'all'); // -> []
 */
    self.match = function(str, regex, scope) {
        var global = (regex.global && scope !== 'one') || scope === 'all',
            cacheFlags = (global ? 'g' : '') + (regex.sticky ? 'y' : ''),
            result,
            r2;

        regex[REGEX_DATA] = regex[REGEX_DATA] || getBaseProps();

        // Shares cached copies with `XRegExp.exec`/`replace`
        r2 = regex[REGEX_DATA][cacheFlags || 'noGY'] || (
            regex[REGEX_DATA][cacheFlags || 'noGY'] = copy(regex, {
                add: cacheFlags,
                remove: scope === 'one' ? 'g' : ''
            })
        );

        result = nativ.match.call(toObject(str), r2);

        if (regex.global) {
            regex.lastIndex = (
                (scope === 'one' && result) ?
                    // Can't use `r2.lastIndex` since `r2` is nonglobal in this case
                    (result.index + result[0].length) : 0
            );
        }

        return global ? (result || []) : (result && result[0]);
    };

/**
 * Retrieves the matches from searching a string using a chain of regexes that successively search
 * within previous matches. The provided `chain` array can contain regexes and objects with `regex`
 * and `backref` properties. When a backreference is specified, the named or numbered backreference
 * is passed forward to the next regex or returned.
 * @memberOf XRegExp
 * @param {String} str String to search.
 * @param {Array} chain Regexes that each search for matches within preceding results.
 * @returns {Array} Matches by the last regex in the chain, or an empty array.
 * @example
 *
 * // Basic usage; matches numbers within <b> tags
 * XRegExp.matchChain('1 <b>2</b> 3 <b>4 a 56</b>', [
 *   XRegExp('(?is)<b>.*?</b>'),
 *   /\d+/
 * ]);
 * // -> ['2', '4', '56']
 *
 * // Passing forward and returning specific backreferences
 * html = '<a href="http://xregexp.com/api/">XRegExp</a>\
 *         <a href="http://www.google.com/">Google</a>';
 * XRegExp.matchChain(html, [
 *   {regex: /<a href="([^"]+)">/i, backref: 1},
 *   {regex: XRegExp('(?i)^https?://(?<domain>[^/?#]+)'), backref: 'domain'}
 * ]);
 * // -> ['xregexp.com', 'www.google.com']
 */
    self.matchChain = function(str, chain) {
        return (function recurseChain(values, level) {
            var item = chain[level].regex ? chain[level] : {regex: chain[level]},
                matches = [],
                addMatch = function(match) {
                    if (item.backref) {
                        /* Safari 4.0.5 (but not 5.0.5+) inappropriately uses sparse arrays to hold
                         * the `undefined`s for backreferences to nonparticipating capturing
                         * groups. In such cases, a `hasOwnProperty` or `in` check on its own would
                         * inappropriately throw the exception, so also check if the backreference
                         * is a number that is within the bounds of the array.
                         */
                        if (!(match.hasOwnProperty(item.backref) || +item.backref < match.length)) {
                            throw new ReferenceError('Backreference to undefined group: ' + item.backref);
                        }

                        matches.push(match[item.backref] || '');
                    } else {
                        matches.push(match[0]);
                    }
                },
                i;

            for (i = 0; i < values.length; ++i) {
                self.forEach(values[i], item.regex, addMatch);
            }

            return ((level === chain.length - 1) || !matches.length) ?
                matches :
                recurseChain(matches, level + 1);
        }([str], 0));
    };

/**
 * Returns a new string with one or all matches of a pattern replaced. The pattern can be a string
 * or regex, and the replacement can be a string or a function to be called for each match. To
 * perform a global search and replace, use the optional `scope` argument or include flag g if
 * using a regex. Replacement strings can use `${n}` for named and numbered backreferences.
 * Replacement functions can use named backreferences via `arguments[0].name`. Also fixes browser
 * bugs compared to the native `String.prototype.replace` and can be used reliably cross-browser.
 * @memberOf XRegExp
 * @param {String} str String to search.
 * @param {RegExp|String} search Search pattern to be replaced.
 * @param {String|Function} replacement Replacement string or a function invoked to create it.
 *   Replacement strings can include special replacement syntax:
 *     <li>$$ - Inserts a literal $ character.
 *     <li>$&, $0 - Inserts the matched substring.
 *     <li>$` - Inserts the string that precedes the matched substring (left context).
 *     <li>$' - Inserts the string that follows the matched substring (right context).
 *     <li>$n, $nn - Where n/nn are digits referencing an existent capturing group, inserts
 *       backreference n/nn.
 *     <li>${n} - Where n is a name or any number of digits that reference an existent capturing
 *       group, inserts backreference n.
 *   Replacement functions are invoked with three or more arguments:
 *     <li>The matched substring (corresponds to $& above). Named backreferences are accessible as
 *       properties of this first argument.
 *     <li>0..n arguments, one for each backreference (corresponding to $1, $2, etc. above).
 *     <li>The zero-based index of the match within the total search string.
 *     <li>The total string being searched.
 * @param {String} [scope='one'] Use 'one' to replace the first match only, or 'all'. If not
 *   explicitly specified and using a regex with flag g, `scope` is 'all'.
 * @returns {String} New string with one or all matches replaced.
 * @example
 *
 * // Regex search, using named backreferences in replacement string
 * var name = XRegExp('(?<first>\\w+) (?<last>\\w+)');
 * XRegExp.replace('John Smith', name, '${last}, ${first}');
 * // -> 'Smith, John'
 *
 * // Regex search, using named backreferences in replacement function
 * XRegExp.replace('John Smith', name, function(match) {
 *   return match.last + ', ' + match.first;
 * });
 * // -> 'Smith, John'
 *
 * // String search, with replace-all
 * XRegExp.replace('RegExp builds RegExps', 'RegExp', 'XRegExp', 'all');
 * // -> 'XRegExp builds XRegExps'
 */
    self.replace = function(str, search, replacement, scope) {
        var isRegex = self.isRegExp(search),
            global = (search.global && scope !== 'one') || scope === 'all',
            cacheFlags = (global ? 'g' : '') + (search.sticky ? 'y' : ''),
            s2 = search,
            result;

        if (isRegex) {
            search[REGEX_DATA] = search[REGEX_DATA] || getBaseProps();

            // Shares cached copies with `XRegExp.exec`/`match`. Since a copy is used,
            // `search`'s `lastIndex` isn't updated *during* replacement iterations
            s2 = search[REGEX_DATA][cacheFlags || 'noGY'] || (
                search[REGEX_DATA][cacheFlags || 'noGY'] = copy(search, {
                    add: cacheFlags,
                    remove: scope === 'one' ? 'g' : ''
                })
            );
        } else if (global) {
            s2 = new RegExp(self.escape(String(search)), 'g');
        }

        // Fixed `replace` required for named backreferences, etc.
        result = fixed.replace.call(toObject(str), s2, replacement);

        if (isRegex && search.global) {
            // Fixes IE, Safari bug (last tested IE 9, Safari 5.1)
            search.lastIndex = 0;
        }

        return result;
    };

/**
 * Performs batch processing of string replacements. Used like {@link #XRegExp.replace}, but
 * accepts an array of replacement details. Later replacements operate on the output of earlier
 * replacements. Replacement details are accepted as an array with a regex or string to search for,
 * the replacement string or function, and an optional scope of 'one' or 'all'. Uses the XRegExp
 * replacement text syntax, which supports named backreference properties via `${name}`.
 * @memberOf XRegExp
 * @param {String} str String to search.
 * @param {Array} replacements Array of replacement detail arrays.
 * @returns {String} New string with all replacements.
 * @example
 *
 * str = XRegExp.replaceEach(str, [
 *   [XRegExp('(?<name>a)'), 'z${name}'],
 *   [/b/gi, 'y'],
 *   [/c/g, 'x', 'one'], // scope 'one' overrides /g
 *   [/d/, 'w', 'all'],  // scope 'all' overrides lack of /g
 *   ['e', 'v', 'all'],  // scope 'all' allows replace-all for strings
 *   [/f/g, function($0) {
 *     return $0.toUpperCase();
 *   }]
 * ]);
 */
    self.replaceEach = function(str, replacements) {
        var i, r;

        for (i = 0; i < replacements.length; ++i) {
            r = replacements[i];
            str = self.replace(str, r[0], r[1], r[2]);
        }

        return str;
    };

/**
 * Splits a string into an array of strings using a regex or string separator. Matches of the
 * separator are not included in the result array. However, if `separator` is a regex that contains
 * capturing groups, backreferences are spliced into the result each time `separator` is matched.
 * Fixes browser bugs compared to the native `String.prototype.split` and can be used reliably
 * cross-browser.
 * @memberOf XRegExp
 * @param {String} str String to split.
 * @param {RegExp|String} separator Regex or string to use for separating the string.
 * @param {Number} [limit] Maximum number of items to include in the result array.
 * @returns {Array} Array of substrings.
 * @example
 *
 * // Basic use
 * XRegExp.split('a b c', ' ');
 * // -> ['a', 'b', 'c']
 *
 * // With limit
 * XRegExp.split('a b c', ' ', 2);
 * // -> ['a', 'b']
 *
 * // Backreferences in result array
 * XRegExp.split('..word1..', /([a-z]+)(\d+)/i);
 * // -> ['..', 'word', '1', '..']
 */
    self.split = function(str, separator, limit) {
        return fixed.split.call(toObject(str), separator, limit);
    };

/**
 * Executes a regex search in a specified string. Returns `true` or `false`. Optional `pos` and
 * `sticky` arguments specify the search start position, and whether the match must start at the
 * specified position only. The `lastIndex` property of the provided regex is not used, but is
 * updated for compatibility. Also fixes browser bugs compared to the native
 * `RegExp.prototype.test` and can be used reliably cross-browser.
 * @memberOf XRegExp
 * @param {String} str String to search.
 * @param {RegExp} regex Regex to search with.
 * @param {Number} [pos=0] Zero-based index at which to start the search.
 * @param {Boolean|String} [sticky=false] Whether the match must start at the specified position
 *   only. The string `'sticky'` is accepted as an alternative to `true`.
 * @returns {Boolean} Whether the regex matched the provided value.
 * @example
 *
 * // Basic use
 * XRegExp.test('abc', /c/); // -> true
 *
 * // With pos and sticky
 * XRegExp.test('abc', /c/, 0, 'sticky'); // -> false
 */
    self.test = function(str, regex, pos, sticky) {
        // Do this the easy way :-)
        return !!self.exec(str, regex, pos, sticky);
    };

/**
 * Uninstalls optional features according to the specified options. All optional features start out
 * uninstalled, so this is used to undo the actions of {@link #XRegExp.install}.
 * @memberOf XRegExp
 * @param {Object|String} options Options object or string.
 * @example
 *
 * // With an options object
 * XRegExp.uninstall({
 *   // Disables support for astral code points in Unicode addons
 *   astral: true,
 *
 *   // Restores native regex methods
 *   natives: true
 * });
 *
 * // With an options string
 * XRegExp.uninstall('astral natives');
 */
    self.uninstall = function(options) {
        options = prepareOptions(options);

        if (features.astral && options.astral) {
            setAstral(false);
        }

        if (features.natives && options.natives) {
            setNatives(false);
        }
    };

/**
 * Returns an XRegExp object that is the union of the given patterns. Patterns can be provided as
 * regex objects or strings. Metacharacters are escaped in patterns provided as strings.
 * Backreferences in provided regex objects are automatically renumbered to work correctly. Native
 * flags used by provided regexes are ignored in favor of the `flags` argument.
 * @memberOf XRegExp
 * @param {Array} patterns Regexes and strings to combine.
 * @param {String} [flags] Any combination of XRegExp flags.
 * @returns {RegExp} Union of the provided regexes and strings.
 * @example
 *
 * XRegExp.union(['a+b*c', /(dogs)\1/, /(cats)\1/], 'i');
 * // -> /a\+b\*c|(dogs)\1|(cats)\2/i
 */
    self.union = function(patterns, flags) {
        var parts = /(\()(?!\?)|\\([1-9]\d*)|\\[\s\S]|\[(?:[^\\\]]|\\[\s\S])*]/g,
            output = [],
            numCaptures = 0,
            numPriorCaptures,
            captureNames,
            pattern,
            rewrite = function(match, paren, backref) {
                var name = captureNames[numCaptures - numPriorCaptures];

                // Capturing group
                if (paren) {
                    ++numCaptures;
                    // If the current capture has a name, preserve the name
                    if (name) {
                        return '(?<' + name + '>';
                    }
                // Backreference
                } else if (backref) {
                    // Rewrite the backreference
                    return '\\' + (+backref + numPriorCaptures);
                }

                return match;
            },
            i;

        if (!(isType(patterns, 'Array') && patterns.length)) {
            throw new TypeError('Must provide a nonempty array of patterns to merge');
        }

        for (i = 0; i < patterns.length; ++i) {
            pattern = patterns[i];

            if (self.isRegExp(pattern)) {
                numPriorCaptures = numCaptures;
                captureNames = (pattern[REGEX_DATA] && pattern[REGEX_DATA].captureNames) || [];

                // Rewrite backreferences. Passing to XRegExp dies on octals and ensures patterns
                // are independently valid; helps keep this simple. Named captures are put back
                output.push(self(pattern.source).source.replace(parts, rewrite));
            } else {
                output.push(self.escape(pattern));
            }
        }

        return self(output.join('|'), flags);
    };

/* ==============================
 * Fixed/extended native methods
 * ============================== */

/**
 * Adds named capture support (with backreferences returned as `result.name`), and fixes browser
 * bugs in the native `RegExp.prototype.exec`. Calling `XRegExp.install('natives')` uses this to
 * override the native method. Use via `XRegExp.exec` without overriding natives.
 * @private
 * @param {String} str String to search.
 * @returns {Array} Match array with named backreference properties, or `null`.
 */
    fixed.exec = function(str) {
        var origLastIndex = this.lastIndex,
            match = nativ.exec.apply(this, arguments),
            name,
            r2,
            i;

        if (match) {
            // Fix browsers whose `exec` methods don't return `undefined` for nonparticipating
            // capturing groups. This fixes IE 5.5-8, but not IE 9's quirks mode or emulation of
            // older IEs. IE 9 in standards mode follows the spec
            if (!correctExecNpcg && match.length > 1 && indexOf(match, '') > -1) {
                r2 = copy(this, {remove: 'g'});
                // Using `str.slice(match.index)` rather than `match[0]` in case lookahead allowed
                // matching due to characters outside the match
                nativ.replace.call(String(str).slice(match.index), r2, function() {
                    var len = arguments.length, i;
                    // Skip index 0 and the last 2
                    for (i = 1; i < len - 2; ++i) {
                        if (arguments[i] === undefined) {
                            match[i] = undefined;
                        }
                    }
                });
            }

            // Attach named capture properties
            if (this[REGEX_DATA] && this[REGEX_DATA].captureNames) {
                // Skip index 0
                for (i = 1; i < match.length; ++i) {
                    name = this[REGEX_DATA].captureNames[i - 1];
                    if (name) {
                        match[name] = match[i];
                    }
                }
            }

            // Fix browsers that increment `lastIndex` after zero-length matches
            if (this.global && !match[0].length && (this.lastIndex > match.index)) {
                this.lastIndex = match.index;
            }
        }

        if (!this.global) {
            // Fixes IE, Opera bug (last tested IE 9, Opera 11.6)
            this.lastIndex = origLastIndex;
        }

        return match;
    };

/**
 * Fixes browser bugs in the native `RegExp.prototype.test`. Calling `XRegExp.install('natives')`
 * uses this to override the native method.
 * @private
 * @param {String} str String to search.
 * @returns {Boolean} Whether the regex matched the provided value.
 */
    fixed.test = function(str) {
        // Do this the easy way :-)
        return !!fixed.exec.call(this, str);
    };

/**
 * Adds named capture support (with backreferences returned as `result.name`), and fixes browser
 * bugs in the native `String.prototype.match`. Calling `XRegExp.install('natives')` uses this to
 * override the native method.
 * @private
 * @param {RegExp|*} regex Regex to search with. If not a regex object, it is passed to `RegExp`.
 * @returns {Array} If `regex` uses flag g, an array of match strings or `null`. Without flag g,
 *   the result of calling `regex.exec(this)`.
 */
    fixed.match = function(regex) {
        var result;

        if (!self.isRegExp(regex)) {
            // Use the native `RegExp` rather than `XRegExp`
            regex = new RegExp(regex);
        } else if (regex.global) {
            result = nativ.match.apply(this, arguments);
            // Fixes IE bug
            regex.lastIndex = 0;

            return result;
        }

        return fixed.exec.call(regex, toObject(this));
    };

/**
 * Adds support for `${n}` tokens for named and numbered backreferences in replacement text, and
 * provides named backreferences to replacement functions as `arguments[0].name`. Also fixes
 * browser bugs in replacement text syntax when performing a replacement using a nonregex search
 * value, and the value of a replacement regex's `lastIndex` property during replacement iterations
 * and upon completion. Note that this doesn't support SpiderMonkey's proprietary third (`flags`)
 * argument. Calling `XRegExp.install('natives')` uses this to override the native method. Use via
 * `XRegExp.replace` without overriding natives.
 * @private
 * @param {RegExp|String} search Search pattern to be replaced.
 * @param {String|Function} replacement Replacement string or a function invoked to create it.
 * @returns {String} New string with one or all matches replaced.
 */
    fixed.replace = function(search, replacement) {
        var isRegex = self.isRegExp(search),
            origLastIndex,
            captureNames,
            result;

        if (isRegex) {
            if (search[REGEX_DATA]) {
                captureNames = search[REGEX_DATA].captureNames;
            }
            // Only needed if `search` is nonglobal
            origLastIndex = search.lastIndex;
        } else {
            search += ''; // Type-convert
        }

        // Don't use `typeof`; some older browsers return 'function' for regex objects
        if (isType(replacement, 'Function')) {
            // Stringifying `this` fixes a bug in IE < 9 where the last argument in replacement
            // functions isn't type-converted to a string
            result = nativ.replace.call(String(this), search, function() {
                var args = arguments, i;
                if (captureNames) {
                    // Change the `arguments[0]` string primitive to a `String` object that can
                    // store properties. This really does need to use `String` as a constructor
                    args[0] = new String(args[0]);
                    // Store named backreferences on the first argument
                    for (i = 0; i < captureNames.length; ++i) {
                        if (captureNames[i]) {
                            args[0][captureNames[i]] = args[i + 1];
                        }
                    }
                }
                // Update `lastIndex` before calling `replacement`. Fixes IE, Chrome, Firefox,
                // Safari bug (last tested IE 9, Chrome 17, Firefox 11, Safari 5.1)
                if (isRegex && search.global) {
                    search.lastIndex = args[args.length - 2] + args[0].length;
                }
                // Should pass `undefined` as context; see
                // <https://bugs.ecmascript.org/show_bug.cgi?id=154>
                return replacement.apply(undefined, args);
            });
        } else {
            // Ensure that the last value of `args` will be a string when given nonstring `this`,
            // while still throwing on `null` or `undefined` context
            result = nativ.replace.call(this == null ? this : String(this), search, function() {
                // Keep this function's `arguments` available through closure
                var args = arguments;
                return nativ.replace.call(String(replacement), replacementToken, function($0, $1, $2) {
                    var n;
                    // Named or numbered backreference with curly braces
                    if ($1) {
                        /* XRegExp behavior for `${n}`:
                         * 1. Backreference to numbered capture, if `n` is an integer. Use `0` for
                         *    for the entire match. Any number of leading zeros may be used.
                         * 2. Backreference to named capture `n`, if it exists and is not an
                         *    integer overridden by numbered capture. In practice, this does not
                         *    overlap with numbered capture since XRegExp does not allow named
                         *    capture to use a bare integer as the name.
                         * 3. If the name or number does not refer to an existing capturing group,
                         *    it's an error.
                         */
                        n = +$1; // Type-convert; drop leading zeros
                        if (n <= args.length - 3) {
                            return args[n] || '';
                        }
                        // Groups with the same name is an error, else would need `lastIndexOf`
                        n = captureNames ? indexOf(captureNames, $1) : -1;
                        if (n < 0) {
                            throw new SyntaxError('Backreference to undefined group ' + $0);
                        }
                        return args[n + 1] || '';
                    }
                    // Else, special variable or numbered backreference without curly braces
                    if ($2 === '$') { // $$
                        return '$';
                    }
                    if ($2 === '&' || +$2 === 0) { // $&, $0 (not followed by 1-9), $00
                        return args[0];
                    }
                    if ($2 === '`') { // $` (left context)
                        return args[args.length - 1].slice(0, args[args.length - 2]);
                    }
                    if ($2 === "'") { // $' (right context)
                        return args[args.length - 1].slice(args[args.length - 2] + args[0].length);
                    }
                    // Else, numbered backreference without curly braces
                    $2 = +$2; // Type-convert; drop leading zero
                    /* XRegExp behavior for `$n` and `$nn`:
                     * - Backrefs end after 1 or 2 digits. Use `${..}` for more digits.
                     * - `$1` is an error if no capturing groups.
                     * - `$10` is an error if less than 10 capturing groups. Use `${1}0` instead.
                     * - `$01` is `$1` if at least one capturing group, else it's an error.
                     * - `$0` (not followed by 1-9) and `$00` are the entire match.
                     * Native behavior, for comparison:
                     * - Backrefs end after 1 or 2 digits. Cannot reference capturing group 100+.
                     * - `$1` is a literal `$1` if no capturing groups.
                     * - `$10` is `$1` followed by a literal `0` if less than 10 capturing groups.
                     * - `$01` is `$1` if at least one capturing group, else it's a literal `$01`.
                     * - `$0` is a literal `$0`.
                     */
                    if (!isNaN($2)) {
                        if ($2 > args.length - 3) {
                            throw new SyntaxError('Backreference to undefined group ' + $0);
                        }
                        return args[$2] || '';
                    }
                    throw new SyntaxError('Invalid token ' + $0);
                });
            });
        }

        if (isRegex) {
            if (search.global) {
                // Fixes IE, Safari bug (last tested IE 9, Safari 5.1)
                search.lastIndex = 0;
            } else {
                // Fixes IE, Opera bug (last tested IE 9, Opera 11.6)
                search.lastIndex = origLastIndex;
            }
        }

        return result;
    };

/**
 * Fixes browser bugs in the native `String.prototype.split`. Calling `XRegExp.install('natives')`
 * uses this to override the native method. Use via `XRegExp.split` without overriding natives.
 * @private
 * @param {RegExp|String} separator Regex or string to use for separating the string.
 * @param {Number} [limit] Maximum number of items to include in the result array.
 * @returns {Array} Array of substrings.
 */
    fixed.split = function(separator, limit) {
        if (!self.isRegExp(separator)) {
            // Browsers handle nonregex split correctly, so use the faster native method
            return nativ.split.apply(this, arguments);
        }

        var str = String(this),
            output = [],
            origLastIndex = separator.lastIndex,
            lastLastIndex = 0,
            lastLength;

        /* Values for `limit`, per the spec:
         * If undefined: pow(2,32) - 1
         * If 0, Infinity, or NaN: 0
         * If positive number: limit = floor(limit); if (limit >= pow(2,32)) limit -= pow(2,32);
         * If negative number: pow(2,32) - floor(abs(limit))
         * If other: Type-convert, then use the above rules
         */
        // This line fails in very strange ways for some values of `limit` in Opera 10.5-10.63,
        // unless Opera Dragonfly is open (go figure). It works in at least Opera 9.5-10.1 and 11+
        limit = (limit === undefined ? -1 : limit) >>> 0;

        self.forEach(str, separator, function(match) {
            // This condition is not the same as `if (match[0].length)`
            if ((match.index + match[0].length) > lastLastIndex) {
                output.push(str.slice(lastLastIndex, match.index));
                if (match.length > 1 && match.index < str.length) {
                    Array.prototype.push.apply(output, match.slice(1));
                }
                lastLength = match[0].length;
                lastLastIndex = match.index + lastLength;
            }
        });

        if (lastLastIndex === str.length) {
            if (!nativ.test.call(separator, '') || lastLength) {
                output.push('');
            }
        } else {
            output.push(str.slice(lastLastIndex));
        }

        separator.lastIndex = origLastIndex;
        return output.length > limit ? output.slice(0, limit) : output;
    };

/* ==============================
 * Built-in syntax/flag tokens
 * ============================== */

    add = self.addToken;

/* Letter identity escapes that natively match literal characters: `\a`, `\A`, etc. These should be
 * SyntaxErrors but are allowed in web reality. XRegExp makes them errors for cross-browser
 * consistency and to reserve their syntax, but lets them be superseded by addons.
 */
    add(
        /\\([ABCE-RTUVXYZaeg-mopqyz]|c(?![A-Za-z])|u(?![\dA-Fa-f]{4})|x(?![\dA-Fa-f]{2}))/,
        function(match, scope) {
            // \B is allowed in default scope only
            if (match[1] === 'B' && scope === defaultScope) {
                return match[0];
            }
            throw new SyntaxError('Invalid escape ' + match[0]);
        },
        {scope: 'all'}
    );

/* Empty character class: `[]` or `[^]`. This fixes a critical cross-browser syntax inconsistency.
 * Unless this is standardized (per the ES spec), regex syntax can't be accurately parsed because
 * character class endings can't be determined.
 */
    add(
        /\[(\^?)]/,
        function(match) {
            // For cross-browser compatibility with ES3, convert [] to \b\B and [^] to [\s\S].
            // (?!) should work like \b\B, but is unreliable in some versions of Firefox
            return match[1] ? '[\\s\\S]' : '\\b\\B';
        }
    );

/* Comment pattern: `(?# )`. Inline comments are an alternative to the line comments allowed in
 * free-spacing mode (flag x).
 */
    add(
        /\(\?#[^)]*\)/,
        function(match, scope, flags) {
            // Keep tokens separated unless the following token is a quantifier
            return isQuantifierNext(match.input, match.index + match[0].length, flags) ?
                '' : '(?:)';
        }
    );

/* Whitespace and line comments, in free-spacing mode (aka extended mode, flag x) only.
 */
    add(
        /\s+|#.*/,
        function(match, scope, flags) {
            // Keep tokens separated unless the following token is a quantifier
            return isQuantifierNext(match.input, match.index + match[0].length, flags) ?
                '' : '(?:)';
        },
        {flag: 'x'}
    );

/* Dot, in dotall mode (aka singleline mode, flag s) only.
 */
    add(
        /\./,
        function() {
            return '[\\s\\S]';
        },
        {flag: 's'}
    );

/* Named backreference: `\k<name>`. Backreference names can use the characters A-Z, a-z, 0-9, _,
 * and $ only.
 */
    add(
        /\\k<([\w$]+)>/,
        function(match) {
            // Groups with the same name is an error, else would need `lastIndexOf`
            var index = isNaN(match[1]) ? (indexOf(this.captureNames, match[1]) + 1) : +match[1],
                endIndex = match.index + match[0].length;
            if (!index || index > this.captureNames.length) {
                throw new SyntaxError('Backreference to undefined group ' + match[0]);
            }
            // Keep backreferences separate from subsequent literal numbers
            return '\\' + index + (
                endIndex === match.input.length || isNaN(match.input.charAt(endIndex)) ?
                    '' : '(?:)'
            );
        }
    );

/* Numbered backreference or octal, plus any following digits: `\0`, `\11`, etc. Octals except `\0`
 * not followed by 0-9 and backreferences to unopened capture groups throw an error. Other matches
 * are returned unaltered. IE < 9 doesn't support backreferences above `\99` in regex syntax.
 */
    add(
        /\\(\d+)/,
        function(match, scope) {
            if (
                !(
                    scope === defaultScope &&
                    /^[1-9]/.test(match[1]) &&
                    +match[1] <= this.captureNames.length
                ) &&
                match[1] !== '0'
            ) {
                throw new SyntaxError('Cannot use octal escape or backreference to undefined group ' +
                    match[0]);
            }
            return match[0];
        },
        {scope: 'all'}
    );

/* Named capturing group; match the opening delimiter only: `(?<name>`. Capture names can use the
 * characters A-Z, a-z, 0-9, _, and $ only. Names can't be integers. Supports Python-style
 * `(?P<name>` as an alternate syntax to avoid issues in recent Opera (which natively supports the
 * Python-style syntax). Otherwise, XRegExp might treat numbered backreferences to Python-style
 * named capture as octals.
 */
    add(
        /\(\?P?<([\w$]+)>/,
        function(match) {
            // Disallow bare integers as names because named backreferences are added to match
            // arrays and therefore numeric properties may lead to incorrect lookups
            if (!isNaN(match[1])) {
                throw new SyntaxError('Cannot use integer as capture name ' + match[0]);
            }
            if (match[1] === 'length' || match[1] === '__proto__') {
                throw new SyntaxError('Cannot use reserved word as capture name ' + match[0]);
            }
            if (indexOf(this.captureNames, match[1]) > -1) {
                throw new SyntaxError('Cannot use same name for multiple groups ' + match[0]);
            }
            this.captureNames.push(match[1]);
            this.hasNamedCapture = true;
            return '(';
        }
    );

/* Capturing group; match the opening parenthesis only. Required for support of named capturing
 * groups. Also adds explicit capture mode (flag n).
 */
    add(
        /\((?!\?)/,
        function(match, scope, flags) {
            if (flags.indexOf('n') > -1) {
                return '(?:';
            }
            this.captureNames.push(null);
            return '(';
        },
        {optionalFlags: 'n'}
    );

/* ==============================
 * Expose XRegExp
 * ============================== */

    return self;

}());