Flagrant Badassery

Five years ago I posted Mimicking Lookbehind in JavaScript on this blog, wherein I detailed several ways to emulate positive and negative lookbehind in JavaScript. My approaches back then were all fairly rough, and it was complicated to properly customize any of them to work with a given pattern. Plus, they were only designed to simulate lookbehind in a regex-based replacement.

To make it much easier to use lookbehind, I recently posted a collection of short functions on GitHub. They use XRegExp v2, so you should check that out, too.

Here's the code:

// Simulating infinite-length leading lookbehind in JavaScript. Uses XRegExp.
// Captures within lookbehind are not included in match results. Lazy
// repetition in lookbehind may lead to unexpected results.

(function (XRegExp) {

    function prepareLb(lb) {
        // Allow mode modifier before lookbehind
        var parts = /^((?:\(\?[\w$]+\))?)\(\?<([=!])([\s\S]*)\)$/.exec(lb);
        return {
            // $(?!\s) allows use of (?m) in lookbehind
            lb: XRegExp(parts ? parts[1] + "(?:" + parts[3] + ")$(?!\\s)" : lb),
            // Positive or negative lookbehind. Use positive if no lookbehind group
            type: parts ? parts[2] === "=" : !parts
        };
    }

    XRegExp.execLb = function (str, lb, regex) {
        var pos = 0, match, leftContext;
        lb = prepareLb(lb);
        while (match = XRegExp.exec(str, regex, pos)) {
            leftContext = str.slice(0, match.index);
            if (lb.type === lb.lb.test(leftContext)) {
                return match;
            }
            pos = match.index + 1;
        }
        return null;
    };

    XRegExp.testLb = function (str, lb, regex) {
        return !!XRegExp.execLb(str, lb, regex);
    };

    XRegExp.searchLb = function (str, lb, regex) {
        var match = XRegExp.execLb(str, lb, regex);
        return match ? match.index : -1;
    };

    XRegExp.matchAllLb = function (str, lb, regex) {
        var matches = [], pos = 0, match, leftContext;
        lb = prepareLb(lb);
        while (match = XRegExp.exec(str, regex, pos)) {
            leftContext = str.slice(0, match.index);
            if (lb.type === lb.lb.test(leftContext)) {
                matches.push(match[0]);
                pos = match.index + (match[0].length || 1);
            } else {
                pos = match.index + 1;
            }
        }
        return matches;
    };

    XRegExp.replaceLb = function (str, lb, regex, replacement) {
        var output = "", pos = 0, lastEnd = 0, match, leftContext;
        lb = prepareLb(lb);
        while (match = XRegExp.exec(str, regex, pos)) {
            leftContext = str.slice(0, match.index);
            if (lb.type === lb.lb.test(leftContext)) {
                // Doesn't work correctly if lookahead in regex looks outside of the match
                output += str.slice(lastEnd, match.index) + XRegExp.replace(match[0], regex, replacement);
                lastEnd = match.index + match[0].length;
                if (!regex.global) {
                    break;
                }
                pos = match.index + (match[0].length || 1);
            } else {
                pos = match.index + 1;
            }
        }
        return output + str.slice(lastEnd);
    };

}(XRegExp));

That's less than 0.5 KB after minification and gzipping. It provides a collection of functions that make it simple to emulate leading lookbehind:

• XRegExp.execLb
• XRegExp.testLb
• XRegExp.searchLb
• XRegExp.matchAllLb
• XRegExp.replaceLb

Each of these functions takes three arguments: the string to search, the lookbehind pattern as a string (can use XRegExp syntax extensions), and the main regex. XRegExp.replaceLb takes a fourth argument for the replacement value, which can be a string or function.

Usage examples follow:

XRegExp.execLb("Fluffy cat", "(?i)(?<=fluffy\\W+)", XRegExp("(?i)(?<first>c)at"));
// -> ["cat", "c"]
// Result has named backref: result.first -> "c"

XRegExp.execLb("Fluffy cat", "(?i)(?<!fluffy\\W+)", /cat/i);
// -> null

XRegExp.testLb("Fluffy cat", "(?i)(?<=fluffy\\W+)", /cat/i);
// -> true

XRegExp.testLb("Fluffy cat", "(?i)(?<!fluffy\\W+)", /cat/i);
// -> false

XRegExp.searchLb("Catwoman's fluffy cat", "(?i)(?<=fluffy\\W+)", /cat/i);
// -> 18

XRegExp.searchLb("Catwoman's fluffy cat", "(?i)(?<!fluffy\\W+)", /cat/i);
// -> 0

XRegExp.matchAllLb("Catwoman's cats are fluffy cats", "(?i)(?<=fluffy\\W+)", /cat\w*/i);
// -> ["cats"]

XRegExp.matchAllLb("Catwoman's cats are fluffy cats", "(?i)(?<!fluffy\\W+)", /cat\w*/i);
// -> ["Catwoman", "cats"]

XRegExp.replaceLb("Catwoman's fluffy cat is a cat", "(?i)(?<=fluffy\\W+)", /cat/ig, "dog");
// -> "Catwoman's fluffy dog is a cat"

XRegExp.replaceLb("Catwoman's fluffy cat is a cat", "(?i)(?<!fluffy\\W+)", /cat/ig, "dog");
// -> "dogwoman's fluffy cat is a dog"

XRegExp.replaceLb("Catwoman's fluffy cat is a cat", "(?i)(?<!fluffy\\W+)", /cat/ig, function ($0) {
    var first = $0.charAt(0);
    return first === first.toUpperCase() ? "Dog" : "dog";
});
// -> "Dogwoman's fluffy cat is a dog"

Easy peasy lemon squeezy. 🙂

Recently, I've added three new addons for XRegExp v2.0 (currently in release candidate stage on GitHub):

• XRegExp.build — Lets you build regexes using named subpatterns. Inspired by Lea Verou's RegExp.create.
• XRegExp Prototype Methods — Adds a collection of methods to be inherited by XRegExp regexes: apply, call, forEach, globalize, xexec, and xtest. These also work for native RegExps copied by XRegExp.
• XRegExp Unicode Properties — Includes the remaining nine properties (beyond what's already available in other XRegExp addons) required for Level-1 Unicode support: Alphabetic, Uppercase, Lowercase, White_Space, Noncharacter_Code_Point, Default_Ignorable_Code_Point, Any, ASCII, and Assigned.

Jumping right into some code, the following demonstrates how the new XRegExp.build addon can be used to create a grammatical pattern for matching real numbers:

// Approach 1: Make all of the subpatterns reusable

var lib = {
    digit:             /[0-9]/,
    exponentIndicator: /[Ee]/,
    digitSeparator:    /[_,]/,
    sign:              /[+-]/,
    point:             /[.]/
};
lib.preexponent = XRegExp.build('(?xn)\
    {{sign}} ?              \
    (?= {{digit}}           \
      | {{point}}           \
    )                       \
    ( {{digit}} {1,3}       \
      ( {{digitSeparator}} ?\
        {{digit}} {3}       \
      ) *                   \
    ) ?                     \
    ( {{point}}             \
      {{digit}} +           \
    ) ?                     ',
    lib
);
lib.exponent = XRegExp.build('(?x)\
    {{exponentIndicator}}\
    {{sign}} ?           \
    {{digit}} +          ',
    lib
);
lib.real = XRegExp.build('(?x)\
    ^              \
    {{preexponent}}\
    {{exponent}} ? \
    $              ',
    lib
);

// Approach 2: No need to reuse the subpatterns. {{sign}} and {{digit}} are
// defined twice, but that can be avoided by defining them before constructing
// the main pattern (see Approach 1).

var real = XRegExp.build('(?x)\
    ^              \
    {{preexponent}}\
    {{exponent}} ? \
    $              ',
    {
        preexponent: XRegExp.build('(?xn)\
            {{sign}} ?              \
            (?= {{digit}}           \
              | {{point}}           \
            )                       \
            ( {{digit}} {1,3}       \
              ( {{digitSeparator}} ?\
                {{digit}} {3}       \
              ) *                   \
            ) ?                     \
            ( {{point}}             \
              {{digit}} +           \
            ) ?                     ',
            {
                sign:           /[+-]/,
                digit:          /[0-9]/,
                digitSeparator: /[_,]/,
                point:          /[.]/
            }
        ),
        exponent: XRegExp.build('(?x)\
            {{exponentIndicator}}\
            {{sign}} ?           \
            {{digit}} +          ',
            {
                sign:              /[+-]/,
                digit:             /[0-9]/,
                exponentIndicator: /[Ee]/
            }
        )
    }
);

The real and lib.real regexes created by the above code are identical. Here are a few examples of strings they match:

• -1
• 1,000
• 10_000_000
• 1,111.1111
• 01.0
• .1
• 1e2
• +1.1e-2

And here are a few examples of strings they don't match:

• ,100
• 10,00
• 1,0000
• 1.
• 1.1,111
• 1k

Grammatical patterns like this are easier to read, write, and maintain, and look more like a BNF than the typical line-noisy regular expressions that some people have come to hate.

Note that the {{…}} syntax shown here works only for regexes created by XRegExp.build. Named subpatterns can be provided as strings or regex objects (strings are passed to the XRegExp constructor). The provided patterns are automatically wrapped in (?:…) so they can be quantified as a unit and don't interfere with the surrounding pattern in unexpected ways. A leading ^ and trailing unescaped $ are stripped from subpatterns if both are present, which allows embedding independently useful anchored patterns. Flags can be provided via XRegExp.build's optional third (flags) argument. Native flags used by provided subpatterns are ignored in favor of the flags argument. Backreferences in the outer pattern and provided subpatterns are automatically renumbered to work correctly within the larger combined pattern. The syntax ({{name}}) works as shorthand for named capture via (?<name>{{name}}). The {{…}} syntax can be escaped with a backslash.

Play around with the above details a bit, and I think you'll find that XRegExp.build works intuitively and handles any edge cases you throw at it.

Feel free to share how you might alter the above regexes. And make sure to check out the fancy new XRegExp v2.0 and its upgraded addons at GitHub.