Skip to content

Latest commit

 

History

History
379 lines (275 loc) · 17.9 KB

README.md

File metadata and controls

379 lines (275 loc) · 17.9 KB
Raw

node-re2 NPM version

This project provides bindings for RE2: fast, safe alternative to backtracking regular expression engines written by Russ Cox. To learn more about RE2, start with an overview Regular Expression Matching in the Wild. More resources can be found at his Implementing Regular Expressions page.

RE2's regular expression language is almost a superset of what is provided by RegExp (see Syntax), but it lacks two features: backreferences and lookahead assertions. See below for more details.

RE2 always works in the Unicode mode, which means that all matches that use character codes are interpret as Unicode code points, not as binary values of UTF-16. See RE2.unicodeWarningLevel below for more details.

RE2 object emulates standard RegExp making it a practical drop-in replacement in most cases. RE2 is extended to provide String-based regular expression methods as well. To help to convert RegExp objects to RE2 its constructor can take RegExp directly honoring all properties.

It can work with node.js buffers directly reducing overhead on recoding and copying characters, and making processing/parsing long files fast.

All documentation can be found in this README and in the wiki.

Why use node-re2?

The built-in Node.js regular expression engine can run in exponential time with a special combination:

  • A vulnerable regular expression
  • "Evil input"

This can lead to what is known as a Regular Expression Denial of Service (ReDoS). To tell if your regular expressions are vulnerable, you might try the one of these projects:

However, neither project is perfect.

node-re2 can protect your Node.js application from ReDoS. node-re2 makes vulnerable regular expression patterns safe by evaluating them in RE2 instead of the built-in Node.js regex engine.

Standard features

RE2 object can be created just like RegExp:

Supported properties:

Supported methods:

Starting with 1.6.0 following well-known symbol-based methods are supported (see Symbols):

It allows to use RE2 instances on strings directly, just like RegExp instances:

var re = new RE2("1");
"213".match(re);        // [ '1', index: 1, input: '213' ]
"213".search(re);       // 1
"213".replace(re, "+"); // 2+3
"213".split(re);        // [ '2', '3' ]

Array.from("2131".matchAll(re)); // returns a generator!
// [['1', index: 1, input: '2131'], ['1', index: 3, input: '2131']]

Starting with 1.8.0 named groups are supported.

Extensions

Shortcut construction

RE2 object can be created from a regular expression:

var re1 = new RE2(/ab*/ig); // from a RegExp object
var re2 = new RE2(re1);     // from another RE2 object

String methods

Standard String defines four more methods that can use regular expressions. RE2 provides them as methods exchanging positions of a string, and a regular expression:

Starting with 1.6.0, these methods added as well-known symbol-based methods to be used transparently with ES6 string/regex machinery.

Buffer support

In order to support Buffer directly, most methods can accept buffers instead of strings. It speeds up all operations. Following signatures are supported:

  • re2.exec(buf)
  • re2.test(buf)
  • re2.match(buf)
  • re2.search(buf)
  • re2.split(buf[, limit])
  • re2.replace(buf, replacer)

Differences with their string-based versions:

  • All buffers are assumed to be encoded as UTF-8 (ASCII is a proper subset of UTF-8).
  • Instead of strings they return Buffer objects, even in composite objects. A buffer can be converted to a string with buf.toString().
  • All offsets and lengths are in bytes, rather than characters (each UTF-8 character can occupy from 1 to 4 bytes). This way users can properly slice buffers without costly recalculations from characters to bytes.

When re2.replace() is used with a replacer function, the replacer can return a buffer, or a string. But all arguments (except for an input object) will be strings, and an offset will be in characters. If you prefer to deal with buffers and byte offsets in a replacer function, set a property useBuffers to true on the function:

function strReplacer(match, offset, input) {
	// typeof match == "string"
	return "<= " + offset + " characters|";
}

RE2("б").replace("абв", strReplacer);
// "а<= 1 characters|в"

function bufReplacer(match, offset, input) {
	// typeof match == "string"
	return "<= " + offset + " bytes|";
}
bufReplacer.useBuffers = true;

RE2("б").replace("абв", bufReplacer);
// "а<= 2 bytes|в"

This feature works for string and buffer inputs. If a buffer was used as an input, its output will be returned as a buffer too, otherwise a string will be returned.

Calculate length

Two functions to calculate string sizes between UTF-8 and UTF-16 are exposed on RE2:

  • RE2.getUtf8Length(str) — calculates a buffer size in bytes to encode a UTF-16 string as a UTF-8 buffer.
  • RE2.getUtf16Length(buf) — calculates a string size in characters to encode a UTF-8 buffer as a UTF-16 string.

JavaScript supports UCS-2 strings with 16-bit characters, while node.js 0.11 supports full UTF-16 as a default string.

Property: internalSource

Starting 1.8.0 property source emulates the same property of RegExp, meaning that it can be used to create an identical RE2 or RegExp instance. Sometimes, for troubleshooting purposes, a user wants to inspect a RE2 translated source. It is available as a read-only property called internalSource.

Unicode warning level

RE2 engine always works in the Unicode mode. In most cases either there is no difference or the Unicode mode is actually preferred. But sometimes a user wants a tight control over their regular expressions. For those cases, there is a static string property RE2.unicodeWarningLevel.

Regular expressions in the Unicode mode work as usual. But if a regular expression lacks the Unicode flag, it is always added silently.

const x = /./;
x.flags; // ''
const y = new RE2(x);
y.flags; // 'u'

In the latter case RE2 can do following actions depending on RE2.unicodeWarningLevel:

  • 'nothing' (the default): no warnings or notifications of any kind, a regular expression will be created with 'u' flag.
  • 'warnOnce': warns exactly once the very first time, a regular expression will be created with 'u' flag.
    • Assigning this value resets an internal flag, so RE2 will warn once again.
  • 'warn': warns every time, a regular expression will be created with 'u' flag.
  • 'throw': throws a SyntaxError every time.
  • All other warning level values are silently ignored on asignment leaving the previous value unchanged.

Warnings and exceptions help to audit an application for stray non-Unicode regular expressions.

How to install

Installation:

npm install --save re2

While the project is known to work with other package managers, it is not guaranteed nor tested. For example, yarn is known to fail in some scenarios (see this Wiki article).

Precompiled artifacts

When installing re2 the install script attempts to download a prebuilt artifact for your system from the Github releases. The download location can be overridden by setting the RE2_DOWNLOAD_MIRROR environment variable as seen in the install script.

If all attempts to download the prebuilt artifact for your system fails the script attempts to built re2 locally on your machine using node-gyp.

How to use

It is used just like a RegExp object.

var RE2 = require("re2");

// with default flags
var re = new RE2("a(b*)");
var result = re.exec("abbc");
console.log(result[0]); // "abb"
console.log(result[1]); // "bb"

result = re.exec("aBbC");
console.log(result[0]); // "a"
console.log(result[1]); // ""

// with explicit flags
re = new RE2("a(b*)", "i");
result = re.exec("aBbC");
console.log(result[0]); // "aBb"
console.log(result[1]); // "Bb"

// from regular expression object
var regexp = new RegExp("a(b*)", "i");
re = new RE2(regexp);
result = re.exec("aBbC");
console.log(result[0]); // "aBb"
console.log(result[1]); // "Bb"

// from regular expression literal
re = new RE2(/a(b*)/i);
result = re.exec("aBbC");
console.log(result[0]); // "aBb"
console.log(result[1]); // "Bb"

// from another RE2 object
var rex = new RE2(re);
result = rex.exec("aBbC");
console.log(result[0]); // "aBb"
console.log(result[1]); // "Bb"

// shortcut
result = new RE2("ab*").exec("abba");

// factory
result = RE2("ab*").exec("abba");

Limitations (things RE2 does not support)

RE2 consciously avoids any regular expression features that require worst-case exponential time to evaluate. These features are essentially those that describe a Context-Free Language (CFL) rather than a Regular Expression, and are extensions to the traditional regular expression language because some people don't know when enough is enough.

The most noteworthy missing features are backreferences and lookahead assertions. If your application uses these features, you should continue to use RegExp. But since these features are fundamentally vulnerable to ReDoS, you should strongly consider replacing them.

RE2 will throw a SyntaxError if you try to declare a regular expression using these features. If you are evaluating an externally-provided regular expression, wrap your RE2 declarations in a try-catch block. It allows to use RegExp, when RE2 misses a feature:

var re = /(a)+(b)*/;
try {
  re = new RE2(re);
  // use RE2 as a drop-in replacement
} catch (e) {
  // suppress an error, and use
  // the original RegExp
}
var result = re.exec(sample);

In addition to these missing features, RE2 also behaves somewhat differently from the built-in regular expression engine in corner cases.

Backreferences

RE2 doesn't support backreferences, which are numbered references to previously matched groups, like so: \1, \2, and so on. Example of backrefrences:

/(cat|dog)\1/.test("catcat"); // true
/(cat|dog)\1/.test("dogdog"); // true
/(cat|dog)\1/.test("catdog"); // false
/(cat|dog)\1/.test("dogcat"); // false

Lookahead assertions

RE2 doesn't support lookahead assertions, which are ways to allow a matching dependent on subsequent contents.

/abc(?=def)/; // match abc only if it is followed by def
/abc(?!def)/; // match abc only if it is not followed by def

Mismatched behavior

RE2 and the built-in regex engines disagree a bit. Before you switch to RE2, verify that your regular expressions continue to work as expected. They should do so in the vast majority of cases.

Here is an example of a case where they may not:

var RE2  = require("../re2");

var pattern = '(?:(a)|(b)|(c))+';

var built_in = new RegExp(pattern);
var re2 = new RE2(pattern);

var input = 'abc';

var bi_res = built_in.exec(input);
var re2_res = re2.exec(input);

console.log('bi_res: ' + bi_res);    // prints: bi_res: abc,,,c
console.log('re2_res : ' + re2_res); // prints: re2_res : abc,a,b,c

Unicode

RE2 always works in the Unicode mode. See RE2.unicodeWarningLevel above for more details on how to control warnings about this feature.

Release history

  • 1.21.4 Fixed a regression reported by caroline-matsec, thx! Added pre-compilation targets for Alpine Linux on ARM. Updated deps.
  • 1.21.3 Fixed an empty string regression reported by Rhys Arkins, thx! Updated deps.
  • 1.21.2 Fixed another memory regression reported by matthewvalentine, thx! Updated deps. Added more tests and benchmarks.
  • 1.21.1 Fixed a memory regression reported by matthewvalentine, thx! Updated deps.
  • 1.21.0 Fixed the performance problem reported by matthewvalentine (thx!). The change improves performance for multiple use cases.
  • 1.20.12 Updated deps. Maintenance chores. Fixes for buffer-related bugs: exec() index (reported by matthewvalentine, thx) and match() index.
  • 1.20.11 Updated deps. Added support for Node 22 (thx, Elton Leong).
  • 1.20.10 Updated deps. Removed files the pack used for development (thx, Haruaki OTAKE). Added arm64 Linux prebilds (thx, Christopher M). Fixed non-npm corepack problem (thx, Steven).
  • 1.20.9 Updated deps. Added more absail-cpp files that manifested itself on NixOS. Thx, Laura Hausmann.
  • 1.20.8 Updated deps: install-artifact-from-github. A default HTTPS agent is used for fetching precompiled artifacts avoiding unnecessary long wait times.
  • 1.20.7 Added more absail-cpp files that manifested itself on ARM Alpine. Thx, Laura Hausmann.
  • 1.20.6 Updated deps, notably node-gyp.
  • 1.20.5 Updated deps, added Node 21 and retired Node 16 as pre-compilation targets.
  • 1.20.4 Updated deps. Fix: the 2nd argument of the constructor overrides flags. Thx, gost-serb.
  • 1.20.3 Fix: subsequent numbers are incorporated into group if they would form a legal group reference. Thx, Oleksii Vasyliev.
  • 1.20.2 Fix: added a missing C++ file, which caused a bug on Alpine Linux. Thx, rbitanga-manticore.
  • 1.20.1 Fix: files included in the npm package to build the C++ code.
  • 1.20.0 Updated RE2. New version uses abseil-cpp and required the adaptation work. Thx, Stefano Rivera.

The rest can be consulted in the project's wiki Release history.

License

BSD-3-Clause