Skip to content

lydell/js-tokens

Repository files navigation

js-tokens

The tiny, regex powered, lenient, almost spec-compliant JavaScript tokenizer that never fails.

const jsTokens = require("js-tokens");

const jsString = 'JSON.stringify({k:3.14**2}, null /*replacer*/, "\\t")';

Array.from(jsTokens(jsString), (token) => token.value).join("|");
// JSON|.|stringify|(|{|k|:|3.14|**|2|}|,| |null| |/*replacer*/|,| |"\t"|)

Installation

npm install js-tokens

import jsTokens from "js-tokens";
// or:
const jsTokens = require("js-tokens");

Usage

jsTokens(string, options?)
Option Type Default Description
jsx boolean false Enable JSX support.

This package exports a generator function, jsTokens, that turns a string of JavaScript code into token objects.

For the empty string, the function yields nothing (which can be turned into an empty list). For any other input, the function always yields something, even for invalid JavaScript, and never throws. Concatenating the token values reproduces the input.

The package is very close to being fully spec compliant (it passes all but 3 of test262-parser-tests), but has taken a couple of shortcuts. See the following sections for limitations of some tokens.

// Loop over tokens:
for (const token of jsTokens("hello, !world")) {
  console.log(token);
}

// Get all tokens as an array:
const tokens = Array.from(jsTokens("hello, !world"));

Tokens

Spec: ECMAScript Language: Lexical Grammar + Additional Syntax

export default function jsTokens(input: string): Iterable<Token>;

type Token =
  | { type: "StringLiteral"; value: string; closed: boolean }
  | { type: "NoSubstitutionTemplate"; value: string; closed: boolean }
  | { type: "TemplateHead"; value: string }
  | { type: "TemplateMiddle"; value: string }
  | { type: "TemplateTail"; value: string; closed: boolean }
  | { type: "RegularExpressionLiteral"; value: string; closed: boolean }
  | { type: "MultiLineComment"; value: string; closed: boolean }
  | { type: "SingleLineComment"; value: string }
  | { type: "HashbangComment"; value: string }
  | { type: "IdentifierName"; value: string }
  | { type: "PrivateIdentifier"; value: string }
  | { type: "NumericLiteral"; value: string }
  | { type: "Punctuator"; value: string }
  | { type: "WhiteSpace"; value: string }
  | { type: "LineTerminatorSequence"; value: string }
  | { type: "Invalid"; value: string };

StringLiteral

Spec: StringLiteral

If the ending " or ' is missing, the token has closed: false. JavaScript strings cannot contain (unescaped) newlines, so unclosed strings simply end at the end of the line.

Escape sequences are supported, but may be invalid. For example, "\u" is matched as a StringLiteral even though it contains an invalid escape.

Examples:

"string"
'string'
""
''
"\""
'\''
"valid: \u00a0, invalid: \u"
'valid: \u00a0, invalid: \u'
"multi-\
line"
'multi-\
line'
" unclosed
' unclosed

NoSubstitutionTemplate / TemplateHead / TemplateMiddle / TemplateTail

Spec: NoSubstitutionTemplate / TemplateHead / TemplateMiddle / TemplateTail

A template without interpolations is matched as is. For, example:

  • `abc`: NoSubstitutionTemplate
  • `abc: NoSubstitutionTemplate with closed: false

A template with interpolations is matched as many tokens. For example, `head${1}middle${2}tail` is matched as follows (apart from the two NumericLiterals):

  • `head${: TemplateHead
  • }middle${: TemplateMiddle
  • }tail`: TemplateTail

TemplateMiddle is optional, and TemplateTail can be unclosed. For example, `head${1}tail (note the missing ending `):

  • `head${: TemplateHead
  • }tail: TemplateTail with closed: false

Templates can contain unescaped newlines, so unclosed templates go on to the end of input.

Just like for StringLiteral, templates can also contain invalid escapes. `\u` is matched as a NoSubstitutionTemplate even though it contains an invalid escape. Also note that in tagged templates, invalid escapes are not syntax errors: x`\u` is syntactically valid JavaScript.

RegularExpressionLiteral

Spec: RegularExpressionLiteral

Regex literals may contain invalid regex syntax. They are still matched as regex literals. The specification even says:

The productions below [...] are used by the input element scanner to find the end of the regular expression literal. The source text [...] are subsequently parsed again using the more stringent ECMAScript Regular Expression grammar.

If the ending / is missing, the token has closed: false. JavaScript regex literals cannot contain newlines (not even escaped ones), so unclosed regex literals simply end at the end of the line.

According to the specification, the flags of regular expressions are IdentifierParts (unknown and repeated regex flags become errors at a later stage).

Differentiating between regex and division in JavaScript is really tricky. js-tokens looks at the previous token to tell them apart. As long as the previous tokens are valid, it should do the right thing. For invalid code, js-tokens might be confused and start matching division as regex or vice versa.

Examples:

/a/
/a/gimsuy
/a/Inva1id
/+/
/[/]\//
/[\p{Decimal_Number}--[0-9]]/v

MultiLineComment

Spec: MultiLineComment

If the ending */ is missing, the token has closed: false. Unclosed multi-line comments go on to the end of the input.

Examples:

/* comment */
/* console.log(
    "commented", out + code);
    */
/**/
/* unclosed

SingleLineComment

Spec: SingleLineComment

Examples:

// comment
// console.log("commented", out + code);
//

HashbangComment

Spec: HashbangComment

Note that a HashbangComment can only occur at the very start of the string that is being tokenized. Anywhere else you will likely get an Invalid token # followed by a Punctuator token !.

Examples:

#!/usr/bin/env node
#! console.log("commented", out + code);
#!

IdentifierName

Spec: IdentifierName

Keywords, reserved words, null, true, false, variable names and property names.

Examples:

if
for
var
instanceof
package
null
true
false
Infinity
undefined
NaN
$variab1e_name
π

ಠ_ಠ
\u006C\u006F\u006C\u0077\u0061\u0074

PrivateIdentifier

Spec: PrivateIdentifier

Any IdentifierName preceded by a #.

Examples:

#if
#for
#var
#instanceof
#package
#null
#true
#false
#Infinity
#undefined
#NaN
#$variab1e_name
#π
#
#ಠ_ಠ
#\u006C\u006F\u006C\u0077\u0061\u0074

NumericLiteral

Spec: NumericLiteral

Examples:

0
1.5
1
1_000
12e9
0.123e-32
0xDead_beef
0b110
12n
07
09.5

Punctuator

Spec: Punctuator + DivPunctuator + RightBracePunctuator

All possible values:

--  ++
.   ?.
<   <=   >   >=
!=  !==  ==  ===
   +   -   %   &   |   ^   /   *   **   <<   >>   >>>   &&   ||   ??
=  +=  -=  %=  &=  |=  ^=  /=  *=  **=  <<=  >>=  >>>=  &&=  ||=  ??=
(  )  [  ]  {  }
!  ?  :  ;  ,  ~  ...  =>

WhiteSpace

Spec: WhiteSpace

Unlike the specification, multiple whitespace characters in a row are matched as one token, not one token per character.

LineTerminatorSequence

Spec: LineTerminatorSequence

CR, LF and CRLF, plus \u2028 and \u2029.

Invalid

Spec: n/a

Single code points not matched in another token.

Examples:

#
@
💩

JSX Tokens

Spec: JSX Specification

export default function jsTokens(
  input: string,
  options: { jsx: true },
): Iterable<Token | JSXToken>;

export declare type JSXToken =
  | { type: "JSXString"; value: string; closed: boolean }
  | { type: "JSXText"; value: string }
  | { type: "JSXIdentifier"; value: string }
  | { type: "JSXPunctuator"; value: string }
  | { type: "JSXInvalid"; value: string };
  • The tokenizer switches between outputting runs of Token and runs of JSXToken.
  • Runs of JSXToken can also contain WhiteSpace, LineTerminatorSequence, MultiLineComment and SingleLineComment.

JSXString

Spec: " JSXDoubleStringCharacters " + ' JSXSingleStringCharacters '

If the ending " or ' is missing, the token has closed: false. JSX strings can contain unescaped newlines, so unclosed JSX strings go on to the end of input.

Note that JSX don’t support escape sequences as part of the token grammar. A " or ' always closes the string, even with a backslash before.

Examples:

"string"
'string'
""
''
"\"
'\'
"multi-
line"
'multi-
line'
" unclosed
' unclosed

JSXText

Spec: JSXText

Anything but <, >, { and }.

JSXIdentifier

Spec: JSXIdentifier

Examples:

div
class
xml
x-element
x------
$htm1_element
ಠ_ಠ

JSXPunctuator

Spec: n/a

All possible values:

<
>
/
.
:
=
{
}

JSXInvalid

Spec: n/a

Single code points not matched in another token.

Examples in JSX tags:

1
`
+
,
#
@
💩

All possible values in JSX children:

>
}

Compatibility

ECMAScript

The intention is to always support the latest ECMAScript version whose feature set has been finalized.

Currently, ECMAScript 2024 is supported.

Annex B and C (strict mode)

Section B: Additional ECMAScript Features for Web Browsers of the spec is optional if the ECMAScript host is not a web browser, and specifies some additional syntax. Section C: The Strict Mode of ECMAScript disallows certain syntax in Strict Mode.

  • Numeric literals: js-tokens supports legacy octal and octal like numeric literals, regardless of Strict Mode.
  • String literals: js-tokens supports legacy octal escapes, since it allows any invalid escapes.
  • HTML-like comments: Not supported. js-tokens prefers treating 5<!--x as 5 < !(--x) rather than as 5 //x.
  • Regular expression patterns: js-tokens doesn’t care what’s between the starting / and ending /, so this is supported.

TypeScript

Supporting TypeScript is not an explicit goal, but js-tokens and Babel both tokenize this TypeScript fixture and this TSX fixture the same way, with one edge case:

type A = Array<Array<string>>
type B = Array<Array<Array<string>>>

Both lines above should end with a couple of > tokens, but js-tokens instead matches the >> and >>> operators.

JSX

JSX is supported: jsTokens("<p>Hello, world!</p>", { jsx: true }).

JavaScript runtimes

js-tokens should work in any JavaScript runtime that supports Unicode property escapes.

Known errors

Here are a couple of tricky cases:

// Case 1:
switch (x) {
  case x: {}/a/g;
  case x: {}<div>x</div>/g;
}

// Case 2:
label: {}/a/g;
label: {}<div>x</div>/g;

// Case 3:
(function f() {}/a/g);
(function f() {}<div>x</div>/g);

This is what they mean:

// Case 1:
switch (x) {
  case x:
    {
    }
    /a/g;
  case x:
    {
    }
    <div>x</div> / g;
}

// Case 2:
label: {
}
/a/g;
label: {
}
<div>x</div> / g;

// Case 3:
(function f() {}) / a / g;
(function f() {}) < div > x < /div>/g;

But js-tokens thinks they mean:

// Case 1:
switch (x) {
  case x:
    ({}) / a / g;
  case x:
    ({}) < div > x < /div>/g;
}

// Case 2:
label: ({}) / a / g;
label: ({}) < div > x < /div>/g;

// Case 3:
function f() {}
/a/g;
function f() {}
<div>x</div> / g;

In other words, js-tokens:

  • Mis-identifies regex as division and JSX as comparison in case 1 and 2.
  • Mis-identifies division as regex and comparison as JSX in case 3.

This happens because js-tokens looks at the previous token when deciding between regex and division or JSX and comparison. In these cases, the previous token is }, which either means “end of block” (→ regex/JSX) or “end of object literal” (→ division/comparison). How does js-tokens determine if the } belongs to a block or an object literal? By looking at the token before the matching {.

In case 1 and 2, that’s a :. A : usually means that we have an object literal or ternary:

let some = weird ? { value: {}/a/g } : {}/a/g;

But : is also used for case and labeled statements.

One idea is to look for case before the : as an exception to the rule, but it’s not so easy:

switch (x) {
  case weird ? true : {}/a/g: {}/a/g
}

The first {}/a/g is a division, while the second {}/a/g is an empty block followed by a regex. Both are preceded by a colon with a case on the same line, and it does not seem like you can distinguish between the two without implementing a parser.

Labeled statements are similarly difficult, since they are so similar to object literals:

{
  label: {}/a/g
}

({
  key: {}/a/g
})

Finally, case 3 ((function f() {}/a/g);) is also difficult, because a ) before a { means that the { is part of a block, and blocks are usually statements:

if (x) {
}
/a/g;

function f() {}
/a/g;

But function expressions are of course not statements. It’s difficult to tell an function expression from a function statement without parsing.

Luckily, none of these edge cases are likely to occur in real code.

Known failures

js-tokens advertises that it “never fails”. Tell you what, it can fail on extreme inputs. The regex engine of the runtime can eventually give up. js-tokens has worked around it to some extent by changing its regexes to be easier on the regex engine. To solve completely, js-tokens would have to stop using regex, but then it wouldn’t be tiny anymore which is the whole point. Luckily, only extreme inputs can fail, hopefully ones you’ll never encounter.

For example, if you try to parse the string literal "\n\n\n" but with 10 million \n instead of just 3, the regex engine gives up with RangeError: Maximum call stack size exceeded (or similar). Try it out:

Array.from(require("js-tokens")(`"${"\\n".repeat(1e7)}"`));

(Yes, that is the regex engine of the runtime giving up. js-tokens has no recursive functions.)

However, if you repeat a instead of \n 10 million times ("aaaaaa…"), it works:

Array.from(require("js-tokens")(`"${"a".repeat(1e7)}"`));

That’s good, because it’s much more common to have lots of non-escapes in a row in a big string literal, than having mostly escapes. (Obfuscated code might have only escapes though.)

Safari warning

I’ve seen Safari give up instead of throwing an error.

In Safari, Chrome, Firefox and Node.js the following code successfully results in a match:

/(#)(?:a|b)+/.exec("#" + "a".repeat(1e5));

But for the following code (with 1e7 instead of 1e5), the runtimes differ:

/(#)(?:a|b)+/.exec("#" + "a".repeat(1e7));
  • Chrome, Firefox and Node.js all throw RangeError: Maximum call stack size exceeded (or similar).
  • Safari returns null (at the time of writing), silently giving up on matching the regex. It’s kind of lying that the regex did not match, while in reality it would given enough computing resources.

This means that in Safari, js-tokens might not fail but instead give you unexpected tokens.

Performance

With @babel/parser for comparison. Node.js 21.6.1 on a MacBook Pro M1 (Sonoma).

Lines of code Size [email protected] @babel/[email protected]
~100 ~4.0 KiB ~2 ms ~10 ms
~1 000 ~39 KiB ~5 ms ~27 ms
~10 000 ~353 KiB ~44 ms ~108 ms
~100 000 ~5.1 MiB ~333 ms ~2.0 s
~2 400 000 ~138 MiB ~7 s ~4 m 9 s (*)

(*) Required increasing the Node.js the memory limit (I set it to 8 GiB).

See benchmark.js if you want to run benchmarks yourself.