moo

What is moo?

The moo npm package is a tokenizer for compilers and other tools that operate on code. It takes a string of text and breaks it down into tokens that can be easily parsed or analyzed. This is useful for creating programming languages, interpreters, compilers, and other syntax-aware utilities.

What are moo's main functionalities?

Tokenization

This code sample demonstrates how to define a simple lexer with the moo package. It includes rules for whitespace, comments, numbers, strings, parentheses, keywords, and newlines. The lexer is then used to tokenize a sample string.

{"lexer": "const moo = require('moo');\nlet lexer = moo.compile({\n  WS: /[ \t]+/,\n  comment: /\/\/.*?$/,\n  number: /\d+/,\n  string: /\"(\\\\.|[^\\\\\"])*\"/,\n  lparen: '('\n  rparen: ')',\n  keyword: ['while', 'if', 'else', 'moo'],\n  NL: { match: /\n/, lineBreaks: true },\n});\nlexer.reset('if (42) moo');\nconsole.log(lexer.next()); // -> { type: 'keyword', value: 'if' }"}

Custom Error Handling

This code sample shows how to handle errors in tokenization with moo. An 'error' token type is defined using moo.error, which is used to throw an exception when the lexer encounters an invalid syntax.

{"lexerWithErrorHandling": "const moo = require('moo');\nlet lexer = moo.compile({\n  WS: /[ \t]+/,\n  comment: /\/\/.*?$/,\n  number: /\d+/,\n  string: /\"(\\\\.|[^\\\\\"])*\"/,\n  error: moo.error,\n});\nlexer.reset('invalid input');\ntry {\n  while (true) {\n    let token = lexer.next();\n    if (!token) break;\n    if (token.type === 'error') throw Error('Invalid syntax');\n  }\n} catch (e) {\n  console.error(e.message);\n}"}

Other packages similar to moo

chevrotain

Chevrotain is a JavaScript parser building toolkit which is similar to moo but more powerful and complex. It provides not only tokenization but also parsing capabilities, allowing users to define both the lexer and the parser for their language. Chevrotain is typically used when you need a full parser rather than just a tokenizer.

nearley

Nearley is a simple, fast, and powerful parsing toolkit for JavaScript. It is similar to moo in that it can be used to tokenize input strings, but it also includes a parser that is generated from a grammar specification. Nearley is often used for more complex parsing tasks where a grammar-based approach is beneficial.

pegjs

PEG.js is a parser generator for JavaScript based on the parsing expression grammar formalism. It generates parsers with excellent performance and error reporting from a grammar. While moo focuses on tokenization, PEG.js provides a full parsing solution with a focus on creating parsers from a domain-specific language (DSL).

README

Moo!

Moo is a highly-optimised tokenizer/lexer generator. Use it to tokenize your strings, before parsing 'em with a parser like nearley or whatever else you're into.

• Fast
• Convenient
• uses Regular Expressions
• tracks Line Numbers
• handles Keywords
• supports States
• custom Errors
• is even Iterable
• has no dependencies
• 4KB minified + gzipped
• Moo!

Is it fast?

Yup! Flying-cows-and-singed-steak fast.

Moo is the fastest JS tokenizer around. It's ~2–10x faster than most other tokenizers; it's a couple orders of magnitude faster than some of the slower ones.

Define your tokens using regular expressions. Moo will compile 'em down to a single RegExp for performance. It uses the new ES6 sticky flag where possible to make things faster; otherwise it falls back to an almost-as-efficient workaround. (For more than you ever wanted to know about this, read adventures in the land of substrings and RegExps.)

You might be able to go faster still by writing your lexer by hand rather than using RegExps, but that's icky.

Oh, and it avoids parsing RegExps by itself. Because that would be horrible.

Usage

First, you need to do the needful: $ npm install moo, or whatever will ship this code to your computer. Alternatively, grab the moo.js file by itself and slap it into your web page via a <script> tag; moo is completely standalone.

Then you can start roasting your very own lexer/tokenizer:

    const moo = require('moo')

    let lexer = moo.compile({
      WS:      /[ \t]+/,
      comment: /\/\/.*?$/,
      number:  /0|[1-9][0-9]*/,
      string:  /"(?:\\["\\]|[^\n"\\])*"/,
      lparen:  '(',
      rparen:  ')',
      keyword: ['while', 'if', 'else', 'moo', 'cows'],
      NL:      { match: /\n/, lineBreaks: true },
    })

And now throw some text at it:

    lexer.reset('while (10) cows\nmoo')
    lexer.next() // -> { type: 'keyword', value: 'while' }
    lexer.next() // -> { type: 'WS', value: ' ' }
    lexer.next() // -> { type: 'lparen', value: '(' }
    lexer.next() // -> { type: 'number', value: '10' }
    // ...

When you reach the end of Moo's internal buffer, next() will return undefined. You can always reset() it and feed it more data when that happens.

On Regular Expressions

RegExps are nifty for making tokenizers, but they can be a bit of a pain. Here are some things to be aware of:

• You often want to use non-greedy quantifiers: e.g. *? instead of *. Otherwise your tokens will be longer than you expect:

  let lexer = moo.compile({
    string: /".*"/,   // greedy quantifier *
    // ...
  })
  
  lexer.reset('"foo" "bar"')
  lexer.next() // -> { type: 'string', value: 'foo" "bar' }
  

  Better:

  let lexer = moo.compile({
    string: /".*?"/,   // non-greedy quantifier *?
    // ...
  })
  
  lexer.reset('"foo" "bar"')
  lexer.next() // -> { type: 'string', value: 'foo' }
  lexer.next() // -> { type: 'space', value: ' ' }
  lexer.next() // -> { type: 'string', value: 'bar' }
  

• The order of your rules matters. Earlier ones will take precedence.

  moo.compile({
      identifier:  /[a-z0-9]+/,
      number:  /[0-9]+/,
  }).reset('42').next() // -> { type: 'identifier', value: '42' }
  
  moo.compile({
      number:  /[0-9]+/,
      identifier:  /[a-z0-9]+/,
  }).reset('42').next() // -> { type: 'number', value: '42' }
  

• Moo uses multiline RegExps. This has a few quirks: for example, the dot /./ doesn't include newlines. Use [^] instead if you want to match newlines too.

• Since an excluding character ranges like /[^ ]/ (which matches anything but a space) will include newlines, you have to be careful not to include them by accident! In particular, the whitespace metacharacter \s includes newlines.

Line Numbers

Moo tracks detailed information about the input for you.

It will track line numbers, as long as you apply the lineBreaks: true option to any rules which might contain newlines. Moo will try to warn you if you forget to do this.

Note that this is false by default, for performance reasons: counting the number of lines in a matched token has a small cost. For optimal performance, only match newlines inside a dedicated token:

    newline: {match: '\n', lineBreaks: true},

Token Info

Token objects (returned from next()) have the following attributes:

• type: the name of the group, as passed to compile.
• text: the string that was matched.
• value: the string that was matched, transformed by your value function (if any).
• offset: the number of bytes from the start of the buffer where the match starts.
• lineBreaks: the number of line breaks found in the match. (Always zero if this rule has lineBreaks: false.)
• line: the line number of the beginning of the match, starting from 1.
• col: the column where the match begins, starting from 1.

Value vs. Text

The value is the same as the text, unless you provide a value transform.

const moo = require('moo')

const lexer = moo.compile({
  ws: /[ \t]+/,
  string: {match: /"(?:\\["\\]|[^\n"\\])*"/, value: s => s.slice(1, -1)},
})

lexer.reset('"test"')
lexer.next() /* { value: 'test', text: '"test"', ... } */

Reset

Calling reset() on your lexer will empty its internal buffer, and set the line, column, and offset counts back to their initial value.

If you don't want this, you can save() the state, and later pass it as the second argument to reset() to explicitly control the internal state of the lexer.

    lexer.reset('some line\n')
    let info = lexer.save() // -> { line: 10 }
    lexer.next() // -> { line: 10 }
    lexer.next() // -> { line: 11 }
    // ...
    lexer.reset('a different line\n', info)
    lexer.next() // -> { line: 10 }

Keywords

Moo makes it convenient to define literals.

    moo.compile({
      lparen:  '(',
      rparen:  ')',
      keyword: ['while', 'if', 'else', 'moo', 'cows'],
    })

It'll automatically compile them into regular expressions, escaping them where necessary.

Keywords should be written using the keywords transform.

    moo.compile({
      IDEN: {match: /[a-zA-Z]+/, type: moo.keywords({
        KW: ['while', 'if', 'else', 'moo', 'cows'],
      })},
      SPACE: {match: /\s+/, lineBreaks: true},
    })

Why?

You need to do this to ensure the longest match principle applies, even in edge cases.

Imagine trying to parse the input className with the following rules:

    keyword: ['class'],
    identifier: /[a-zA-Z]+/,

You'll get two tokens — ['class', 'Name'] -- which is not what you want! If you swap the order of the rules, you'll fix this example; but now you'll lex class wrong (as an identifier).

The keywords helper checks matches against the list of keywords; if any of them match, it uses the type 'keyword' instead of 'identifier' (for this example).

Keyword Types

Keywords can also have individual types.

    let lexer = moo.compile({
      name: {match: /[a-zA-Z]+/, type: moo.keywords({
        'kw-class': 'class',
        'kw-def': 'def',
        'kw-if': 'if',
      })},
      // ...
    })
    lexer.reset('def foo')
    lexer.next() // -> { type: 'kw-def', value: 'def' }
    lexer.next() // space
    lexer.next() // -> { type: 'name', value: 'foo' }

You can use Object.fromEntries to easily construct keyword objects:

Object.fromEntries(['class', 'def', 'if'].map(k => ['kw-' + k, k]))

States

Moo allows you to define multiple lexer states. Each state defines its own separate set of token rules. Your lexer will start off in the first state given to moo.states({}).

Rules can be annotated with next, push, and pop, to change the current state after that token is matched. A "stack" of past states is kept, which is used by push and pop.

• next: 'bar' moves to the state named bar. (The stack is not changed.)
• push: 'bar' moves to the state named bar, and pushes the old state onto the stack.
• pop: 1 removes one state from the top of the stack, and moves to that state. (Only 1 is supported.)

Only rules from the current state can be matched. You need to copy your rule into all the states you want it to be matched in.

For example, to tokenize JS-style string interpolation such as a${{c: d}}e, you might use:

    let lexer = moo.states({
      main: {
        strstart: {match: '`', push: 'lit'},
        ident:    /\w+/,
        lbrace:   {match: '{', push: 'main'},
        rbrace:   {match: '}', pop: 1},
        colon:    ':',
        space:    {match: /\s+/, lineBreaks: true},
      },
      lit: {
        interp:   {match: '${', push: 'main'},
        escape:   /\\./,
        strend:   {match: '`', pop: 1},
        const:    {match: /(?:[^$`]|\$(?!\{))+/, lineBreaks: true},
      },
    })
    // <= `a${{c: d}}e`
    // => strstart const interp lbrace ident colon space ident rbrace rbrace const strend

The rbrace rule is annotated with pop, so it moves from the main state into either lit or main, depending on the stack.

Errors

If none of your rules match, Moo will throw an Error; since it doesn't know what else to do.

If you prefer, you can have moo return an error token instead of throwing an exception. The error token will contain the whole of the rest of the buffer.

    moo.compile({
      // ...
      myError: moo.error,
    })

    moo.reset('invalid')
    moo.next() // -> { type: 'myError', value: 'invalid', text: 'invalid', offset: 0, lineBreaks: 0, line: 1, col: 1 }
    moo.next() // -> undefined

You can have a token type that both matches tokens and contains error values.

    moo.compile({
      // ...
      myError: {match: /[\$?`]/, error: true},
    })

Formatting errors

If you want to throw an error from your parser, you might find formatError helpful. Call it with the offending token:

throw new Error(lexer.formatError(token, "invalid syntax"))

It returns a string with a pretty error message.

Error: invalid syntax at line 2 col 15:

  totally valid `syntax`
                ^

Iteration

Iterators: we got 'em.

    for (let here of lexer) {
      // here = { type: 'number', value: '123', ... }
    }

Create an array of tokens.

    let tokens = Array.from(lexer);

Use itt's iteration tools with Moo.

    for (let [here, next] of itt(lexer).lookahead()) { // pass a number if you need more tokens
      // enjoy!
    }

Transform

Moo doesn't allow capturing groups, but you can supply a transform function, value(), which will be called on the value before storing it in the Token object.

    moo.compile({
      STRING: [
        {match: /"""[^]*?"""/, lineBreaks: true, value: x => x.slice(3, -3)},
        {match: /"(?:\\["\\rn]|[^"\\])*?"/, lineBreaks: true, value: x => x.slice(1, -1)},
        {match: /'(?:\\['\\rn]|[^'\\])*?'/, lineBreaks: true, value: x => x.slice(1, -1)},
      ],
      // ...
    })

Contributing

Do check the FAQ.

Before submitting an issue, remember...