Hamms and wooorm Remove support for whitespace in reference
Closes GH-355.

Reviewed-by: Titus Wormer <tituswormer@gmail.com>
Latest commit ddc948a Sep 23, 2018


remark-parse Build Status Coverage Status Chat

Parser for unified. Parses markdown to an MDAST syntax tree. Used in the remark processor. Can be extended to change how markdown is parsed.



npm install remark-parse


var unified = require('unified');
var createStream = require('unified-stream');
var markdown = require('remark-parse');
var html = require('remark-html');

var processor = unified()
  .use(markdown, {commonmark: true})


Table of Contents


processor.use(parse[, options])

Configure the processor to read markdown as input and process an MDAST syntax tree.


Options are passed directly, or passed later through processor.data().

hello ~~hi~~ world

GFM mode (boolean, default: true) turns on:

This is a paragraph
    and this is also part of the preceding paragraph.

CommonMark mode (boolean, default: false) allows:

  • Empty lines to split blockquotes
  • Parentheses (( and )) around for link and image titles
  • Any escaped ASCII-punctuation character
  • Closing parenthesis ()) as an ordered list marker
  • URL definitions (and footnotes, when enabled) in blockquotes

CommonMark mode disallows:

  • Code directly following a paragraph
  • ATX-headings (# Hash headings) without spacing after opening hashes or and before closing hashes
  • Setext headings (Underline headings\n---) when following a paragraph
  • Newlines in link and image titles
  • White space in link and image URLs in auto-links (links in brackets, < and >)
  • Lazy blockquote continuation, lines not preceded by a closing angle bracket (>), for lists, code, and thematicBreak
Something something[^or something?].

And something else[^1].

[^1]: This reference footnote contains a paragraph...

    * ...and a list

Footnotes mode (boolean, default: false) enables reference footnotes and inline footnotes. Both are wrapped in square brackets and preceded by a caret (^), and can be referenced from inside other footnotes.


Blocks (Array.<string>, default: list of block HTML elements) exposes let’s users define block-level HTML elements.

Check out some_file_name.txt

Pedantic mode (boolean, default: false) turns on:

  • Emphasis (_alpha_) and importance (__bravo__) with underscores in words
  • Unordered lists with different markers (*, -, +)
  • If commonmark is also turned on, ordered lists with different markers (., ))
  • And pedantic mode removes less spaces in list-items (at most four, instead of the whole indent)


Access to the parser, if you need it.

Extending the Parser

Most often, using transformers to manipulate a syntax tree produces the desired output. Sometimes, mainly when introducing new syntactic entities with a certain level of precedence, interfacing with the parser is necessary.

If the remark-parse plugin is used, it adds a Parser constructor to the processor. Other plugins can add tokenizers to the parser’s prototype to change how markdown is parsed.

The below plugin adds a tokenizer for at-mentions.

module.exports = mentions;

function mentions() {
  var Parser = this.Parser;
  var tokenizers = Parser.prototype.inlineTokenizers;
  var methods = Parser.prototype.inlineMethods;

  /* Add an inline tokenizer (defined in the following example). */
  tokenizers.mention = tokenizeMention;

  /* Run it just before `text`. */
  methods.splice(methods.indexOf('text'), 0, 'mention');


An object mapping tokenizer names to tokenizers. These tokenizers (for example: fencedCode, table, and paragraph) eat from the start of a value to a line ending.

See #blockMethods below for a list of methods that are included by default.


Array of blockTokenizers names (string) specifying the order in which they run.

  • newline
  • indentedCode
  • fencedCode
  • blockquote
  • atxHeading
  • thematicBreak
  • list
  • setextHeading
  • html
  • footnote
  • definition
  • table
  • paragraph


An object mapping tokenizer names to tokenizers. These tokenizers (for example: url, reference, and emphasis) eat from the start of a value. To increase performance, they depend on locators.

See #inlineMethods below for a list of methods that are included by default.


Array of inlineTokenizers names (string) specifying the order in which they run.

  • escape
  • autoLink
  • url
  • html
  • link
  • reference
  • strong
  • emphasis
  • deletion
  • code
  • break
  • text

function tokenizer(eat, value, silent)

tokenizeMention.notInLink = true;
tokenizeMention.locator = locateMention;

function tokenizeMention(eat, value, silent) {
  var match = /^@(\w+)/.exec(value);

  if (match) {
    if (silent) {
      return true;

    return eat(match[0])({
      type: 'link',
      url: 'https://social-network/' + match[1],
      children: [{type: 'text', value: match[0]}]

The parser knows two types of tokenizers: block level and inline level. Block level tokenizers are the same as inline level tokenizers, with the exception that the latter must have a locator.

Tokenizers test whether a document starts with a certain syntactic entity. In silent mode, they return whether that test passes. In normal mode, they consume that token, a process which is called “eating”. Locators enable tokenizers to function faster by providing information on where the next entity may occur.

  • Node? = tokenizer(eat, value)
  • boolean? = tokenizer(eat, value, silent)
  • eat (Function) — Eat, when applicable, an entity
  • value (string) — Value which may start an entity
  • silent (boolean, optional) — Whether to detect or consume
  • locator (Function) — Required for inline tokenizers
  • onlyAtStart (boolean) — Whether nodes can only be found at the beginning of the document
  • notInBlock (boolean) — Whether nodes cannot be in blockquotes, lists, or footnote definitions
  • notInList (boolean) — Whether nodes cannot be in lists
  • notInLink (boolean) — Whether nodes cannot be in links
  • In silent mode, whether a node can be found at the start of value
  • In normal mode, a node if it can be found at the start of value

tokenizer.locator(value, fromIndex)

function locateMention(value, fromIndex) {
  return value.indexOf('@', fromIndex);

Locators are required for inline tokenization to keep the process performant. Locators enable inline tokenizers to function faster by providing information on the where the next entity occurs. Locators may be wrong, it’s OK if there actually isn’t a node to be found at the index they return, but they must skip any nodes.

  • value (string) — Value which may contain an entity
  • fromIndex (number) — Position to start searching at

Index at which an entity may start, and -1 otherwise.


var add = eat('foo');

Eat subvalue, which is a string at the start of the tokenized value (it’s tracked to ensure the correct value is eaten).

  • subvalue (string) - Value to eat.


add(node[, parent])

var add = eat('foo');
add({type: 'text', value: 'foo'});

Add positional information to node and add it to parent.

  • node (Node) - Node to patch position on and insert
  • parent (Node, optional) - Place to add node to in the syntax tree. Defaults to the currently processed node

The given node.


Get the positional information which would be patched on node by add.



add.reset(node[, parent])

add, but resets the internal location. Useful for example in lists, where the same content is first eaten for a list, and later for list items

  • node (Node) - Node to patch position on and insert
  • parent (Node, optional) - Place to add node to in the syntax tree. Defaults to the currently processed node

The given node.

Turning off a tokenizer

In rare situations, you may want to turn off a tokenizer to avoid parsing that syntactic feature. This can be done by deleting the tokenizer from your Parser’s blockTokenizers (or blockMethods) or inlineTokenizers (or inlineMethods).

The following example turns off indented code blocks:

delete remarkParse.Parser.prototype.blockTokenizers.indentedCode;


MIT © Titus Wormer