@khanacademy/simple-markdown

@khanacademy/simple-markdown

simple-markdown is a markdown-like parser designed for simplicity and extensibility.

Philosophy

Most markdown-like parsers aim for speed or edge case handling. simple-markdown aims for extensibility and simplicity.

What does this mean? Many websites using markdown-like languages have custom extensions, such as @mentions or issue number linking. Unfortunately, most markdown-like parsers don't allow extension without forking, and can be difficult to modify even when forked. simple-markdown is designed to allow simple addition of custom extensions without needing to be forked.

At Khan Academy, we use simple-markdown to format over half of our math exercises, because we need [markdown extensions][PerseusMarkdown] for math text and interactive widgets.

simple-markdown is MIT licensed.

Getting started

First, let's parse and output some generic markdown using simple-markdown.

If you want to run these examples in node, you should run pnpm install in the simple-markdown folder or pnpm add simple-markdown in your project's folder. Then you can acquire the SimpleMarkdown variable with:

import SimpleMarkdown from '@khanacademy/simple-markdown'

Then let's get a basic markdown parser and outputter. SimpleMarkdown provides default parsers/outputters for generic markdown:

    var mdParse = SimpleMarkdown.defaultBlockParse;
    var mdOutput = SimpleMarkdown.defaultOutput;

mdParse can give us a syntax tree:

    var syntaxTree = mdParse("Here is a paragraph and an *em tag*.");

Let's inspect our syntax tree:

    // pretty-print this with 4-space indentation:
    console.log(JSON.stringify(syntaxTree, null, 4));
    => [
        {
            "content": [
                {
                    "content": "Here is a paragraph and an ",
                    "type": "text"
                },
                {
                    "content": [
                        {
                            "content": "em tag",
                            "type": "text"
                        }
                    ],
                    "type": "em"
                },
                {
                    "content": ".",
                    "type": "text"
                }
            ],
            "type": "paragraph"
        }
    ]

Then to turn that into an array of React elements, we can call mdOutput:

    mdOutput(syntaxTree)
    => [ { type: 'div',
        key: null,
        ref: null,
        _owner: null,
        _context: {},
        _store: { validated: false, props: [Object] } } ]

Adding a simple extension

Let's add an underline extension! To do this, we'll need to create a new rule and then make a new parser/outputter. The next section will explain how all of these steps work in greater detail. (To follow along with these examples, you'll also need underscore.)

First, we create a new rule. We'll look for double underscores surrounding text.

We'll put underlines right before ems, so that __ will be parsed before _ for emphasis/italics.

A regex to capture this would look something like /^__([\s\S]+?)__(?!_)/. This matches __, followed by any content until it finds another __ not followed by a third _.

    var underlineRule = {
        // Specify the order in which this rule is to be run
        order: SimpleMarkdown.defaultRules.em.order - 0.5,

        // First we check whether a string matches
        match: function(source) {
            return /^__([\s\S]+?)__(?!_)/.exec(source);
        },

        // Then parse this string into a syntax node
        parse: function(capture, parse, state) {
            return {
                content: parse(capture[1], state)
            };
        },

        // Finally transform this syntax node into a
        // React element
        react: function(node, output) {
            return React.DOM.u(null, output(node.content));
        },

        // Or an html element:
        // (Note: you may only need to make one of `react:` or
        // `html:`, as long as you never ask for an outputter
        // for the other type.)
        html: function(node, output) {
            return '<u>' + output(node.content) + '</u>';
        },
    };

Then, we need to add this rule to the other rules:

    var rules = _.extend({}, SimpleMarkdown.defaultRules, {
        underline: underlineRule
    });

Finally, we need to build our parser and outputters:

    var rawBuiltParser = SimpleMarkdown.parserFor(rules);
    var parse = function(source) {
        var blockSource = source + "\n\n";
        return rawBuiltParser(blockSource, {inline: false});
    };
    // You probably only need one of these: choose depending on
    // whether you want react nodes or an html string:
    var reactOutput = SimpleMarkdown.outputFor(rules, 'react');
    var htmlOutput = SimpleMarkdown.outputFor(rules, 'html');

Now we can use our custom parse and output functions to parse markdown with underlines!

    var syntaxTree = parse("__hello underlines__");
    console.log(JSON.stringify(syntaxTree, null, 4));
    => [
        {
            "content": [
                {
                    "content": [
                        {
                            "content": "hello underlines",
                            "type": "text"
                        }
                    ],
                    "type": "underline"
                }
            ],
            "type": "paragraph"
        }
    ]

    reactOutput(syntaxTree)
    => [ { type: 'div',
        key: null,
        ref: null,
        _owner: null,
        _context: {},
        _store: { validated: false, props: [Object] } } ]

    htmlOutput(syntaxTree)

    => '<div class="paragraph"><u>hello underlines</u></div>'

Basic parsing/output API

SimpleMarkdown.defaultBlockParse(source)

Returns a syntax tree of the result of parsing source with the default markdown rules. Assumes a block scope.

SimpleMarkdown.defaultInlineParse(source)

Returns a syntax tree of the result of parsing source with the default markdown rules, where source is assumed to be inline text. Does not emit <p> elements. Useful for allowing inline markdown formatting in one-line fields where paragraphs, lists, etc. are disallowed.

SimpleMarkdown.defaultImplicitParse(source)

Parses source as block if it ends with \n\n, or inline if not.

SimpleMarkdown.defaultOutput(syntaxTree)

Returns React-renderable output for syntaxTree.

Note: raw html output will be coming soon

Extension Overview

Elements in simple-markdown are generally created from rules. For parsing, rules must specify match and parse methods. For output, rules must specify a react or html method (or both), depending on which outputter you create afterwards.

Here is an example rule, a slightly modified version of what simple-markdown uses for parsing strong (bold) text:

    strong: {
        match: function(source, state, lookbehind) {
            return /^\*\*([\s\S]+?)\*\*(?!\*)/.exec(source);
        },
        parse: function(capture, recurseParse, state) {
            return {
                content: recurseParse(capture[1], state)
            };
        },
        react: function(node, recurseOutput) {
            return React.DOM.strong(null, recurseOutput(node.content));
        },
        html: function(node, recurseOutput) {
            return '<strong>' + recurseOutput(node.content) + '</strong>';
        },
    },

Let's look at those three methods in more detail.

match(source, state, lookbehind)

simple-markdown calls your match function to determine whether the upcoming markdown source matches this rule or not.

source is the upcoming source, beginning at the current position of parsing (source[0] is the next character).

state is a mutable state object to allow for more complicated matching and parsing. The most common field on state is inline, which all of the default rules set to true when we are in an inline scope, and false or undefined when we are in a block scope.

DEPRECATED - use state.prevCapture instead. lookbehind is the string previously captured at this parsing level, to allow for lookbehind. For example, lists check that lookbehind ends with /^$|\n *$/ to ensure that lists only match at the beginning of a new line.

If this rule matches, match should return an object, array, or array-like object, which we'll call capture, where capture[0] is the full matched source, and any other fields can be used in the rule's parse function. The return value from Regexp.prototype.exec fits this requirement, and the common use case is to return the result of someRegex.exec(source).

If this rule does not match, match should return null.

NOTE: If you are using regexes in your match function, your regex should always begin with ^. Regexes without leading ^s can cause unexpected output or infinite loops.

parse(capture, recurseParse, state)

parse takes the output of match and transforms it into a syntax tree node object, which we'll call node here.

capture is the non-null result returned from match.

recurseParse is a function that can be called on sub-content and state to recursively parse the sub-content. This returns an array.

state is the mutable state threading object, which can be examined or modified, and should be passed as the third argument to any recurseParse calls.

For example, to parse inline sub-content, you can add inline: true to state, or inline: false to force block parsing (to leave the parsing scope alone, you can just pass state with no modifications). For example:

    var innerText = capture[1];
    recurseParse(innerText, _.defaults({
        inline: true
    }, state));

parse should return a node object, which can have custom fields that will be passed to output, below. The one reserved field is type, which designates the type of the node, which will be used for output. If no type is specified, simple-markdown will use the current rule's type (the common case). If you have multiple ways to parse a single element, it can be useful to have multiple rules that all return nodes of the same type.

react(node, recurseOutput, state)

react takes a syntax tree node and transforms it into React-renderable output.

node is the return value from parse, which has a type field of the same type as the current rule, as well as any custom fields created by parse.

recurseOutput is a function to recursively output sub-tree nodes created by using recurseParse in parse.

state is the mutable state threading object, which can be examined or modified, and should be passed as the second argument to any recurseOutput calls.

The simple-markdown API contains several helper methods for creating rules, as well as methods for creating parsers and outputters from rules.

Extension API

simple-markdown includes access to the default list of rules, as well as several functions to allow you to create parsers and outputters from modifications of those default rules, or even from a totally custom rule list.

These functions are separated so that you can customize intermediate steps in the parsing/output process, if necessary.

SimpleMarkdown.defaultRules

The default rules, specified as an object, where the keys are the rule types, and the values are objects containing order, match, parse, react, and html fields (these rules can be used for both parsing and outputting).

SimpleMarkdown.parserFor(rules)

Takes a rules object and returns a parser for the rule types in the rules object, in order of increasing order fields, which must be present and a finite number for each rule. In the case of order field ties, rules are ordered lexicographically by rule name. Each of the rules in the rules object must contain a match and a parse function.

SimpleMarkdown.outputFor(rules, key)

Takes a rules object and a key that indicates which key in the rules object is mapped to the function that generates the output type you want. This will be 'react' or 'html' unless you are defining a custom output type.

It returns a function that outputs a single syntax tree node of any type that is in the rules object, given a node and a recursive output function.

Putting it all together

Given a set of rules, one can create a single function that takes an input content string and outputs a React-renderable as follows. Note that since many rules expect blocks to end in "\n\n", we append that to source input manually, in addition to specifying inline: false (inline: false is technically optional for all of the default rules, which assume inline is false if it is undefined).

var rules = {
    ...SimpleMarkdown.defaultRules,
    paragraph: {
        ...SimpleMarkdown.defaultRules.paragraph,
        react: (node, output, state) => {
            return <p key={state.key}>{output(node.content, state)}</p>;
        }
    }
};

var parser = SimpleMarkdown.parserFor(rules);
var reactOutput = SimpleMarkdown.outputFor(rules, 'react'));
var htmlOutput = SimpleMarkdown.outputFor(rules, 'html'));

var blockParseAndOutput = function(source) {
    // Many rules require content to end in \n\n to be interpreted
    // as a block.
    var blockSource = source + "\n\n";
    var parseTree = parser(blockSource, {inline: false});
    var outputResult = htmlOutput(parseTree);
    // Or for react output, use:
    // var outputResult = reactOutput(parseTree);
    return outputResult;
};

LICENSE

MIT. See the LICENSE file for text.