Skip to content

The only Markdown escaper passing backtranslation tests.

License

Notifications You must be signed in to change notification settings

orchitech/gfm-escape

Repository files navigation

GfmEscape

...the only escaper passing backtranslation tests.

GfmEscape is an enterprise-grade library for transforming untagged plain text to CommonMark and GitHub Flavored Markdown (GFM).

Why GfmEscape?

There are neat and configurable markup converters like Turndown, which even allows transforming any markup that can be converted to HTML first.

While conversion of inline and block constructs is well covered, little attention is paid to transforming text content itself. And this is tricky especially with non-delimited "extended" autolinks, which make escaping heavily context-dependent.

In short:

GfmEscape addresses these issues without significant performance penalty, as it is based on UnionReplacer. See below for more details.

Outline

Installation and usage

In browsers:

<script src="https://unpkg.com/union-replacer/dist/union-replacer.umd.js"></script>
<script src="https://unpkg.com/gfm-escape/dist/gfm-escape.umd.js"></script>

Using npm:

npm install gfm-escape

In Node.js:

const GfmEscape = require('gfm-escape');

Synopsis

escaper = new GfmEscape(escapingOptions[, syntax])
newStr = escaper.escape(input[, gfmContext[, metadata]])

A created GfmEscape instance is intended to be reused and shared in your code.

Parameters

escapingOptions: option object defining how to perform escaping, its keys correspond to individual replaces. When a replace option is set to any truthy value, suboption defaults are applied and can be overriden by passed suboptions. A single option object can be reused for instantiating escapers for different syntaxes, some options would just render irrelevant. The current full options are:

{
  strikethrough: { // default false
    optimizeForDoubleTilde: false,
  },
  extAutolink: { // default false
     breakUrl: false,
     breakWww: false,
     breaker: '<!-- -->',
     allowedTransformations: [ 'entities', 'commonmark' ],
     allowAddHttpScheme: false,
     inImage: false,
  },
  table: true, // default false
  emphasisNonDelimiters: { // default true
    maxIntrawordUnderscoreRun: undefined,
  },
  linkTitle: { // default true
    delimiters: [ '"', '\'', '()' ],
    alwaysEscapeDelimiters: [],
  },
}

See below for more details.

syntax: suggests the syntax escaper is built for. The predefined syntaxes are available as members of GfmEscape.Syntax:

  • text: normal text, the default.
  • linkDestination: text rendered [sometext](here).
  • cmAutolink: text rendered <here>. Please note that a valid CommonMark must contain a URI scheme, which cannot be addressed by the escaper. When deciding if CommonMark autolink is an appropriate construct to use, we suggest to use the isEncodable(input) and wouldBeUnaltered(input) methods on the Syntax.cmAutolink object.
  • codeSpan: text rendered `here`.
  • linkTitle: text rendered [text](destination "here") or [text](destination 'here') or [text](destination (here)).

input: the string to escape. Please note that correct escaping is currently only guaranteed when the input is trimmed and normalized in terms of whitespace. The library does not perfos qrm whitespace normalizing on its own, as it is often ensured by the source's origin, e.g. textContent of a normalized HTML DOM. Manual normalizing can be done with input.trim().replace(/[ \t\n\r]+/g, ' '). If it is intended to keep the source somewhat organized in lines, the minimum treatment to make escaping safe would be input.replace(/^[ \t\r\n]+|[ \t]+$/gm, ''). In such case, the caller has a responsibility to place the output correctly in the generated document. I.e. to indent all the lines when the context requires indenting.

gfmContext: extra contextual information to be considered. The contexts have no defaults, i.e. they are falsy by default. The following contexts are available:

{
  inLink: true, // indicates suppressing nested links
  inImage: true, // similar to inLink for ![this image text](img.png)
  inTable: true, // indicates extra escaping of table contents
}

When escaping, metadata is extra input-output parameter that collects metadata about the actual escaping. Currently metadata are used for codeSpan syntax and linkTitle syntax.

const escaper = new GfmEscape({ table: true }, GfmEscape.Syntax.codeSpan);
const x = {}; // not necessary as the surrounding delimiter is always '`'
const context = { inTable: true };
const escaped = escaper.escape('`array|string`', context, x);
console.log(`\`${escaped}\``); // `` `array\|string` ``
console.log(`${x.extraBacktickString.length} backtickts and ${x.extraSpace.length} spaces added.`);
// 1 backticks and 1 spaces added.

const linkTitleEscaper = new GfmEscape({}, GfmEscape.Syntax.linkTitle);
const x = {}; // needed as we let GfmEscape decide the surrounding delimiter
let escaped = escaper.escape('cool "link \'title\'"', context, x);
console.log(`${x.startDelimiter}${escaped}${x.endDelimiter}`);
// (cool "link 'title'")

escaped = escaper.escape('average link title', context, x);
console.log(`${x.startDelimiter}${escaped}${x.endDelimiter}`);
// "average link title"

const rigidLinkTitleEscaper = new GfmEscape({
  linkTitle: {
    delimiters: '"',
  }
}, GfmEscape.Syntax.linkTitle);
// metadata not necessary, as the surronding delimiter will be always '"'
escaped = escaper.escape('cool "link \'title\'"');
console.log(`"${escaped}"`);
// "cool \"link 'title'\""

Escaping options: strikethrough

Defaults to false, i.e. '~' is not special and it is not escaped.

Suboptions:

  • optimizeForDoubleTilde: only eventual sequences of two tildes are escaped. Default false.

Escaping options: extAutolink

Defaults to false, i.e. autolinks are not detected and do not form special case for escaping.

Suboptions:

  • breakUrl: if a string capable of forming extended url autolink is encountered, it is broken to prevent that. E.g. https://orchi.tech becomes https://<!-- -->orchi.tech. Default false.
  • breakWww: if a string capable of forming extended www autolink is encountered, it is broken to prevent that. E.g. www.orchi.tech becomes www.<!-- -->orchi.tech. Default false.
  • breaker: a sequence used to break extended autolinks, used both for breaking and terminating. Default <!-- -->. Please note that some Markdown renderers like Redcarpet do not support HTML comments - tag sequences like <span></span> or artificial <nolink> can be used instead.
  • allowedTransformations: array of transformations that are allowed if an extended autolink-like string needs to be transformed to retain the expected target and text. The order indicates priority. Defaults to ['entities', 'commonmark']. Available transformations are:
    • 'keep': always the most preferred, no reason to set it explicitly.
    • 'entites': entity name references are used to escape trailing characters. E.g. *http://orchi.tech,* becomes \*http://orchi.tech,&ast;.
    • 'commonmark': a CommonMark autolink is used to delimit the actual link part. E.g. *http://orchi.tech,* becomes \*<http://orchi.tech>,\*.
    • 'breakup': autolink-like string is broken, so that it is not interpreted as an autolink. E.g. *https://orchi.tech,* becomes \*https://<!-- -->orchi.tech,\*.
    • 'breakafter': autolink-like string is terminated after the actual link part. E.g. *https://orchi.tech,* becomes \*https://orchi.tech<!-- -->,\*. This transformation is the default fallback, no reason to set it explicitly.
  • allowAddHttpScheme: add http:// scheme when a transformation needs it to work. E.g. *www.orchi.tech,* would become \*<http://www.orchi.tech>,\* with the commonmark transformation.
  • inImage: suggest if extended autolink treatment should be applied within image text. Although the CommonMark spec says links are interpreted and just the stripped plain text part renders to the alt attribute, cmark-gfm actually does not do it for extended autolinks, so the default is false.

How to choose the options:

  1. Consider rendering details of the target Markdown flavor. Backtranslation test should pass on text. And if a link is produced, it should match the input.
  2. Consider user expectations. The users probably don't expect HTML comments all over their documents. They probably don't expect HTML entity references too, but see also the next point.
  3. Consider declared semantics. Transforming to CommonMark autolinks looks quite well, but CommonMark autolinks form explicit link demarkation when the input was not explicitly link-demarked. 'breakafter' might be better option in some situations.
  4. Last, but not least - consider the origin of your input. If you transform HTML rendered from another markup language that supports autolinking too, you may expect that an autolink-suppression mechanism was used if an autolink-like string is encountered in plain text. Then it might be better to break it too.
    And if the original renderer supports url autolinks, but not www autolinks, it might be better to set only 'breakUrl', as users may still expect www links to be autolinked in the plain text.

Escaping options: emphasisNonDelimiters

Defaults to true, i.e. intraword emphasis delimiters are not escaped if it is safe not to escape them. E.g. in My account is joe_average., the underscore stays unescaped as joe_average, not joe\_average.

Suboptions:

  • maxIntrawordUnderscoreRun: if defined as a number, it sets the maximum length of intraword underscores to be kept as is. E.g. for 1 and input joe_average or joe__average, the output would be joe_average or joe\_\_average. This is helpful for some renderers like Redcarpet. Both undefined and false mean no limit on unescaped intraword underscore run length. Defaults to undefined.

Escaping options: table

Defaults to false, i.e. table pipes are not escaped. If enabled, rendering of table delimiter rows is suppressed by escaping its pipes and all pipes are escaped when in table context.

Escaping options: linkTitle

Suboptions:

  • delimiters: array of allowed delimiter to be chosen from or a single delimiter. Delimiters are ", ' and (). When more delimiters are allowed, GfmEscape picks the least interferring one. The picked delimiter is returned in metadata, as shown in the example above.
  • alwaysEscapeDelimiters: array of delimiters that are always escaped.

GFM escaping details

Terminology:

Specs:

Reference implementations examined:

Implementation notes

Remarks on cmark-gfm library

While cmark-gfm is somewhat a reference implementation of GFM Spec, we have found a few interesting details...

  • cmark_gfm-001: Contrary to the GFM spec stating All such recognized autolinks can only come at the beginning of a line, after whitespace, or any of the delimiting characters *, _, ~, and (, it seems this applies just to extended www autolinks in cmark-gfm. E.g. .https://orchi.tech is recognized as an autolink by this library. We follow this.
  • cmark_gfm-002: Contrary to the GFM spec, extended autolinks in cmark-gfm do not treat [\v\f] as space, while CM autolinks do. We follow this.
  • cmark_gfm-003: cmark-gfm considers < as valid for autolink detection and trims the resulting link afterwards. So https://or_chi.tech.< leads to autolinking of https://or_chi.tech, although this wouldn't form autolink without the trailing <. We follow this, but non-explicit extended autolink transformations would break the autolink detection - which is probaly good. E.g. with the default settings, https://or_chi.tech.< leads to https://or_chi.tech.&lt; (wouldn't be detected as extended autolink by cmark-gfm), while https://or_chi.tech.<~ leads to <https://or_chi.tech>\<\~ (forced CM autolink).
  • cmark_gfm-004: GFM spec says If an autolink ends in a semicolon (;), we check to see if it appears to resemble an entity reference; if the preceding text is & followed by one or more alphanumeric characters. If so, it is excluded from the autolink... Alphabetic references cmark-gfm
  • cmark_gfm-005: Backslash escape in link destination, e.g. [foo](http://orchi.tech/foo\&lowbar;bar) does not prevent entity reference from interpreting in rendered HTML. We use entity encoding instead, i.e. &amp;. The same applies to link titles.

TODO

  • Minification for browsers.
  • Complete and polish implementation remarks.