BSD-3-Clause licensed by John MacFarlane
Maintained by [email protected]
This version can be pinned in stack with:commonmark-0.2.2@sha256:e6b0b1889a56a53f8513a6b94f5d2f1eecd23f52bef90bcfe72fe322a8db7b74,3454

commonmark

hackage release

This package provides the core parsing functionality for commonmark, together with HTML renderers.

:construction: This library is still in an experimental state. Comments on the API and implementation are very much welcome. Further changes should be expected.

The library is fully commonmark-compliant and passes the test suite for version 0.30 of the commonmark spec. It is designed to be customizable and easily extensible. To customize the output, create an AST, or support a new output format, one need only define some new typeclass instances. It is also easy to add new syntax elements or modify existing ones.

Accurate information about source positions is available for all block and inline elements. Thus the library can be used to create an accurate syntax highlighter or an editor with synced live preview.

Finally, the library has been designed for robust performance even in pathological cases. The parser behaves well on pathological cases that tend to cause stack overflows or exponential slowdowns in other parsers, with parsing speed that varies linearly with input length.

Related libraries

  • commonmark-extensions provides a set of useful extensions to core commonmark syntax, including all GitHub-flavored Markdown extensions and many pandoc extensions. For convenience, the package of extensions defining GitHub-flavored Markdown is exported as gfmExtensions.

  • commonmark-pandoc defines type instances for parsing commonmark as a Pandoc AST.

  • commonmark-cli is a command-line program that uses this library to convert and syntax-highlight commonmark documents.

Simple usage example

This program reads commonmark from stdin and renders HTML to stdout:

{-# LANGUAGE ScopedTypeVariables #-}
import Commonmark
import Data.Text.IO as TIO
import Data.Text.Lazy.IO as TLIO

main = do
  res <- commonmark "stdin" <$> TIO.getContents
  case res of
    Left e                  -> error (show e)
    Right (html :: Html ()) -> TLIO.putStr $ renderHtml html

Notes on the design

The input is a token stream ([Tok]), which can be be produced from a Text using tokenize. The Tok elements record source positions, making these easier to track.

Extensibility is emphasized throughout. There are two ways in which one might want to extend a commonmark converter. First, one might want to support an alternate output format, or to change the output for a given format. Second, one might want to add new syntactic elements (e.g., definition lists).

To support both kinds of extension, we export the function

parseCommonmarkWith :: (Monad m, IsBlock il bl, IsInline il)
                    => SyntaxSpec m il bl -- ^ Defines syntax
                    -> [Tok] -- ^ Tokenized commonmark input
                    -> m (Either ParseError bl)  -- ^ Result or error

The parser function takes two arguments: a SyntaxSpec which defines parsing for the various syntactic elements, and a list of tokens. Output is polymorphic: you can convert commonmark to any type that is an instance of the IsBlock typeclass. This gives tremendous flexibility. Want to produce HTML? You can use the Html () type defined in Commonmark.Types for basic HTML, or Html SourceRange for HTML with source range attributes on every element.

GHCI> :set -XOverloadedStrings
GHCI>
GHCI> parseCommonmarkWith defaultSyntaxSpec (tokenize "source" "Hi there") :: IO (Either ParseError (Html ()))
Right <p>Hi there</p>
> parseCommonmarkWith defaultSyntaxSpec (tokenize "source" "Hi there") :: IO (Either ParseError (Html SourceRange))
Right <p data-sourcepos="source@1:1-1:9">Hi there</p>

Want to produce a Pandoc AST? You can use the type Cm a Text.Pandoc.Builder.Blocks defined in commonmark-pandoc.

GHCI> parseCommonmarkWith defaultSyntaxSpec (tokenize "source" "Hi there") :: Maybe (Either ParseError (Cm () B.Blocks))
Just (Right (Cm {unCm = Many {unMany = fromList [Para [Str "Hi",Space,Str "there"]]}}))
GHCI> parseCommonmarkWith defaultSyntaxSpec (tokenize "source" "Hi there") :: Maybe (Either ParseError (Cm SourceRange B.Blocks))
Just (Right (Cm {unCm = Many {unMany = fromList [Div ("",[],[("data-pos","source@1:1-1:9")]) [Para [Span ("",[],[("data-pos","source@1:1-1:3")]) [Str "Hi"],Span ("",[],[("data-pos","source@1:3-1:4")]) [Space],Span ("",[],[("data-pos","source@1:4-1:9")]) [Str "there"]]]]}}))

If you want to support another format (for example, Haddock’s DocH), just define typeclass instances of IsBlock and IsInline for your type.

Supporting a new syntactic element generally requires (a) adding a SyntaxSpec for it and (b) defining relevant type class instances for the element. See the examples in Commonmark.Extensions.*. Note that SyntaxSpec is a Monoid, so you can specify myNewSyntaxSpec <> defaultSyntaxSpec.

Performance

Here are some benchmarks on real-world commonmark documents, using make benchmark. To get benchmark.md, we concatenated a number of real-world commonmark documents. The resulting file was 355K. The bench tool was used to run the benchmarks.

program time (ms)
cmark 12
cheapskate 105
commonmark.js 217
commonmark-hs 229
pandoc -f commonmark 948

It would be good to improve performance. I’d welcome help with this.

Changes

Changelog for commonmark

0.2.2

  • Blocks: export getParentListType [API change].
  • Require unicode-data >= 0.3.
  • Change mkFormattingSpecMap so it integrates different FormattingSpecs that use the same character (#87). Otherwise we have problems if you have one formatting spec that reacts to single delimiters and another that reacts to pairs; if the first fails to match, the fallback behavior is produced and the second never matches.
  • Use unicode-data’s faster versions of Data.Char functions. This speeds up benchmarks for tokenize considerably; little difference in other benchmarks. unicode-data is already a transitive dependency, via unicode-transforms.
  • Increase strictness in tokenize/go.
  • Remove legacy cpp needed to support ghc < 8.4.

0.2.1.1

  • Fix bug in prettyShow for SourceRange (#80). The bug led to an infinite loop in certain cases.

0.2.1

  • Use official 0.30 spec.txt.
  • Update HTML block parser for recent spec changes.
  • Fix test case from commonmark/cmark#383. We need to index the list of stack bottoms not just by the length mod 3 of the closer but by whether it can be an opener, since this goes into the calculation of whether the delimiters can match.

0.2

  • Commonmark.Inlines: export LinkInfo(..) [API change].
  • Commonmark.Inlines: export pLink [API chage].
  • Comonmark.ReferenceMap: Add linkPos field to LinkInfo [API change].
  • Commonmark.Tokens: normalize unicode to NFC before tokenizing (#57). Normalization might affect detection of flankingness, recognition of reference links, etc.
  • Commonmark.Html: add data-prefix to non-HTML5 attributes, as pandoc does.
  • Remove unnecessary build-depends.
  • Use lightweight tasty-bench instead of criterion for benchmarks.

0.1.1.4

  • Fix build with GHC 9.0.1 (Simon Jakobi, #72).

0.1.1.3

  • Fix bug in links with spaces at the beginning or end of link description (#67). We were putting flankingness constraints on the link delimiters, but this isn’t requried by the spec.

0.1.1.2

  • Fix bug in fix to #65 (#66).

0.1.1.1

  • Fixed corner case with link suffix parsing, which could result in dropped tokens in certain cases (#65).

0.1.1

  • Export reverseSubforests from Commonmark.Blocks [API change] (#64).

0.1.0.2

  • Fix tight/loose list detection with multiple blank lines at end (#56).

0.1.0.1

  • Set source position when we add a token in gobbleSpaces (#54). This fixes a bug in gobbling indented spaces in some nested contexts.
  • Drop support for ghc 7.10/base 4.8. We need StrictData. Move SCC annotations; ghc 8.0.x doesn’t support them on declarations.

0.1.0.0

  • Initial release