Version 8 (modified by cblp, 5 years ago) (diff)


Literate Markdown

Markdown has grown in popularity since github started encouraging people to write their documentation with it. Github highlights the source according to how it's labeled, so the haskell code blocks look nice, as do the HTML blocks. As an example, here is a blog post written in markdown and rendered by github's source view:

And here is the source:

As you can see, the source is a haskell module, and if GHC accepted .md or .markdown as input, this blog post would be executable.

Current Literate Processing

Haskell already supports literate files, using two different styles:

Using "bird-tracks":

This is a comment.  Lines starting with '>' are the actual code.

> average xs = sum xs / length xs

Or, using the LaTeX compatible notation:

This is a comment.

average = sum xs / length xs

Unfortunately, neither of this is compatible with mark-down: in mark-down the bird-tracks signify quoting (just like in e-mail clients) and, of course, \begin{code} is LaTeX.

The Proposal

The idea is to extend Haskell's literate notation so that it is compatible with markdown, in the same way that \begin{code} makes it work with LaTeX. This is great for two reasons:

  1. markdown is a simple language that is used by many programmers
  2. there are many existing tools that know how to process the markdown notation (e.g., github, pandoc, etc.)

To support literate Haskell written in markdown we need two changes:

  1. A new way to indicate what are the code parts in a literate Haskell file
  2. (Optional, but nice.) Disable bird-tracks style Haskell blocks in markdown files, so that GHC does not accidentally interpret quotes as code.

How to Recognize Code Blocks

A new code block would be started by ``` or ```haskell at the beginning of a line, the code starts on the next line. The code block is terminated by a line that starts with ```. For example:

# Heading

This is a comment, and next there will be some code:

x :: Int
x  = 10

Markdown supports snippets of code in different languages, which is why there is haskell after the ticks. If a language is not explicitly specified, then we assume that the block contains Haskell---after all, we are working with literate Haskell files.

When to Use Literate Markdown

Currently, GHC supports mixing various styles code blocks---one may have a literate Haskell file that contains both bird-tracks and LaTeX style literate Haskell. It is very unclear if this generality is ever useful.

For literate markdown we propose a different approach: the style of the literate Haskell file will be determined by the file's extension. So, if the compiler is handed a file ending in .md or .markdown, the common suffixes for markdown files, then it will unlit the file using the markdown convention. Ordinary .lhs files will be processed as usual, so this is fully backward compatible.


The above design has been implemented in .

The implementation involved the following:

  1. Modify unlit to support the new form of code block
  2. Don't get confused with CPP handline (see bug #4836)
  3. Make GHC look for modules in .md and .markdown files.
  4. TODO when unliting markdown, do not recognize bird-tracks as code blocks.