wiki:LiterateMarkdown

Version 6 (modified by diatchki, 10 months 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:

https://github.com/elliottt/elliottt.github.com/blob/source/source/_posts/2013-02-19-serenade-in-haskell.markdown

And here is the source:

https://raw.github.com/elliottt/elliottt.github.com/source/source/_posts/2013-02-19-serenade-in-haskell.markdown

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.

\begin{code}
average = sum xs / length xs
\end{code}

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.

There is one open ticket on the bug tracker about markdown as a literate format, #4836, that concerns unlit not processing markdown correctly. As unlit attempts to preserve CPP in the whole file unless otherwise instructed, markdown headings are treated as CPP and left in.

# Heading

```haskell
x :: Int
x  = 10
```

Is translated to the following by unlit:

# Heading


x :: Int
x  = 10

Handling Markdown

As markdown has two commonly used extensions, .md and .markdown, it seems reasonable to instruct unlit to not attempt to keep lines that it views as CPP from within code blocks. Unlit has two flags that should be able to handle this, -r and -#, which when combined will not leave any markdown headers in the resulting source.

Additionally, if the .md and .markdown extensions are to be supported, it should be reasonable to start ghci simply using the module name, and have ghc also search for files with those extensions.

Patch

The above suggestion has been implemented in https://github.com/elliottt/ghc/tree/literate-markdown .