Hakyll setup

Main image

tl;dr: How I use hakyll. Abbreviations, typography corrections, multi-language, use index.html, etc…

This website is done with Hakyll.

Hakyll can be considered as a minimal cms. But more generally it is a library helping file generation. We can view it as an advanced build system (like make).

From the user perspective I blog this way:

  1. I open an editor (vim in my case) and edit a markdown file. It looks like this
A First Level Header

A Second Level Header

Who would cross the Bridge of Death must answer me
these questions three, ere the other side he see.
This is just a regular paragraph.

Ask me the questions, bridgekeeper. I am not afraid.

### Header 3

> This is a blockquote.
> This is the second paragraph in the blockquote.
> ## This is an H2 in a blockquote
  1. I open a browser and reload time to time to see the change.
  2. Once I finished I’ve written a very minimal script which mainly do a git push. My blog is hosted on github.

Being short sighted one could reduce the role of Hakyll to:

create (resp. update) html file when I create (resp. change) a markdown file.

While it sounds easy, there are a lot of hidden details:

  • Add metadatas like keywords.
  • Create an archive page containing a list of all the posts.
  • Deal with static files.
  • Creating an rss feed.
  • Filter the content with some function.
  • Dealing with dependencies.

The work of Hakyll is to help you with these. But let’s start with the basic concepts.

The concepts and syntax


For each file you create, you have to provide:

  • a destination path
  • a list of content filters.

First, let’s start with the simplest case: static files (images, fonts, etc…). Generally, you have a source directory (here is the current directory) and a destination directory _site.

The Hakyll code is:

-- for each file in the static directory
match "static/*" do
  -- don't change its name nor directory
  route   idRoute
  -- don't change its content
  compile copyFileCompiler

This program will copy static/foo.jpg to _site/static/foo.jpg. I concede this is a bit overkill for a simple cp. Now how to write a markdown file and generate an html one?

-- for each file with md extension in the "posts/" directory
match "posts/*.md" do
  -- change its extension to html
  route $ setExtension "html"
  -- use pandoc library to compile the markdown content into html
  compile $ pandocCompiler

If you create a file posts/foo.md, it will create a file _site/posts/foo.html.

If the file posts/foo.md contains

# Cthulhu

ph'nglui mglw'nafh Cthulhu R'lyeh wgah'nagl fhtagn

the file _site/posts/foo.html, will contain

<p>ph'nglui mglw'nafh Cthulhu R'lyeh wgah'nagl fhtagn</p>

But horror! _site/posts/cthulhu.html is not a complete html file. It doesn’t have any header nor footer, etc… This is where you use templates. I simply add a new directive in the compile block.

match "posts/*.md" do
  route $ setExtension "html"
  compile $ pandocCompiler
    -- use the template with the current content
    >>= loadAndApplyTemplate "templates/post.html" defaultContext

Now if templates/posts.html contains:

    <title>How could I get the title?</title>

our cthulhu.html contains (indentation added for readability):

    <title>How could I get the title?</title>
    <p>ph'nglui mglw'nafh Cthulhu R'lyeh wgah'nagl fhtagn</p>

See, it’s easy But we have a problem. How could we change the title or add keywords?

The solution is to use Contexts. For this, we first need to add some metadatas to our markdown1.

title: Cthulhu
# Cthulhu

ph'nglui mglw'nafh Cthulhu R'lyeh wgah'nagl fhtagn

And modify slightly our template:


As Sir Robin said just before dying before the Bridge of Death:

“That’s EASY!”

Sir Robin, the Not-Quite-So-Brave-As-Sir-Lancelot

Real customization

Now that we understand the basic functionality. How to:

  • use SASS?
  • add keywords?
  • simplify url?
  • create an archive page?
  • create an rss feed?
  • filter the content?
  • add abbreviations support?
  • manage two languages?


That’s easy. Simply call the executable using unixFilter. Of course you’ll have to install SASS (gem install sass). And we also use compressCss to gain some space.

match "css/*" $ do
    route   $ setExtension "css"
    compile $ getResourceString >>=
              withItemBody (unixFilter "sass" ["--trace"]) >>=
              return . fmap compressCss

Add keywords

In order to help to reference your website on the web, it is nice to add some keywords as meta datas to your html page.

<meta name="keywords"
      content="Cthulhu, Yog-Sothoth, Shub-Niggurath">

In order to add keywords, we could not directly use the markdown metadatas. Because, without any, there should be any meta tag in the html.

An easy answer is to create a Context that will contains the meta tag.

-- metaKeywordContext will return a Context containing a String
metaKeywordContext :: Context String
-- can be reached using $metaKeywords$ in the templates
-- Use the current item (markdown file)
metaKeywordContext = field "metaKeywords" $ \item -> do
  -- tags contains the content of the "tags" metadata
  -- inside the item (understand the source)
  tags <- getMetadataField (itemIdentifier item) "tags"
  -- if tags is empty return an empty string
  -- in the other case return
  --   <meta name="keywords" content="$tags$">
  return $ maybe "" showMetaTags tags
      showMetaTags t = "<meta name=\"keywords\" content=\""
                       ++ t ++ "\">\n"

Then we pass this Context to the loadAndApplyTemplate function:

match "posts/*.md" do
  route $ setExtension "html"
  compile $ pandocCompiler
    -- use the template with the current content
    >>= loadAndApplyTemplate "templates/post.html"
            (defaultContext <> metaKeywordContext)

☞ Here are the imports I use for this tutorial.

{-# LANGUAGE OverloadedStrings #-}
import           Control.Monad          (forM,forM_)
import           Data.List              (sortBy,isInfixOf)
import           Data.Monoid            ((<>),mconcat)
import           Data.Ord               (comparing)
import           Hakyll
import           System.Locale          (defaultTimeLocale)
import           System.FilePath.Posix  (takeBaseName,takeDirectory

Simplify url

What I mean is to use url of the form:


I prefer this than having to add file with .html extension. We have to change the default Hakyll route behavior. We create another function niceRoute.

-- replace a foo/bar.md by foo/bar/index.html
-- this way the url looks like: foo/bar in most browsers
niceRoute :: Routes
niceRoute = customRoute createIndexRoute
    createIndexRoute ident =
        takeDirectory p </> takeBaseName p </> "index.html"
    where p=toFilePath ident

Not too difficult. But! There might be a problem. What if there is a foo/index.html link instead of a clean foo/ in some content?

Very simple, we simply remove all /index.html to all our links.

-- replace url of the form foo/bar/index.html by foo/bar
removeIndexHtml :: Item String -> Compiler (Item String)
removeIndexHtml item = return $ fmap (withUrls removeIndexStr) item
    removeIndexStr :: String -> String
    removeIndexStr url = case splitFileName url of
        (dir, "index.html") | isLocal dir -> dir
        _                                 -> url
        where isLocal uri = not (isInfixOf "://" uri)

And we apply this filter at the end of our compilation

match "posts/*.md" do
  route $ niceRoute
  compile $ pandocCompiler
    -- use the template with the current content
    >>= loadAndApplyTemplate "templates/post.html" defaultContext
    >>= removeIndexHtml

Create an archive page

Creating an archive start to be difficult. There is an example in the default Hakyll example. Unfortunately, it assumes all posts prefix their name with a date like in 2013-03-20-My-New-Post.md.

I migrated from an older blog and didn’t want to change my url. Also I prefer not to use any filename convention. Therefore, I add the date information in the metadata published. And the solution is here:

match "archive.md" $ do
  route $ niceRoute
  compile $ do
    body <- getResourceBody
    return $ renderPandoc body
      >>= loadAndApplyTemplate "templates/archive.html" archiveCtx
      >>= loadAndApplyTemplate "templates/base.html" defaultContext
      >>= removeIndexHtml

Where templates/archive.html contains



And base.html is a standard template (simpler than post.html).

archiveCtx provide a context containing an html representation of a list of posts in the metadata named posts. It will be used in the templates/archive.html file with $posts$.

archiveCtx =
  defaultContext <>
  metaKeywordContext <>
  field "posts" (\_ -> postList createdFirst)

postList returns an html representation of a list of posts given an Item sort function. The representation will apply a minimal template on all posts. Then it concatenate all the results. The template is post-item.html:

<li><a href="$url$">$published$ - $title$</a></li>

Here is how it is done:

postList :: [Item String] -> Compiler [Item String]
            -> Compiler String
postList sortFilter = do
    -- sorted posts
    posts   <- loadAll "post/*" >>= sortFilter
    itemTpl <- loadBody "templates/post-item.html"
    -- we apply the template to all post
    -- and we concatenate the result.
    -- list is a string
    list    <- applyTemplateList itemTpl defaultContext posts
    return list

createdFirst sort a list of item and put it inside Compiler context. We need to be in the Compiler context to access metadatas.

createdFirst :: [Item String] -> Compiler [Item String]
createdFirst items = do
  -- itemsWithTime is a list of couple (date,item)
  itemsWithTime <- forM items $ \item -> do
    -- getItemUTC will look for the metadata "published" or "date"
    -- then it will try to get the date from some standard formats
    utc <- getItemUTC defaultTimeLocale $ itemIdentifier item
    return (utc,item)
  -- we return a sorted item list
  return $ map snd $ reverse $ sortBy (comparing fst) itemsWithTime

It wasn’t so easy. But it works pretty well.

Create an rss feed

To create an rss feed, we have to:

  • select only the lasts posts.
  • generate partially rendered posts (no css, js, etc…)

We could then render the posts twice. One for html rendering and another time for rss. Remark we need to generate the rss version to create the html one.

One of the great feature of Hakyll is to be able to save snapshots. Here is how:

match "posts/*.md" do
  route $ setExtension "html"
  compile $ pandocCompiler
    -- save a snapshot to be used later in rss generation
    >>= saveSnapshot "content"
    >>= loadAndApplyTemplate "templates/post.html" defaultContext

Now for each post there is a snapshot named “content” associated. The snapshots are created before applying a template and after applying pandoc. Furthermore feed don’t need a source markdown file. Then we create a new file from no one. Instead of using match, we use create:

create ["feed.xml"] $ do
      route idRoute
      compile $ do
        -- load all "content" snapshots of all posts
        loadAllSnapshots "posts/*" "content"
        -- take the latest 10
        >>= (fmap (take 10)) . createdFirst
        -- renderAntom feed using some configuration
        >>= renderAtom feedConfiguration feedCtx
        feedCtx :: Context String
        feedCtx =  defaultContext <>
                   -- $description$ will render as the post body
                   bodyField "description"

The feedConfiguration contains some general informations about the feed.

feedConfiguration :: FeedConfiguration
feedConfiguration = FeedConfiguration
  { feedTitle = "Great Old Ones"
  , feedDescription = "This feed provide information about Great Old Ones"
  , feedAuthorName = "Abdul Alhazred"
  , feedAuthorEmail = "abdul.alhazred@great-old-ones.com"
  , feedRoot = "http://great-old-ones.com"

Great idea certainly steal from nanoc (my previous blog engine)!

Filter the content

As I just said, nanoc was my preceding blog engine. It is written in Ruby and as Hakyll, it is quite awesome. And one thing Ruby does more naturally than Haskell is regular expressions. I had a lot of filters in nanoc. I lost some because I don’t use them much. But I wanted to keep some. Generally, filtering the content is just a way to apply to the body a function of type String -> String.

Also we generally want prefilters (to filter the markdown) and postfilters (to filter the html after the pandoc compilation).

Here is how I do it:

markdownPostBehavior = do
  route $ niceRoute
  compile $ do
    body <- getResourceBody
    prefilteredText <- return $ (fmap preFilters body)
    return $ renderPandoc prefilteredText
    >>= applyFilter postFilters
    >>= saveSnapshot "content"
    >>= loadAndApplyTemplate "templates/post.html"    yContext
    >>= loadAndApplyTemplate "templates/boilerplate.html" yContext
    >>= relativizeUrls
    >>= removeIndexHtml


applyFilter strfilter str = return $ (fmap $ strfilter) str
preFilters :: String -> String
postFilters :: String -> String

Now we have a simple way to filter the content. Let’s augment the markdown ability.

Add abbreviations support

Comparing to LaTeX, a very annoying markdown limitation is the lack of abbreviations.

Fortunately we can filter our content. And here is the filter I use:

abbreviationFilter :: String -> String
abbreviationFilter = replaceAll "%[a-zA-Z0-9_]*" newnaming
    newnaming matched = case M.lookup (tail matched) abbreviations of
                          Nothing -> matched
                          Just v -> v
abbreviations :: Map String String
abbreviations = M.fromList
    [ ("html", "<span class=\"sc\">html</span>")
    , ("css", "<span class=\"sc\">css</span>")
    , ("svg", "<span class=\"sc\">svg</span>")
    , ("xml", "<span class=\"sc\">xml</span>")
    , ("xslt", "<span class=\"sc\">xslt</span>") ]

It will search for all string starting by ‘%’ and it will search in the Map if there is a corresponding abbreviation. If there is one, we replace the content. Otherwise we do nothing.

Do you really believe I type

<span style="text-transform: uppercase">L<sup style="vertical-align: 0.15em; margin-left: -0.36em; margin-right: -0.15em; font-size: .85em">a</sup>T<sub style="vertical-align: -0.5ex; margin-left: -0.1667em; margin-right: -0.125em; font-size: 1em">e</sub>X</span>

each time I write LaTeX?

Manage two languages

Generally I write my post in English and French. And this is more difficult than it appears. For example, I need to filter the language in order to get the right list of posts. I also use some words in the templates and I want them to be translated.

<a href="$otherLanguagePath$"
	$changeLanguage$ </a>

First I create a Map containing all translations.

data Trad = Trad { frTrad :: String, enTrad :: String }

trads :: Map String Trad
trads = M.fromList $ map toTrad [
      , "Français"))
      ("Changer de theme"
      ,"Change Theme"))
      ("Ces liens sociaux préservent votre vie privée"
      ,"These social sharing links preserve your privacy"))
    toTrad (key,(french,english)) =
      (key, Trad { frTrad = french , enTrad = english })

Then I create a context for all key:

tradsContext :: Context a
tradsContext = mconcat (map addTrad (M.keys trads))
    addTrad :: String -> Context a
    addTrad name =
      field name $ \item -> do
          lang <- itemLang item
          case M.lookup name trads of
              Just (Trad lmap) -> case M.lookup (L lang) lmap of
                          Just tr -> return tr
                          Nothing -> return ("NO TRANSLATION FOR " ++ name)
              Nothing -> return ("NO TRANSLATION FOR " ++ name)


The full code is here. And except from the main file, I use literate Haskell. This way the code should be easier to understand.

If you want to know why I switched from nanoc:

My preceding nanoc website was a bit too messy. So much in fact, that the dependency system recompiled the entire website for any change.

So I had to do something about it. I had two choices:

  1. Correct my old code (in Ruby)
  2. Duplicate the core functionalities with Hakyll (in Haskell)

I added too much functionalities in my nanoc system. Starting from scratch (almost) remove efficiently a lot of unused crap.

So far I am very happy with the switch. A complete build is about 4x faster. I didn’t broke the dependency system this time. As soon as I modify and save the markdown source, I can reload the page in the browser.

I removed a lot of feature thought. Some of them will be difficult to achieve with Hakyll. A typical example:

In nanoc I could take a file like this as source:

# Title


<code file="foo.hs">
main = putStrLn "Cthulhu!"

And it will create a file foo.hs which could then be downloaded.



<a href="code/foo.hs">Download foo.hs</a>
<pre><code>main = putStrLn "Cthulhu!"</code></pre>

  1. We could also add the metadatas in an external file (foo.md.metadata).


comments powered by Disqus
Published on 2013-03-16
Follow @yogsototh
Yann Esposito©
Done with Vim & nanoc Hakyll