New blog with Hakyll!

 David Sferruzza -   (last updated: ) -  en

Update: I do not use this system anymore, but this article may still be helpful.


I had this « make yourself a decent website » task on my todo list for more than a year. As I am more and more interested in what I do, I decided it was time to begin to give back and share some stuff I like.

Also, the « I should totally write a blog post about that! » situation happened too many times repeatedly these last weeks.

I think I will mostly write about Computer Science/Programming, and projects I like and/or make. But who knows...


The first thing I want to write about is how I made this website/blog.

I used a Haskell library called Hakyll. Hakyll makes it easy to build a static website generator.

There are lots of static website generators[^1], but Hakyll is the best I tried. It is fast, while being highly customizable. And it can handle many cool stuff almost out-of-the-box, including:

  • templating
  • markdown (and many other formats, because it uses Pandoc)
  • incremental build (you don't have to rebuild everything on every modification)
  • CSS compression
  • tags and metadata
  • RSS/Atom feeds
  • ...

Making a static website by hand can be painful. To avoid repeating code, people often use dynamic languages (such as PHP). But this sounds like a waste to me: on each request, the server has to compute the same operations over and over to serve the same page, just because the developer wants DRY[^2] code. This makes sense when you need to handle user inputs, but that is not always the case. For that kind of projects, Hakyll is a really nice tool.

You can find the source code here:

Some little details

Two static website generators are never exactly the same (that's the point). So building one involves picking up some stuff from the example site, or from nice people, and developing the rest.

I will quickly show some of these features I picked up or developed.

I'm still a beginner, so if you have better solutions, please let me know!

Links in feed

As I said, Hakyll is able to generate RSS and Atom feeds[^3]. While generating a post's page, we will tell Hakyll to take a snapshot of the result, just after converting the post's markdown to HTML, and just before applying it to a template. This snapshot will then be put in the feed.

But what if a post contains a relative link?

Well, it will stay relative, and that's what we want for the post's page. But we don't want that for the feed, because feed readers won't be able to resolve it.

A solution would be to use the xml:base[^4] attribute in a root element of the feed. The feed rendering function let me think it would be possible to override the default feed templates. But I don't know if it is possible yet, nor how to do it...

So I picked up another solution from Clément Delafargue's blog. When generating posts' pages, Hakyll will resolve relative links before taking a snapshot for the feed, and then transform them back so that they will be relative in the rendered page (but not in the feed).

I believe that the first solution would be cleaner, but the second works nicely.

Update date

When I update a blog post, I want the date of the update to appear in the blog's page, but also in the feed (using <updated> tags).

Hakyll handles that with a modificationTimeField[^5] you can add to your post's context. But I don't really like the idea of the modification date only be kept on my file system. I want to be able to clone my blog's Git repository, build it and get exactly the same result every time. This would not be the case if I use a modificationTimeField, because Git don't store files' modification date.

So I created a updatedField, which is implemented exactly like dateField except that it only tries to get the date from a updated field in my content's metadata.

updatedField :: String -> String -> Context a
updatedField key format = field key $ \i -> do
    time <- getUpdatedTime locale $ itemIdentifier i
    return $ formatTime locale format time
    locale = defaultTimeLocale

getUpdatedTime :: MonadMetadata m => TimeLocale -> Identifier -> m UTCTime
getUpdatedTime locale id' = do
    metadata <- getMetadata id'
    let tryField k fmt = M.lookup k metadata >>= parseTime' fmt
    maybe empty' return $ msum $ [tryField "updated" fmt | fmt <- formats]
    empty'     = fail $ "getUpdatedTime: " ++
        "could not parse time for " ++ show id'
    parseTime' = parseTime locale
    formats    =
        [ "%a, %d %b %Y %H:%M:%S %Z"
        , "%Y-%m-%dT%H:%M:%S%Z"
        , "%Y-%m-%d %H:%M:%S%Z"
        , "%Y-%m-%d"
        , "%B %e, %Y %l:%M %p"
        , "%B %e, %Y"
        , "%b %d, %Y"

Now, if I want to indicate that I updated a post, I just need to add something like updated: 2014-12-01 12:30:00+02:00 in the post's metadata. It will be synced by Git :)

ISO 8601 dates

I wanted to use <time>[^6] tags in my posts' pages. <time> tags binds a machine-readable date to a human-readable date. This means we need a machine-readable date.

I picked up a quite simple solution from Julien Tanguy's blog. You just need to add a new dateField to the post's context. This dateField would have a format argument like that: (iso8601DateFormat Nothing).

Cabal sandbox

When you install Hakyll following the official tutorial, you type cabal install hakyll, which installs globally the latest version of Hakyll. In a previous project, I had some troubles when I started to update this globally-installed Hakyll.

So I decided I will use a sandbox this time.

Sandboxes ( in cabal > 1.18 ) are self contained environments of Haskell packages.[^7]

My haskell's dependencies (including Hakyll) are written in a .cabal file. Then it's easy:

  • cabal sandbox init to initialize the sandbox
  • cabal install --dependencies-only to install dependencies

It seems like, in a near future, Hakyll will generate the .cabal file when doing a hakyll-init my-site[^8].


Of course, it is not possible to make a comment system using Hakyll. But it is possible to integrate one.

I was lazy so I choose to use Disqus. As it includes a lot of stuff (Google Analytics, ~ 700 ko of data which is around 200% of this page's size without comments), I don't let it load itself automatically.

I made something quick and dirty to do so:

<div id="disqus_thread"><a href="#disqus_thread" onclick="disqus()">Load comments</a></div>

<script type="text/javascript">
	function disqus() {
		// Disqus code

You should give Hakyll a try!

[^1]: Someone even made a website to list them [^2]: Don't Repeat Yourself [^3]: See the official tutorial [^4]: See [^5]: See [^6]: Specification: [^7]: I took this quote from [^8]: According to