Obelisk tutorial, Markdown preview with Reflex

In this article, I’ll describe how to get a full-stack Haskell application up and running. In particular, the app will render user-entered Markdown text in real-time (the final version is hosted at https://commonmark.srid.ca). Notably, our app will use Haskell even on the frontend. This is made possible by the GHCJS compiler, that compiles Haskell code to JavaScript.

We will not work directly with GHCJS, however, and instead will use the Reflex-FRP library, through the excellent Obelisk full-stack framework.

Create an Obelisk project

First and foremost, make sure you have Obelisk installed, and then follow along.

Obelisk includes a command called ob that can be used to initialize a project. We will also use git to keep track of changes:

mkdir MarkdownPreview
cd ./MarkdownPreview
# Create an Obelisk project
ob init
# Add to git
git init
git add .
git commit -m "first commit"

This gives us a project layout with three Haskell packages: backend, common and frontend. As the names indicate, frontend contains the Haskell code that ultimate gets compiled to JavaScript. The common package however contains code that is shared between the backend and the frontend. This is extremely useful for type sharing, which is impossible with something like Elm or PureScript (without explicit conversion).


You will notice that Obelisk uses Nix to build your project. The command ob run (described below) will abstract over the Nix stuff, including any GHCi handling, so you do not have to deal directly with Nix except for overriding dependencies.

Running hello world

Now, it is time to run our app.

ob run

This command may take a while to finish the very first time it is run, as it would need to download packages from the Nix caches. At the end you would expect to see: Frontend running on http://localhost:8000/. Your obelisk app will be accessible at that URL.

Interlude: ob run reloads your code

So what does this ob run command do? Think of it as stack run or cabal run - but it also recompiles changed sources and reloads application. ob run uses ghcid underneath, in combination with custom ghci config to specify the modules to reload.

There is also ob repl which gives you a GHCi repl for your project. As well as ob hoogle providing a local Hoogle server for project and its dependencies.

Add commonmark-hs dependency

Our application will use the Pure Haskell Markdown parser commonmark-hs, which is written by the author of Pandoc, who intends to migrate Pandoc over to it eventually.

We will also use the latest version from Git, instead of Hackage. Some Nix-fu is helpful at this stage. But the main thing you need to know, in order to add a custom Haskell dependency, is the following general workflow:

  • Clone the git repo under a subdirectory
  • “Pack” it using ob thunk
  • Load it in default.nix using hackGet

As briefly as possible, you would do this for commonmark-hs as follows:

# Get the source
git clone https://github.com/jgm/commonmark-hs.git dep/commonmark-hs
# Pack it
ob thunk pack dep/commonmark-hs

Next, edit your default.nix and:

  • Add hackGet and pkgs as arguments to the project if they don’t already exist
  • Use the packed thunk by calling hackGet
  • The git rep contains multiple Haskell packages; use commonmark and commonmark-extensions

Here’s how your default.nix should look:

project ./. ({ pkgs, hackGet, ... }: {
  packages = let 
    commonmarkSrc = hackGet ./dep/commonmark-hs;
  in {
    commonmark = commonmarkSrc + "/commonmark";
    commonmark-extensions = commonmarkSrc + "/commonmark-extensions";

If you run ob run at this point, it may complain about further missing dependencies. Let’s override each of them:

git clone https://github.com/jgm/emojis.git dep/emojis
ob thunk pack dep/emojis

Go back to default.nix, and add this new dependency to the packages attribute:

    emojis = hackGet ./dep/emojis;

As we will be using commonmark directly in the frontend, go ahead and add these to frontend.cabal (under the library stanza):

-- frontend/frontend.cabal
               , commonmark
               , commonmark-extensions

Restart ob run, which should build the the new dependency before starting our app.

Interlude: What is an Obelisk thunk?

If you are familiar with Haskell overrides in Nix, then think of the obelisk thunk mechanism as an abstraction on top. A “packed” thunk is essentially similar to Nix’s fetchGit in that you specify the exact source revision of the dependency to use.

But a thunk can also be “unpacked”, using ob thunk unpack, which – in addition to unpacking it as a git clone – has the effect of adding it to ob run and ob repl sessions. For example, you could unpack the above commonmark thunk using ob thunk unpack dep/commonmark-hs and restart ob run. Now, when you hack on ./dep/commonmark-hs and change its Haskell sources, ob run will automatically reload the app using the modified commonmark-hs. See the Obelisk ChangeLog for details.

Let’s add a textbox

If you are unfamiliar with Reflex, checkout the official guide. All our frontend code is defined in the frontend/src/Frontend.hs file. With ob run running by side, open that module and try changing a few things, like the title - and the app should update. Let’s add a textbox element where the user would write their Markdown text.

Somewhere in _frontend_body, add the following:

markdownText :: Dynamic t T.Text <-
  fmap value $ textAreaElement $
    & initialAttributes .~ ("style" =: "width:50%;height:15em;")

markdownText is a reflex Dynamic that holds the user-entered text.

Parse Markdown

Now it is time to actually use the commonmark library. Let’s import it:

import qualified Commonmark as CM

renderMarkdown :: T.Text -> Either CM.ParseError (CM.Html ())
renderMarkdown =
  CM.commonmark "markdown"

The renderMarkdown function will parse our Markdown and return the HTML representation of it. Calling show on the Right value gets us the raw HTML.

Render to HTML

Finally let’s plug everything together. We want to parse and render the resulting HTML every time the user changes the textbox. Reflex’s Dynamic automatically updates, so let’s use that.

result <- eitherDyn $ fmap renderMarkdown markdownText
dyn_ $ ffor result $ \case
  Left err ->
    dyn_ $ ffor err $ \_ -> text "Parse error"
  Right htmlVal ->
    prerender_ blank $ void $ elDynHtml' "div" $ T.pack . show <$> htmlVal

That’s all it takes! Now as you type the Markdown text, its live preview will automatically update next to the textbox.

You can even hack on commonmark-hs (see the Obelisk thunk interlude above), and have ob run automatically reload when the library sources change. This is extremely handy if you want to play with the internals of the Markdown parser and see its live result in the browser.

Further resources

Links to this page
  • Neuron 0.6 released

    Previous versions of neuron used the mmark parser which advertises itself as being “strict”. This prevented certain links from working as they caused the parser to fail. In version 0.6, I switched over to use commonmark-hs, which is the Haskell implementation of the CommonMark specification, written by the same author of Pandoc. Another reason for switching is that commonmark-hs and its extensions can be compiled to JavaScript via GHCJS*, which is useful when reusing the Haskell code with the web app (see below).

    See Obelisk tutorial, Markdown preview with Reflex for a description of the development workflow involving writing Haskell on the frontend.

#blog #nojs