Standing up a Hakyll blog with Windows and Netlify

Posted on August 1, 2021

This is my second attempt at a static-site-gen blog. So far it’s going a lot better than the first, because I’m working with some technologies I really like, but that’s not to say everything was smooth sailing. That means a story, which means this blog post.

I got up yesterday morning and decided to start messing around with Haskell on my personal desktop (I write Haskell for work, too, and this felt like a decent chance to get better at it.) The first project that popped into my mind was resurrecting my blog. I’ve been meaning to try Hakyll for a few years now, but for various reasons it hasn’t happened.

Context

This is a “what I tried and how I got things to work” story, not a tutorial. You can probably guess from the look of the site that I’m not done yet. As such this won’t be a great reference for googling error messages, but it just might help someone who’s trying to do the same thing. I’ll be happy if it entertains.

Step one: WSL2

My desktop is built for gaming, a bit of digital art, and a lot of shitpoasting on twitter, so it runs Windows 10. This isn’t an operating system well-known for supporting Haskell - before my new job, my tool of choice was F#. Fortunately, Microsoft made 2019 the Year Of Linux On The Desktop with WSL2, so I installed that. Two reboots later I had an ubuntu-flavoured bash prompt in front of me, and I apt-got Haskell-Platform.

Work uses cabal so I figured I would too, but when I ran cabal update it spat out an ominous warning about backward-compatibility and deprecation, and scolded me to use the new- versions of various verbs. I figured I’d try that with cabal new-install hakyll, and bounced off an error about not being able to open a pipe. An hour of fucking around and several rm -rf ~/.cabals later, neither the old nor the new cabal installs would get properly through pulling down Hakyll, so I gave up and uninstalled ubuntu. (This did leave WSL2 enabled, which may be important later.)

Step two: Visual Studio Code devcontainers

Well, I’m not gonna let myself feel ashamed of my gamer-oriented software. I’d planned to work in VS Code anyway - I use it for work, it’s pretty good - and went back to google, where I found an old friend: Igal Tabachnik wrote up a “Haskell on Windows in minutes!” tutorial using a Docker container and VS Code. Even better, a Haskell container had been officially adopted by the VS Code devcontainers crew, a mere nine days (fresh! relevant!) before I’d looked.

So I followed the VS Code tutorial and downloaded Docker Desktop. That tutorial is very clear that you need to give Docker Desktop permission to access your filesystem, at least where your code lives. Unfortunately, there’s no Resources > File Sharing option available, as referenced by the tutorial.

Well, I got this far by dint of a “let’s keep going and see what happens” attitude, and I wasn’t going to let this slow me down. I tried the Node container suggested in the tutorial and then went looking for that Haskell container.

I found a note on Igal’s github to the effect of “don’t use this, use theirs!” Well. As far as I can tell there isn’t a list on a web page.

With a bit of digging through other platforms’ tutorials I discovered the magic sauce:

  1. Open a project directory
  2. Open the VS Code command palette
  3. Select “Remote-Containers: Add Development Container Configuration Files”
  4. Start typing “Hask…” until something pops up

I guess one interpretation of ✨The Magic Of Containers✨ is that they serve as configuration files? Whatever. I’m up and running in a container with ghc installed!

Step three: stack, jacked

The Hakyll tutorials largely use stack over cabal, and I remembered getting bitten by cabal only an hour or two earlier, so I figured I’d try stack. Alas, stack install hakyll ran into three sets of version incompatibilities for Hakyll’s dependencies, and even though stack’s errors helpfully told me how to add exceptions for those particular package versions, two of them asked me to satisfy mutually exclusive version bounds.

Fortunately, stack also told me where the “YOLO” setting was for version constraints, and after enabling that and waiting approximately a week and a half for Pandoc to compile, I had a working install of Hakyll 4.14.

The great virtue of Haskell is that “if it compiles, it runs”, right? On to the next step, building a site skeleton. That went off without a hitch and led me to the meat of the Hakyll workflow, building and running the site generator.

A brief digression on why I’m doing all this

It’s been about four years since I blogged with Jekyll. One of my strongest impressions of that tool is its configuration-orientedness - plugins and yaml files and other activities that are at best programming-adjacent. Hakyll, however, promises to let you write code to change how you want your static site generated. It’s more of a framework for writing an SSG than a stanndalone SSG in itself. That’s why I have to build it here, but it’s also why I’m on this path in the first place.

Building the site generator is as simple as stack build. Except, stack can’t figure out how to build my site generator, because it can’t find version 4.14 of Hakyll on any of the resolvers it knows about.

You know, the package I just laboriously built locally, from which I generated this haskell project.

After half an hour of fruitless fucking around, I gave up on stack and went back to cabal, meticulously prefixing everything with new-. That took another week and a half to compile Pandoc, but worked fine in the end.

Magically, I now have a static site!

Step four: Nifty Netlify

Now that I have some HTML, CSS, and a lambda-looking image, I just need a place to put it. My old blog was hosted on github.io, but since then I acquired the audax-labs.com vanity domain and I haven’t really done anything with it. Time to pop into my AWS account, where the domain’s hosted…

…and remember which authenticator app I used for the admin account…

…and remember that it was the same thing I used for $oldjob, which I deleted in a fit of digital cleanliness after I quit…

fuck!

Fortunately, AWS has a great account recovery process if you’ve kept access to your root account’s email. Five minutes later I was logged in as root and setting up MFA devices once again, by far the fastest problem I overcame in this whole saga.

I did all that so I could hook up Netlify to the domain. For my purposes here, at least, Netlify is a hosting company with a CI back-end that hooks into a git repository (github, in my case). Push to main, Netlify picks that up, and files get copied over. Once AWS negotiated the DNS transfer and blog.audax-labs.com went live, they even provisioned an SSL cert for me. The AWS account recovery might’ve been the fastest problem I solved, but that’s because Netlify didn’t present any problems.

I’m using their free tier, which offers one build agent and 300 CI minutes per month. Building static HTML locally, pushing it to a branch, and copying it out costs approximately zero CI minutes per deploy, so I think I’ll be good for a while.

Step five: tailwind

Now that I had a site and a public place to put it, I wanted to make it look better. That means CSS, and an excuse to play around with Tailwind CSS which I’ve been aware of for a little while.

Unfortunately, I don’t know shit-all about CSS, and Tailwind’s docs are written for people who do this stuff for a living. It took me a few minutes to generate a Tailwind CSS file and start seeing some change in the rendered result, and several hours to figure out how to actually get it to work properly.

In particular, most of the instructions for integrating Tailwind assume that you’re adding it to an existing toolset rather than YOLO’ing it from the command line. This combined with my lack of CSS knowledge - turns out you can’t just magic up @-directives - and Firefox’s cheerful willingness to render as much malformed CSS as it can rather than spit out error messages. Eventually I figured out that I needed to combine Tailwind with Hakyll’s default CSS and run it all through nvx tailwindcss to build a style sheet that’d do what I want. In retrospect, the tailwind CLI docs told me what I needed to know but assumed I knew what workflow to use, which I didn’t.

This is perhaps not a great use for Tailwind - which gives you fine control over laying out and rendering individual components, where Hakyll generates a bunch of bulk HTML for you out of markdown - but screw it, I’m having fun.

Step six: conclusions?

I’m not entirely sure how to end this post. I got an awful lot done just by beating my head against it, but by the same token the barriers to entry for setting up a blog like this one are shockingly low in this year of our lord two thousand and twenty-one. I clearly have a lot of work to do, but at the same time what I’ve done on this blog kinda kicks ass.

Also, this was all inspired by quinoa. Stay frosty for that.