Simon Shine

Agentic Playwright testing with Nix

Sat, 15 Aug 2026 21:31:07 +0200

I don’t do much front-end web development these days, but I enable people who do.

I use a lot of coding agents for my daily work with back-end APIs and service meshes.

I’ve known for a long time that agentic workflows instrumenting Playwright to run headless browser tests based on prompts has advanced quite a lot. I just haven’t got around to try it out.

tl;dr ¶

Playwright, its browsers and its MCP server are all in nixpkgs. No package.json, no npm install, no npx playwright install.
nixpkgs’ playwright-test hardcodes the full browser set into its wrapper, so the naive setup drags 2.2 GB of Chromium, Firefox and WebKit into your devshell whether you want them or not. I managed to push that down to 634 MB.
Headed Chromium and interactive debugging lives behind a flake-parts partition.
The planner/generator/healer agents run on an existing Claude Code session.
I practice TDD workflows for writing end-to-end tests.

The application under test ¶

The thing I’m practising on is a generic client area dashboard for a software service, written in Rust on topcoat with toasty. So I’m doing three things here: Testing topcoat, toying around with client dashboards, and getting my agentic playwright testing workflow up.

Hashpinning

Sat, 08 Aug 2026 01:59:35 +0200

GitHub Actions workflows usually start like this:

- uses: actions/checkout@v6

v6 is a git tag that points to some commit for the release of this version of an action. There is also a v6.0.1 tag. Every time a minor or patch version is released, e.g. v6.0.2 or v6.1.0, the v6 tag is moved to that. Being unspecific means you always get the latest version that begins with v6. This is a feature.

Git config-based hooks with hk-nix

Sat, 01 Aug 2026 00:00:00 +0100

hk is a fast, modern git hook manager and project linter configured in Pkl. hk-nix is to hk what lefthook-nix is to lefthook: you install hk with Nixpkgs, declare your hooks in Nix, pin every tool with Nix, and the hooks install themselves whenever you enter the devshell. It is the same stack I previously built with lefthook, with a nicer configuration language and one fewer workaround.

Synergies ¶

Nix + direnv: You only ever need to install these two manually: Nix being the package manager, and direnv being the thing that activates the Nix devshell. All other tooling is derived, pinned and installed for you.

Using sops-nix templates to support passwords in legacy configs

Sun, 05 Jul 2026 22:51:02 +0200

Note: This article builds on my dendritic sops-nix flake template.

Nix is littered with alternative solutions to things.

Most people who use Nix to manage their irssi configuration would use the home-manager irssi extensions. But I have a principle that I’ve held onto: I like to keep program configuration non-nixified so that they can still hypothetically be used outside of Nix.

The problem: irssi keeps password authentication in the config file ¶

Irssi was first released in 1999. Back then you didn’t think about these things.

Authenticating Tailscale non-interactively via API with sops-nix

Sun, 05 Jul 2026 21:01:12 +0200

For the past three years I’ve connected my Nix machines using Tailscale by running sudo tailscale login once per machine.

I’ve known that you can actually log onto a Tailnet with an API key.

Why bother? Running sudo tailscale login takes ten seconds, your browser’s password manager kicks in, you’re done.

This is actually simple and ships almost out of the box, if you already have sops-nix secrets management in your system configuration.

A dendritic sops-nix flake template

Sun, 05 Jul 2026 02:35:37 +0200

This tutorial sets up a flake using flake-parts, import-tree, devshell and sops-nix.

You can see the template here: https://github.com/sshine/nix-sops-flake-template

Unlike agenix/ragenix, the sops-nix NixOS module does not require importing a Nix file describing recipients, so it composes cleanly inside flake-parts modules without triggering IFDs.

1. A dendritic flake ¶

This flake auto-imports all .nix files in lib/.

You may want for that directory to be named something else, depending on your project.

Dendritic flake partitions

Sun, 21 Jun 2026 13:39:00 +0200

tl;dr ¶

Some of my sandboxed CLI agents aren’t in nixpkgs, so they ride along as flake inputs.
They’re heavy, and were polluting the flake.lock of every machine that consumes the flake.
flake-parts partitions move those inputs into a nested flake that’s only evaluated when you actually build the thing.
Because my flake is dendritic (import-tree), I tuck that nested flake inside the import tree using import-tree’s ignored-paths convention: _inputs/flake.{nix,lock}.

While researching this I found a good guide, Dr. Steve’s Dendritic Design with the Flake Parts Framework: A guide on how to structure your Nix code with Flake Parts using the Dendritic Pattern. It covers aspects, collectors, factories, but doesn’t mention flake-parts partitions or lazy input loading.

Managing Secrets privately with sops-nix

Fri, 15 May 2026 02:37:41 +0200

tl;dr ¶

I want to commit my secrets encrypted to git with sops-nix
I want to publish my infra-as-code, but don’t want to publish my git secrets
So I put my secrets in a private git+file flake input

Motivation ¶

I run Kubernetes clusters managed declaratively with Talos Linux and kubenix. I want to publish the entire repository that defines the cluster. It includes commands that require authentication (kubectl, talosctl, hcloud, etc.) that don’t work without their secrets.

Is the novelty budget dead?

Fri, 24 Apr 2026 12:25:31 +0200

Michael Snoyman wrote a thinkpiece called The Boring Haskell Manifesto (2019) about how to adopt Haskell into your organization, and how to make your organization productive with Haskell for non-wizards.

The article introduced me to the concept of a “novelty budget” (Mark Wotton, 2018): the limit your project has on new, cool stuff. If you exceed your novelty budget you impose too much risk: new software with unknown limitations, spontaneous learning curves, longer onboarding, harder to hire, longer to develop, things you might need to scrap and rewrite in a less fancy way.

NixOS runners on Forgejo Actions

Thu, 23 Apr 2026 22:06:40 +0200

Note: You can see the project in action here: https://git.shine.town/infra/runners

I decided to bootstrap a Kubernetes cluster with Talos and kubenix during my vacation.

A lot of experience came out of that, and I’m trying to unpack it gradually.

One of the things I host on this cluster is Forgejo, which is Codeberg’s fork of Gitea. Forgejo has Forgejo Actions which is a CI system that mimicks GitHub Actions, and it has a Container Registry much like GHCR, or DockerHub.

Three levels of Nix

Sun, 15 Mar 2026 12:34:52 +0100

When using Nix to orchestrate a new project, I find myself choosing between these three levels:

1. Just a `shell.nix` ¶

This is good for starting on, it’s easy to sell to colleagues or project owners you’re not close with or already bought on the idea of Nix. It leaves a very small footprint in terms of lines of code: Every lines in this file is worth its own weight, since it usually just mentions packages to install. That also makes it very tangible. It could look like:

Vendored theme workflow after one month

Sat, 07 Mar 2026 23:03:24 +0100

It’s exactly one month since I wrote:

Vendoring Hugo themes with Nix and Git Feb 7, 2026

I read an interesting article, The Rust Calling Convention We Deserve by Miguel Young de la Sota, and I couldn’t help but think that the website was really nicely made: Inspirational use of CSS3, a personal font choice, and proper underlines under hyperlinks.

Using lefthook and treefmt with Nix

Sat, 07 Mar 2026 21:47:14 +0100

In my recent attempts to modernize my CI pipelines, I’ve experimented with and subsequently let go of pre-commit. The reason why I didn’t stick with pre-commit is that pre-commit drags in several software runtimes and toolchains when building it from scratch. When Nix fails to build something from nixpkgs, it’s an omen of too many chefs.

Alternatives I’m aware of:

lefthook: A generic tool written in Go; defines hooks in a lefthook.yml file.
husky: A generic tool written in JavaScript; enough said.
cargo-husky: A tool specialized to Rust/Cargo; very nice, but not generic enough.

I initially disregarded lefthook because of its special lefthook.yml.

Dendritic Nix With nixos-shell

Sun, 01 Mar 2026 14:03:12 +0100

tl;dr ¶

I’m refactoring my system flake using The Dendritic Pattern
I’m testing the system flake using nixos-shell
I’ll completely wipe my laptop’s NixOS installation when ready

Motivation (some ranting) ¶

Last week I experimented with nixos-anywhere, deploy-rs and disko on Hetzner which lets me easily deploy NixOS on a Hetzner VM even though NixOS is not one of the dashboard options.

Learning about AutoAgents by benchmarking Claude Code vs. OpenCode

Sun, 22 Feb 2026 23:22:30 +0100

tl;dr ¶

I’m extending AutoAgents so that it supports Z.ai, a frontier AI company
Z.ai’s GLM-5 model might or might not compete with Claude Opus 4.5 at a very competitive price.
I’m testing both the OpenCode CLI vs. Claude Code CLI, and Z.ai’s GLM-5 against Claude Opus 4.5.
Claude is considerably faster, but neglects testing. OpenCode’s tests were not end-to-end.
OpenStack writes tests, but not ones that actually contact the API. When I ask it to do that, it does, and the test fails without fixing it, and when I copy-paste the error code from running it, it gets fixed.
Compare their commits here and here.
I’d feel confident to vibe-code the full integration with either stack.

Alternative to OpenClaw in Rust ¶

Lately I’ve been playing around with OpenClaw. I really like the whole idea, and I don’t like the codebase. It uses Node.js and that whole ecosystem just smells like security incidents. I did some research into who is making AI frameworks in Rust, and I found two interesting projects:

nixos-anywhere on Hetzner

Fri, 20 Feb 2026 00:55:22 +0100

tl;dr: ¶

I installed NixOS using nixos-anywhere for the first time.
Along the way I also finally tried (and succeeded) to use disko and deploy-rs.
Deploying from MacBook’s aarch64-darwin to the VM’s x86-64 required a few tweaks.

Tonight I wanted to try and run OpenClaw. I thought I would give it a persistent VM and that it should run NixOS, of course. I already run several NixOS VMs in the cloud, but their deployment was made with nixos-infect, which I promised myself I wouldn’t do again.

Vaultwarden on NixOS

Thu, 19 Feb 2026 08:51:15 +0100

This post is about self-hosting Vaultwarden, an unofficial Bitwarden server, so that you can have self-hosted password management. Bitwarden clients on computers and phones can be configured to use a custom server.

I found a few blog posts that deserve some credit; sadly, none of them provide the simplest possible configuration, or every piece of information I needed.

They either assume you want to host your Vaultwarden on a VPN (making Let’s Encrypt not an option), using secrets management, using Borg as backup system, or using postgresql as the database. All reasonable choices, but hardly the simplest possible.

A git primer for future colleagues

Sat, 14 Feb 2026 15:56:43 +0100

You become a git expert when you learn about git reflog. Not because it’s an advanced feature.

But because it gives you the confidence that whatever command you run can be reverted.

Once you can undo any git command, your willingness to experiment increases significantly, and you become the export.

Besides that, git reflog is not the most useful command, since it’s mostly for when you fucked up, which is less and less.

Vendoring Hugo themes with Nix and Git

Sat, 07 Feb 2026 13:44:17 +0100

tl;dr: ¶

I forked the git repo for my Hugo theme and pushed my feature to a branch
I refactored the blog repo’s Nix code to be multi-file and flake-based
I switched my Nix-vendored theme to use this git repo/branch
I made it possible to use the git repo checked out locally for rapid prototyping
I retained the possibility to use the Nix-vendored theme (same branch, but committed and pushed)
I found a way to easily deploy using either pathway without changing files (--theme, just)

Context ¶

I migrated this website to Hugo in February 2025:

Speeding up CI more with Nix

Mon, 02 Feb 2026 00:44:38 +0100

Recently I wrote Speeding up CI with Nix.

I improved a 2m30s build to 1m30s after it had recently deteriorated to 10m30s.

I have another repository that would benefit from Nix.

Not because of speedups, but because it fails to install Chinese fonts properly.

My Typst résumé’s CI failed to render Chinese fonts for the longest time; for that reason, I never actually made use of the PDFs that I released. It is the fontist/setup-fontist and fontist install --formula noto_serif_cjk parts below (using 3m2s out of 3m20s):

Speeding up CI with Nix

Sun, 25 Jan 2026 12:28:38 +0100

In the article Maintaining libraries with Claude I update the CI pipeline to release the package automatically on Hackage and GitHub when pushing a version tag.

I added a git hook, check-cabal-sync.sh, that checks the version in package.yaml against the version in evm-opcodes.cabal, as well as the version in the git tag, to ensure I ran hpack before releasing; otherwise Hackage upload will fail.

Total CI running time: 2m30s before

Of which haskell-actions/setup takes: 2m5s.

zsh config for NixOS + nix-darwin

Sun, 25 Jan 2026 01:11:55 +0100

tl;dr: ¶

I’m taking Nix more seriously on my MacBook
I want to share my zsh config across all systems
Because my zsh config was NixOS-only, I’ve had to refactor it

Why ¶

I’ve been running NixOS for 3 years now.

During this whole time I’ve also owned a MacBook.

And while I started installing Nix on my MacBook, I didn’t really appreciate it until installing NixOS on an x86-64-based laptop. Nix, compared to Homebrew, just felt complicated and unnecessary as a way to install programs.

Maintaining libraries with Claude

Wed, 14 Jan 2026 23:38:11 +0100

tl;dr ¶

Claude bumps the versions.
Claude finds missing opcodes.
Claude finds even more missing opcodes I didn’t know about.
Claude correctly handles compiler warnings that indicate cascading changes.
Claude correctly handles what files are auto-generated and how to regenerate them.
Stretch Goal #1: Release via CI to Hackage and GitHub
Stretch Goal #2: Validate hpack pre-push

The goal ¶

I maintain an open source library uploaded to a package repository called evm-opcodes. It just exposes an enum with the assembly instructions (opcodes) for Ethereum’s VM, in case you want to build a smart contract compiler in Haskell.

Using NixOS for Exam Computers

Tue, 13 Jan 2026 16:20:48 +0100

I used NixOS to define two exam computers for a programming exam.

One in a presentation room with an examiner and a censor, connected to a projector.
Another one in an adjacent preparation room, just connected to the internet.

Both computers had the same software development environments, and a system user for each student. The presentation computer was connected to a projector with screen mirroring. Some syncing of files was necessary between the two computers. Both computers needed various exam task descriptions on disk.

Kubernetes-valgfaget der ikke blev til noget

Wed, 07 Jan 2026 13:53:14 +0100

Planen var at stoppe som underviser på fuldtid og have et enkelt valgfag som timelærer. Valgfaget blev desværre aflyst efter det var blevet godkendt og var blevet valgt af de studerende, for at spare på eksterne timer. Det var lidt ærgerligt, da jeg havde været transparent omkring planen af hensyn til at give god tid til at finde min afløser.

Titlen skulle have været “Cloud Native Kubernetes” og var det næstmest populære valgfag til foråret.

End of teaching retrospective

Thu, 01 Jan 2026 09:30:15 +0100

I quit my teacher job.

Unlike most of my teaching blog posts, this entry is not in Danish.

I want to write a retrospective blog post about having been a CS teacher for two years, having been on paternity leave, and picking up consulting again, but via an agency instead of freelance, which I had done in the years prior, and not primarily focusing on embedded Rust.

I switched the consulting agency of employment in the last minute before quitting the teacher job. The story that motivated me to stop teaching is tied to an ambitious project that the old agency is doing, and since I wasn’t going to work there after all, the story I told had to change.

Funktionel datamodellering

Thu, 01 Jan 2026 00:00:00 +0100

I Om at blive rekursivt beruset med monoider modellerer jeg superdrinks som en monoid og viser hvordan man kan programmere med dem ved hjælp af simple algebraiske datatyper. Og i Design Patterns i popkultur fandt jeg et eksempel på et design pattern:

PPAP kan modelleres som et frit magma: en algebraisk struktur lukket under komposition, men uden kommutativitet, idempotens, eller identitetselement. Men hvordan gør man det til en konkret datastruktur i et programmeringssprog?

Et frit magma er helt oplagt at modellere som en datatype i et type-sikkert programmeringssprog som Java, Rust eller Haskell. Vi kan udnytte typesystemet til at fange de algebraiske egenskaber vi har analyseret, sådan at datatypen opfører sig efter de regler vi har etableret gælder for det fænomen, vi modellerer.

Animating the NixOS logo

Tue, 30 Dec 2025 03:23:47 +0100

I really like NixOS.

I also like to draw the NixOS logo.

While there are SVGs and PNGs at github:nixos/branding, the logo is very geometric and should be easily expressed as a program. My first attempt with turtle graphics was an experiment to work with HTML Canvas via Wasm. It works by instructing a small robot turtle, Bob, to walk and paint in the shape of a λ; you then repeat that six times:

Design Patterns i popkultur

Thu, 04 Sep 2025 11:34:00 +0200

Jeg har, som mange andre, været grebet af sangen PPAP:

Jeg har ikke kunnet lade sangen slippe, fordi PIKOTARO tydeligvis formidler abstrakt algebra:

Studiet af algebraiske strukturer, hvor man betragter sammensætningen af matematiske objekter, der ikke nødvendigvis er tal.

Abstrakt algebra finder bred anvendelse i programmering. Vi kalder det et designmønster, fordi det er et abstrakt mønster som går igen, når vi arbejder med konkrete datastrukturer. Designmønstre er værd at lære fordi man med tiden kan arbejde på et højere niveau, hvor man kan drage på flere af sine erfaringer, fordi mange ting er lavet på den samme måde.

JSON Schema in Haskell using AutoDoCodec

Sat, 26 Jul 2025 16:00:57 +0100

In April 2022 I made a review of Haskell libraries for JSON Schema.

The outcome of my research was that there weren’t really any good libraries for dealing with JSON Schema from within Haskell. That was a surprising disappointment, because Haskell often has very good libraries (although they may be hard to use). The reason is that the JSON Schema specification itself isn’t very well-made.

I eventually gave up using any of them.

Just: En bedre Makefile

Mon, 21 Jul 2025 23:24:15 +0000

Har du nogensinde arbejdet på et softwareprojekt og set en fil kaldet Makefile? Makefiles bruges traditionelt til at automatisere kompileringsprocesser, men mange udviklere bruger dem også til at køre almindelige projektkommandoer som:

make test
make deploy
make start

Men der findes et bedre alternativ: Justfile

Problemet med Makefiles til projektkommandoer ¶

Makefiles blev oprindeligt designet til at kompilere filer - ikke til at køre projektkommandoer. Når du bruger Make til at køre tasks som make test eller make deploy, misbruger du faktisk værktøjet.

Claude Code og git-bug: Når AI har brug for issue tracking

Mon, 21 Jul 2025 13:03:06 +0000

Jeg prøvede for nyligt at kombinere Claude Code med issue tracking.

Som udvikler er det en udfordring at holde styr på bugs, opgaver og ideer i sine projekter. Vi mennesker bruger grafiske issue tracking-værktøjer som Jira og GitHub Projects. Personligt har jeg været glad for Linear.app, som er gratis op til 250 tickets. Claude Code har sit eget ad-hoc to-do-system, men den er sessionsbundet, og persister ikke med projektet.

Jeg har indtil nu brugt Claude Code til at dumpe dens kontekst ind i tekstfiler såsom CLAUDE.md, PLAN.md og doc/XYZ.md så jeg kan vende tilbage til den forståelse som opbygges omkring delproblemer.

Hvordan understøtter man europæisk cloud i undervisningen?

Tue, 06 May 2025 13:03:06 +0000

tl;dr (den ultra korte version): ¶

I de tekniske fag på Erhvervsakademi København kunne man…

afskaffe brugen af Azure Web Apps, Azure Managed MySQL på 2. semester
introducere Docker Compose og Linux på 2. semester i stedet
driftssætte via Docker Compose på Linux på en VPS på 3. semester

En opsummering af artiklen nedenfor: ¶

Ved at undervise i cloud-agnostisk teknologi som ikke fastlåser én til den enkelte cloud:
- Teknologier som Docker Compose, Linux og SSH, Terraform, Ansible og Kubernetes
- Workflows som GitOps, organisationer såsom CNCF (Cloud Native Computing Foundation)
- VM’er, S3, open-source relationelle databaser, Virtual Private Cloud (netværksopsætning)
Undgå at undervise i cloud-teknologier som er helt unikke for én cloud fordi de er særligt indviklede:
- Databaser: Amazon DynamoDB / Azure Cosmos DB / Google Cloud Firestore
- Proprietære on-premise-løsninger: AWS Outpost
- Web interfaces: Azure Web Apps
- Identitet: Azure AD / Entra ID / AWS IAM / Cloud IAM / Cloud Identity
- Containerized serverless: Google Cloud Run / AWS Fargate / Azure Container Apps
Der findes ingen europæiske cloud-alternativer til de store amerikanske clouds
Der findes suveræne cloud-baserede alternativer, hvis man siger “cloud” er et spektrum
Suverænitet opnås igennem ejerskab frem for outsourcing af drift og intellekt

Hvordan forstås suverænitet? ¶

“The cloud is just someone else’s computer.” - Jeff Atwood (StackOverflow co-founder)

Hvorfor blev jeg datamatiker-lærer?

Thu, 17 Apr 2025 16:41:46 +0100

Artiklen her er introen til min daværende blog, datamatik.blog, som jeg drev i 2025 mens jeg var datamatiker-lærer på KEA, Københavns Erhvervsakademi. Jeg tror måske jeg følte, jeg skulle forsvare at gå ned i løn og undervise på en kort videregående uddannelse.

Hej, jeg hedder Simon Shine, og jeg er en programmør født i 1985.

Jeg er også datamatiker-lærer på Erhvervsakademi København, fusionen mellem KEA og Cphbusiness som foregår her i 2025. Det er det største erhvervsakademi i Danmark. Jeg underviser i programmering og grundlæggende datalogi. Datamatiker er en 2,5-årig uddannelse med fokus på praktisk softwareudvikling. Jeg har været lærer der siden 2024. Før det tilbragte jeg mange år som undervisningsassistent på DIKU, Datalogisk Institut på Københavns Universitet, i stedet for at færdiggøre min uddannelse.

Om at blive rekursivt beruset med monoider

Sun, 13 Apr 2025 20:26:14 +0000

Disclaimer: Jeg er ikke mixolog. Dette er ikke professionel cocktailrådgivning!

Lyt i stedet: Podcast-afsnittet Haskell Weekly, Episode 26: Recursive Monoids.

Jeg læste engang, at man kunne modellere cocktails som monoider.

Det er et virkelig cool eksempel på funktionsorienteret modellering, som jeg gerne vil uddybe.

Først vil jeg demonstrere min uvidenhed og antage, at cocktailopskrifter er frie kommutative monoider over ingredienser:

Rækkefølgen hvori du tilføjer ingredienser har ingen betydning,
Hvis du kombinerer to cocktails, får du en anden cocktail,
Identitetscocktailen er den tomme cocktail uden ingredienser.

Dernæst vil jeg genfortælle en rekursiv cocktailopskrift, som jeg lærte om da jeg studerede datalogi.

Deploying a Rust web service on NixOS

Thu, 10 Apr 2025 11:31:54 +0000

I have a small Rust web service that serves HTTP on a port.

In order to run this in production, I want to

compile and install the web service binary on my server,
run it as a non-interactive user on a local port, and
expose the web service via an nginx reverse proxy

I will use a subdirectory of an already existing website, such as a Hugo static website.

A web service package derivation ¶

In order to build and package my project, I need to express a derivation.

Corne Split Keyboard Layout

Thu, 10 Apr 2025 08:57:09 +0000

I min barsel har jeg leget med et Corne v3 split keyboard, som jeg har købt på Taobao.

Det har 46 taster. Til sammenligning har min MacBook Pro med dansk layout 79 taster.

De knapper man mister vs. de knapper man får ¶

Man fristes til at fokusere på alle de knapper, man mister, frem for alle de knapper man får.

Især får man noget som alle tastaturer burde have: Ekstra knapper til tommelfingrene.

Packaging open-source projects with Nix

Mon, 07 Apr 2025 13:33:50 +0000

Introduction ¶

Problem: You run NixOS and want to download, compile and run some project on GitHub.

While most projects come with a README that tell you what packages to apt-get install if you run Ubuntu, they often don’t explain how to do the same on NixOS. This makes NixOS a harder Linux distro to use, because you need to be able to write Nix to run certain programs.

Hvad er "datamatik"?

Thu, 06 Mar 2025 23:58:01 +0100

datamatik, sb. (læren om) hensigtsmæssig behandling af data på edb-systemer [1966]

I dette første blogindlæg vil jeg sætte spørgsmålstegn ved bloggens navn.

Ordet “datamat” er en anakronisme: Opfundet i 1960’erne, da referencer til computere blev almindelige, og dansk populærkultur endnu ikke var opslugt af amerikansk engelsk. Det hedder “computer” nu, med mindre man har boet i kælderen på datalogisk institut.

Mens “datamat” er en sjov, gammeldags måde at beskrive sin computer på, er “datamatik” gået så meget af mode, at ingen bruger ordet, selv ikke for sjov. Datamatik’s fætter, “datalogi”, bruges af universitetsuddannelsen.

Setting up a Hugo static website with Nix

Mon, 24 Feb 2025 00:00:00 +0000

tl;dr ¶

See the tutorial code at https://github.com/nix-tools/nix-hugo

This tutorial…

Provides a development environment for running interactively
Provides a derivation for deployment on NixOS
Deploys to a webserver using just, ssh and scp

Step 0: Prerequisites ¶

You have Nix installed (required)
You have flakes enabled (required)
You have nix-direnv enabled (recommended)

Step 1: A development environment ¶

A dev shell is provided as a file called shell.nix:

Minimal Sufficient Groups

Sun, 02 Feb 2025 16:11:57 +0100

This is a programming interview question I got some years ago.

The problem ¶

When a certain company creates a new project, they enumerate the skills required for the project, and they form a project group where each of those skills are represented. Project groups have two rules:

Sufficient: Each required skill should be represented with at least one group member.
Minimal: Each group member should be essential, so removing any one member from the team should make the rest of the team no longer sufficient. Members can have overlapping skills, as long as everyone is essential in some capacity.

Why you should be careful with the Default trait/typeclass

Tue, 25 Oct 2022 16:13:28 +0100

You heard it. Default considered harmful!

tl;dr:

Default has no defining or testable properties
Default is not even guaranteed to be the same between runs
Defaults are context-specific: One type does not always have one default
Defaults are useful for transitive derivation, but context drift risks causing errors
Your Default instance is dangerous if you cannot carelessly replace it with any other value
Haskellers: Use the default* pattern instead
Rustaceans: Use Default with care and respect

`Default` in Rust and in Haskell ¶

Rust’s standard library has a Default trait that lets you provide a default value for any type:

A review of JSON Schema libraries for Haskell

Sun, 10 Apr 2022 16:14:27 +0100

Update 2025-07-26: This review is obsolete. Consider using AutoDoCodec.

JSON Schema is a JSON format that describes JSON formats.

It is mostly used to validate that a JSON value has the correct format, but it can also be used to generate random values that fit a schema definition. This may be useful for testing.

The latest version of JSON Schema is called “Draft 2020-12”, and the way to define a schema using this draft:

Has-style traits in Rust

Thu, 31 Mar 2022 16:16:50 +0100

I recently experimented with Rust trait instances on bounded generic types. This is the fancy way of saying impl<T: SomeTrait> AnotherTrait for T, i.e. any number of AnotherTrait instances created as a result of the presence of a SomeTrait instance.

Here is a similar but slightly more complicated example of that.

I have a number of things that all have a common base set of properties. I’d like for the common base set to be described with one trait, and all the ways in which the things are different to be described with another trait that may depend on the common things.

Writing code on doors

Tue, 29 Mar 2022 16:20:20 +0100

Did you ever see pseudo-code in job ads that looked like

if ($you.like_to_work($here) === True) {
 $job.apply();
}

and thought to yourself a bunch of things about variables not being in scope, and whether their actual code has utterly specific library functions? Did you ever try to design an authentic piece of pseudo-code and found how difficult it is to remain concise enough to get the point across, but also idiomatic enough to even care expressing this as code?

A Terraform DNS moment

Sat, 29 Jan 2022 16:21:50 +0100

tl;dr: I stuck the IPv4 and IPv6 address of my VPS into an SPF record field text.

I manage my DNS with Terraform. This means that past domain registration and changing nameservers, I manage DNS with the Terraform command-line interface and git. Today I experienced a moment of happiness when two pieces of personal infrastructure fitted.

The scenario: I’m setting up a send-only email server on a VPS for Gitea notifications. If this works out well, I will consider using it for a newsletter. Doing this on a commercial cloud VPS means that my server’s IPv4 address gets a bad score. One way to improve on that score is to add an SPF record.

Fiat Currencies Are Not One Thing

Fri, 10 Dec 2021 16:25:17 +0100

When we discuss the potential future usefulness of cryptocurrency, we compare against existing use-cases of fiat currency. Given a particular fiat currency – its name, its unit, and its exchange rate – our language enforces us to think of it as one currency.

This is a misconception that will lead to poorly designed alternatives with strong, but unaligned monetary properties at the wrong time. If we don’t think of a fiat currency as many currencies within one brand, then any attempt to re-model those use-cases with cryptocurrencies will come short of being practically useful.

Jq Hack 2 Curling the Right Binary on Github

Sat, 04 Dec 2021 16:26:20 +0100

Something I see quite often is a combination of grep and cut to extract URLs from JSON objects, for example:

$ curl -s https://api.github.com/repos/go-gitea/gitea/releases/latest \
 | grep browser_download_url \
 | cut -d '"' -f 4 \
 | grep '\linux-amd64$' \
 | wget -i -

This is neat because grep and cut are almost always available on Linux. And the approach works for all kinds of plaintext documents with a semi-expectable format, not just JSON, so the approach is worth keeping around. But for JSON in particular, this approach is unnecessarily fragile because of a few unreasonable assumptions:

Why rewriting version control history matters

Thu, 25 Nov 2021 16:27:18 +0100

Life is understood backwards, but must be lived forwards.

— Søren Kierkegaard

A summary on good commit messages ¶

Because version control systems, like git and GitHub, are at the heart of modern software development workflows, most workflows are heavily impacted by and defined in terms of how to operate and cooperate around version control.

Making use of issue trackers, pull requests, review tools, and commit messages, there are many right ways to distribute the communication that revolves around a change in source code. So…

Efficient Probability Testing

Tue, 21 Sep 2021 16:02:16 +0100

I want share two open-source contributions that were lost on their original audience, yet has taught me something about testing functions that use randomness.

I once had a task at work where a website displayed a banner ad from a selection of banner ads with some internal logic related to categories, weights and fallback banners. Since the result was random, how do you test the selection algorithm?

In the concrete task, I ended up decoupling the rand() call from the logic that builds the probability distribution, so that it could be tested deterministically. But below are some other approaches that I really like.

Baking metaphors in the imperative/declarative programming debate

Fri, 17 Sep 2021 16:33:16 +0100

tl;dr:

Cake recipes contain both declarative and imperative parts.
Declarative syntax is better for describing the problem domain.
Each programming paradigm has its place in software architecture.

Baking metaphors ¶

An argument is floating around that imperative programming is easier, more natural for humans for the same reason that baking recipes are imperative. Here is a pie recipe for example:

Typical baking recipes contain three parts:

Extracting the KUBECONFIG for a DigitalOcean Kubernetes cluster from a Terraform .tfstate

Sat, 10 Jul 2021 16:34:07 +0100

When provisioning a Kubernetes cluster from DigitalOcean with Terraform, the .tfstate contains a field called raw_config that authenticates kubectl. It can elegantly be extracted with jq:

$ jq -r '.resources[]
 | select(.type == "digitalocean_kubernetes_cluster")
 | .instances[].attributes.kube_config[].raw_config' \ 
 terraform.tfstate
apiVersion: v1
kind: Config
clusters:
- cluster:
 certificate-authority-data: ...
 server: https://...k8s.ondigitalocean.com
 name: your-cluster-name
contexts:
- context:
 cluster: your-cluster-name
 user: your-cluster-name-admin
 name: your-cluster-name
current-context: your-cluster-name
users:
- name: your-cluster-name-admin
 user:
 token: ...

If you provision multiple clusters, you cannot simply pipe the multiple kind: Configs, but otherwhise, this output can be dumped straight into ~/.kube/config. Otherwise, you may want to dump it to a specific file before you run

Getting recursively drunk with monoids

Sun, 04 Oct 2020 16:05:52 +0100

Disclaimer: I am not a mixologist. This is not professional cocktail advice!

Sam Horvath-Hunt blogged about modelling cocktails as monoids. This is a really cool example of FP modelling that I want to expand on. (Lennart Kolmodin once wrote that the dance steps of Tango form a monoid.)

First, I will demonstrate my ignorance by assuming that cocktail recipes are free commutative monoids over ingredients:

The order in which you add ingredients does not matter,
If you add two cocktails together, you get another cocktail,
The identity cocktail is the empty cocktail with no ingredients in it.

Second, I want to up the ante with a recursive cocktail recipe.

Jq Hack 1 Colored Less

Mon, 28 Sep 2020 16:37:28 +0100

Sometimes you want to look at a lot of JSON output in the terminal. Sometimes this JSON output is nicely formatted, and often it is compacted by a REST endpoint. Examples of verbose output:

$ kubectl get nodes -o json
$ curl 'https://hacker-news.firebaseio.com/v0/topstories.json'

In both cases you can pipe the output through:

$ ... | jq -C . | less -R

The -C parameter to jq forces colors. When omitted, jq will only print colors when an interactive terminal is detected. Piping further into less removes the colors. Additionally, we have to tell less to interpolate those ANSI colors so that we don’t see a bunch of ESC[1;39m codes.

What if, when you grep, it goes into pager mode when results don't fit one screen?

Mon, 28 Sep 2020 16:35:50 +0100

There is a general theme in GNU shell commands that if a program is able to deliver ANSI-colored output, it only does so in interactive shells, or if explicitly instructed. If you want to pipe output but have colors remain for later printing, you sometimes have to augment both the sending and the receiving command.

Of the many alternatives to grep I use ack.

Given the following ~/.ackrc,

--flush
--pager
less -R -F

whenever I ack for something, it uses less as a pager, and if the search results fit a screen, it just prints them, including colors, and otherwise it enters less‘es interactive mode.

'Parse, don't validate' using ViewPatterns

Mon, 14 Sep 2020 16:38:44 +0100

In 2019, Alexis King wrote the Parse, don’t validate article that catches some of the essences of type-driven software development and design. Several books were released on this matter, including Alexander Granin’s Functional Design and Architecture, Sandy Maguire’s Thinking with Types: Type-Level Programming in Haskell, or even Edwin Brady’s Type-Driven Development with Idris for a book about the future.

This blog post is about where Alexis’ three words got me today.

I refactor a Haskell library that handles Ethereum VM opcodes, and in particular, the part of the library that lets you convert a Word8 into an Opcode.

Aggressive refactoring with Haskell

Sat, 12 Sep 2020 16:10:02 +0100

Here is a small example of aggressive refactoring in Haskell. I am in the codebase of a compiler for a small smart contract language, and I’m replacing occurrences of Integer with occurrences of Word256, since the code generator in question targets Ethereum which has 256-bit word sizes.

The compiler was originally made with Integer, and in some cases its very own Word256 which was made in a crude way. Integer because this is a very convenient data type in Haskell, even though arbitrary precision in this case is actually wrong, and the crude Word256 because it seemed simpler to make this type rather than import a library.

Hvorfor må jeg ikke eje en svensker?

Wed, 01 Jan 2003 10:14:48 +0100

Hvorfor må jeg ikke eje en svensker?

The following essay is an adaptation of a letter titled Why Can’t I Own a Canadian? by Dr. Laura Schlessinger in October 2002. This adaptation was originally posted by me to Ateistisk Selskab’s web forums but was later lost from here and copied without attribution.

An archived copy of the article exists on an older version of this website

Kære kristen

Tak fordi du gør så meget for at prædike Guds ord. Jeg har lært en masse ved at gå i kirke, og jeg prøver at dele min viden med så mange som muligt. Når nogen for eksempel forsvarer homoseksualitet henviser jeg dem bare til 3. Mosebog kapitel 18 vers 22 der siger at du ikke må have samleje med en mand, som man har samleje med en kvinde. Det er en vederstyggelighed. Diskussion slut. Jeg har dog nogle spørgsmål om nogle andre love og hvordan man bør følge dem:

Mon, 01 Jan 0001 00:00:00 +0000

Welcome! ¶

I’m Simon.

I’m a computer programmer from Copenhagen, Denmark. I grew up on the island of Amager, and that’s where I still live. I have been coding since 1997, and I studied Computer Science at DIKU in the period 2008-2017. I have been absolutely fascinated by programming, terminal systems and computer networks ever since.

My first website from the late 1990s, long lost, contained fantasy cartography produced by The GIMP. Unfortunately those pages never left my home computer, and they were only available when my 56k modem was connected, via constantly changing IP addresses. I didn’t get a lot of views.

Mon, 01 Jan 0001 00:00:00 +0000

Contact ¶

You can reach me at simon@simonshine.dk.

Mon, 01 Jan 0001 00:00:00 +0000

Retro articles and links ¶

My website was always more of a link collection than a blog.

Since my website has since got several blogs, my link collection has become secondary.

Keeping with nostalgia, this collection has accumulated over 20 years and contains timeless pieces.

Parse, Don't Validate 2019, by Alexis King

Discovering Assumptions 1996, by Paul Niquette

Eigenmorality 2014, by Scott Aaronson

The Theory of Interstellar Trade 1978, by Paul Krugman

Who can name the bigger number? 2005, by Scott Aaronson

The man who loved only numbers 2018, by Paul Hoffman

A pair of dice that never roll 7 2004, by Simon Tatham

Making regex faster with Thompson NFAs 2007, by Russ Cox

The Alice and Bob After Dinner Speech 1984, by John Gordon

OpenBSD on the Dell Inspiron 8600 2004

Simon Shine

Agentic Playwright testing with Nix

tl;dr ¶

The application under test ¶

Hashpinning

Git config-based hooks with hk-nix

Synergies ¶

Using sops-nix templates to support passwords in legacy configs

The problem: irssi keeps password authentication in the config file ¶

Authenticating Tailscale non-interactively via API with sops-nix

A dendritic sops-nix flake template

1. A dendritic flake ¶

Dendritic flake partitions

tl;dr ¶

Managing Secrets privately with sops-nix

tl;dr ¶

Motivation ¶

Is the novelty budget dead?

NixOS runners on Forgejo Actions

Three levels of Nix

1. Just a shell.nix ¶

Vendored theme workflow after one month

Using lefthook and treefmt with Nix

Dendritic Nix With nixos-shell

tl;dr ¶

Table of contents ¶

Motivation (some ranting) ¶

Learning about AutoAgents by benchmarking Claude Code vs. OpenCode

tl;dr ¶

Alternative to OpenClaw in Rust ¶

nixos-anywhere on Hetzner

tl;dr: ¶

Vaultwarden on NixOS

A git primer for future colleagues

Vendoring Hugo themes with Nix and Git

tl;dr: ¶

Context ¶

Speeding up CI more with Nix

Speeding up CI with Nix

zsh config for NixOS + nix-darwin

tl;dr: ¶

Why ¶

Maintaining libraries with Claude

tl;dr ¶

The goal ¶

Using NixOS for Exam Computers

Kubernetes-valgfaget der ikke blev til noget

End of teaching retrospective

Funktionel datamodellering

Animating the NixOS logo

Design Patterns i popkultur

JSON Schema in Haskell using AutoDoCodec

Just: En bedre Makefile

Problemet med Makefiles til projektkommandoer ¶

Claude Code og git-bug: Når AI har brug for issue tracking

Hvordan understøtter man europæisk cloud i undervisningen?

tl;dr (den ultra korte version): ¶

En opsummering af artiklen nedenfor: ¶

Hvordan forstås suverænitet? ¶

Hvorfor blev jeg datamatiker-lærer?

Om at blive rekursivt beruset med monoider

Deploying a Rust web service on NixOS

A web service package derivation ¶

Corne Split Keyboard Layout

De knapper man mister vs. de knapper man får ¶

Packaging open-source projects with Nix

Introduction ¶

Hvad er "datamatik"?

Setting up a Hugo static website with Nix

tl;dr ¶

Step 0: Prerequisites ¶

Step 1: A development environment ¶

Minimal Sufficient Groups

The problem ¶

Why you should be careful with the Default trait/typeclass

Default in Rust and in Haskell ¶

A review of JSON Schema libraries for Haskell

Has-style traits in Rust

Writing code on doors

A Terraform DNS moment

1. Just a `shell.nix` ¶

`Default` in Rust and in Haskell ¶