I thought about posting notes on Syncthing but then I’d have to tangentially talk about NixOS. NixOS is my main Linux distribution atm but Nix/NixOS/Flakes are too hard to write about and I fear they’ve ventured too far into the realm of over–engineering.

What do I mean by over–engineering? An over–engineered tool is one where even the simplest use cases are non–obvious (to most people) and this can happen when it tries to do too many things with “specificity”. The overall concept is elegant though (explainable in lay terms) and can be applied in other contexts.

I feel sorta conflicted about my posted notes on nixops because shortly after that I discovered nixos-rebuild had remote deploys. Now nixos-rebuild itself is just a bash script and taking a peek inside is well, you know.. enlightening. There’s even a NixOS deployment tool written in just nix. I use nixos-rebuild…

nixos-rebuild switch \
  --target-host "nix@remote.host" \
  --build-host "localhost" \
  --no-build-nix

The nix packages collection has a neat library repository. A small nudge can make it completely standalone. Basically it allows experimenting quickly with configuration patterns for custom stuff without an expensive evaluation.

I might not remember correctly but lxc came out around 2008. Docker in 2013. The lxc commands had the trend of dashes in their name — lxc-attach, lxc-snapshot, lxc-copy and so forth. It had lots of boilerplate and a steep learning curve.

Then lxd arrived (in 2015?) to make everything user friendly. lxd is preferred. I still mostly use lxc out of habit.

After daily driving NixOS and the Nix pm for almost three years, it feels like it’ll be simplified by entities external to the project. It’s still in that academic phase (don’t do this/that) and needs software engineering asap.

That usually involves reducing boilerplate ruthlessly while generalizing/capturing fundamental uses cases (setting implicit best current practices). It’s reminiscent of lxc just before docker arrived for the masses.

btw this blog is served to you by NixOS :-)

Surely typing/saying the below Alpine command to install Firefox is way easier than who knows what multi–stage process other operating systems require.

apk add firefox

Google’s search box (rather tongue in cheek) is the most popular command line interface in use today. With NLP, installing and running firefox would just be a phrase equivalent that compiles to:

firefox

On NixOS, we can already do stuff like that. Of course, NixOS lacks a translation/encapsulation layer for imperative to declarative specification — so user friendliness suffers dramatically.

It’s kinda odd being able to remember what things were like before Docker appeared. Docker is the software developer’s take on creating a Linux distribution. It’s useful, but the indirection of containers and pinning can be deadly. Avoiding unnecessary indirection is advantageous.

I’m just now realizing that there might be a schism within the Nix/NixOS ecosystem/community on old versus new interfaces. If that’s remotely true, then somewhere, a great holy war is at play. Here’s an article summary of the old versus the new interface that I stumbled upon recently.

This blog is really really good for robust usage of Nix/NixOS. I stumble upon it every so often. Another excellent blog is “How to Learn Nix” which explores in excruciating detail the painful parts of nix and its documentation. Discovered that one recently.

What’s funny about the NixOS/GNU Guix design is that it tricks developers into writing their own system packages. That would never happen on other Linux distributions. I’m slowly favoring Guix though, since the new nix flake interface couples too tightly with git.