home blog log atom rss about contact

four years on NixOS

It’s another yearly instance of my NixOS journey (2024 instance). I meant to write it around 25.05 but completely forgot!

system maintenance

As usual I don’t remember what I did to my system over the past year, so I’m at the git log for /etc/nixos as it contains all the changes:

• follow hardware.opengl -> hardware.graphics rename
• follow hardware.opengl.driSupport{,32Bit} removal
• follow rename of gnome.adwaita-icon-theme
• follow hardware.pulseaudio -> services.pulseaudio rename
• follow okular -> kdePackages.okular rename
• drop deprecated i18n.supportedLocales
• follow networking.wireless.iwd.settings.General -> networking.wireless.iwd.settings.DriverQuirks rename
• follow services.postfix.config -> services.postfix.settings.main rename
• fix services.postfix.settings.main.mynetworks type (changed from string to array)

It is quite a bit more of renames than last year. I think locale changes actually broke my locales at runtime and I had to figure out what to change to get them back.

I did not have major package build failures that required any local changes.

This time I had the following non-trivial problems in upstream packages:

• duperemove would hang up on NoCOW files: https://github.com/markfasheh/duperemove/pull/376. I had to bisect the regression and fix it upstream. It was easy as I am somewhat familiar with duperemove implementation.
• nix started stripping too much in gcc-14 warning logs: https://github.com/NixOS/nix/pull/13109. It came up after nixpkgs switched to gcc-14. Took some time to figure out what is so special about gcc warnings. Ended up being quite easy.
• perf stated to hang up on my system: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=c21986d33d6beb269a35b38dcb8adaa5bd228527. It came up after I ran system-wide profiler to see what wine does to use 100% CPU on an ancient game. THe fix was trivial once looked at. I don’t remember what changed to expose the hangups. Maybe I just never did it before? Ended up being easy as well.
• mpv started failing GPU rendering due to libplacebo / shaderc incompatibility: https://code.videolan.org/videolan/libplacebo/-/issues/335. Most arcane bug of all: had to bisect the whole system to find the package first, then had to bisect down to the commit. Upstream eventually fixed it. But if nixpkgs had up-to-date shaderc we would not stumble on this bug. The only non-trivial bug from the whole list.

Community support

I still feel that NixOS community is a welcoming place for newcomers, experimenters and people who do grunt maintenance work. NixOS community now had elected their first Steering Committee who can help resolving high-level conflicts.

Some of the amusing things I did over the past year:

• nix language non-determinism in sort built-in
• stdenv fix to handle root directories that start with dash (-): https://github.com/NixOS/nixpkgs/pull/317106. diffoscope-269 release was a great stress test for nixpkgs bash code :)
• Found and fixed ~50 more eval failures in nixpkgs found by the hack. This hack was also the trigger that exposed sort non-determinism above.
• Fixed nixpkgs isMachO helper: https://github.com/NixOS/nixpkgs/pull/432097. Reading 4 bytes from the file in pure bash. How hard could it be?

Just like last year I managed to get about 800 commits into nixpkgs this year.

I stopped reading any Matrix channels completely and only skim through discourse and read github notifications.

Home server experience

I did not have to adapt anything for the past year. Things still Just Work.

Local experiments

I switched to helix editor and to chromium browser. Both were quite smooth transitions.

I continued gcc testing. This year it was gcc-15 branch. nixpkgs still manages to serve as a reasonable vehicle to find bugs. Just like last year I found about 50 compiler bugs. Did not manage to fix any myself.

Parting words

NixOS still works for me.

Give NixOS a go if you did not yet :)