You do not need NixOS on the desktop
aruarian.dance
aruarian.dance
There are many distros for people who don't want to care - Ubuntu, Mint, Elementary OS, MX Linux, etc. I don't see how NixOS not being one of them is NixOS's problem.
> "I believe that the core idea of NixOS is fundamentally opposed to the idea of what the average person wants in their desktop."
What an average person wants in their desktop is Windows - not Linux and certainly not some obscure independent distro. And this is still not a problem of that distro or Linux.
It seems like the author is talking to people who might consider using NixOS for desktops, not towards NixOS itself. Unless I missed something in the article, the author is not saying this is a NixOS problem, but a "I made the wrong choice for me" problem, and now they're sharing the experience of reaching that conclusion.
Don't get me wrong, I (like many) have a love-hate relationship with NixOS, where I use it for all my servers both remote and at home, but my desktop/laptops remain on Arch Linux because I too don't fit it fitting for desktop usage. But I wouldn't argue against people who want to/not want to use it for desktop use, cool that it works or not for them.
> What an average person wants in their desktop is Windows
I think based on the context, the author is talking about the average developer really, not the typical end-user. They do say "someone who wants to use a computer like a regular person to do regular work" which might confuse people, but they really are talking about developers, as you can tell by the rest of the article.
Average developer wants Windows (with WSL) or Mac. Still not Linux. ¯\_(ツ)_/¯
According to the latest Stack Overflow Developer Survey (https://survey.stackoverflow.co/2024/technology#most-popular..., which also is biased admittedly), it seems like for professional use, ~48% use Windows, ~32% macOS and 28% Ubuntu. ~17% also use WSL, which is basically Linux in Windows, so I guess you could say ~45% wants to (or at least, does) use Linux for work.
Seems not so clear cut as you seem to think it is.
With that I mean, the only reason I use WSL is for running containers locally.
All the regular software, outside projects that require container based deployments, is straight Windows software.
Yeah, you're probably not alone in using Linux specifically for containers since Windows cannot run them. Just like I'm probably not alone in using Windows solely for the purpose of running Ableton, as Linux cannot run it :)
https://learn.microsoft.com/en-us/virtualization/windowscont...
Just because GNU/Linux has the biggest mindshare in containers, doesn't mean it is special in any regard.
In fact, my first experience with a container like deployment model was with HP-UX Vaults in 1999.
Ps, also did "containers" before it was cool with Jails :) Not as early as 1999 though so I guess you win this round pjmlp.
I don't dual boot since around 2010.
The average person doesn't even want Windows. They want to click a button and not be bothered with the implementation details.
That is why mobile/tablet is such a popular form of compute these days. People don't even have to learn the basics of interfacing with a file system most of the time. Want to look at pictures you've taken? You can be oblivious to the fact that your camera app puts picture files in a specific directory and embeds a date code in the file name, the photo viewer app takes care of that for you.
10 years ago Jarvis became a part of Vision in Age of Ultron and effectively no longer exists in the MCU. A variety of new AI assistants with new names were made in later movies.
None of the new ones became as recognizable, and I guess Jarvis is also falling into obscurity.
(And as a student I Saturday-jobbed at a cinema, so there's a certain era for which I've seen at the very least many odd scenes out of order for essentially all widely released films...)
I said this one time on hacker news and a nixos fan told me that people like me should basically fuck off from the community because they don’t want people like me.
I used nixos for about a year for work so I definitely see the benefits of this but this ass hole attitude just made me sick of it. Like why can’t I have both? Is it impossible to have both? It may not exist yet but there’s a need here that would benefit everyone if it were filled.
Needless to say that thanks to that person I now really just dislike the nixos community. Rude and no flexibility in changing. Inevitably some guy might respond to this to apologize on behalf of the community and I appreciate that… but at the same time I think most members of the community truly have this attitude of keeping things hard and challenging on nixos and they don’t want things to change at all.
In addition to this.
Everything the guy mentioned that makes nixos hard for desktop can apply to servers as well. It’s not as if everything magically gets better.
* less service interruptions by using chromeOS over a cell carriers network instead of a failing internal network
* improved productivity from fewer OS configuration and update problems by switching to Macs
* Lower costs across the board due to better or less expensive hardware with Chrome and Mac
* Higher dev productivity in a small dev shop by requiring devs to use the exact same os they deploy code to (it forced the devs to learn)
If you want to use it as your desktop, nobody’s stopping you. They’re making the case why they think it’s unsuitable.
I’d been curious about it many times in the past, but found the learning curve to be prohibitive the first several times I tried it. I’m fairly convinced that many of the problems people experience stem from the poor documentation and over-enthusiastic community extolling the virtues of Flakes and Home Manager.
I understand their value and why people like using them, but for a beginner, those capabilities just layer complexity on top of something that already feels unfamiliar, and make it impossible to figure out which thing isn’t working when things go wrong. When I went to a very Vanilla NixOS configuration as a starting point, everything clicked and I was able to build up a solid desktop environment incrementally.
As a tinkerer, the killer feature for me is the ease of experimenting with packages/whole configurations and then reverting back to my known good config. Type `nix-shell -p <package>` and the package is ready for use, and totally gone when I exit the shell.
I don’t think need is the right way to describe my relationship with NixOS as a desktop. But decades of experience dealing with the aftermath of installing/experimenting/tweaking my environment in traditional distros sure makes me appreciate how much easier it can be.
> I believe that the core idea of NixOS is fundamentally opposed to the idea of what the average person wants in their desktop.
NixOS on the desktop isn’t targeted at the average person, or probably even the average Linux desktop user. If it’s causing more pain than it’s solving, it’s probably not the right choice. But I think that will be a very individual/personal calculation.
I started on a work laptop, and moved the configuration to my home desktop; it instantly worked the same way. I copied the configuration to my gaming laptop, and when my desktop broke down, I could continue to use my gaming laptop as if it were the same computer.
Being able to copy-paste a few snippets and have Steam working was amazing.
> If it’s causing more pain than it’s solving, it’s probably not the right choice.
It certainly is causing more pain than it's solving.
But NixOS has a way of converting all your regular problems into Nix problems.
Here I am, trying to get back into Kubernetes.
But all I can focus on is how to install Kubernetes.
My middle ground approach has been to use containers and fall back to a more traditional Linux approach within those containers when I don’t want to be bothered (or just don’t need) to figure out the Nix Way.
If it’s something that’s core to my daily setup, I’ll still invest the time to bake it into my NixOS config, but using fallbacks like this has been a good middle ground for me that allows me to take advantage of the best parts of NixOS without adding undue pain when spinning up experiments.
I keep telling myself that of these days I need to play with the Nix way, but we’ll see.
virtualisation.docker.enable = true;
users.extraGroups.docker.members = [ "green" ];
Update once per year.
> brew, distrobox, flatpak
You can have that stuff on Nix if you need it.
Nix is bad because the language is very limited and extremely hard to debug and the module system is fragile and obscure (just try to write a recursion-safe mkMerge and prove it won't magically drop some parts of your config because the values in the module system are always magic attrsets, not values), but, unfortunately, there is nothing better around.
In other words, the old adage still rings true. The beauty of Nix is that you can do anything. The ugliness of nix is that you can do anything
It turned out to be ansible... which is a pure python tool. Beats me.
There's also some cases where it wants to build Android Studio from source. I've just removed it and now run it in a VM.
I'm sure I'm doing it wrong, and I'm sure nixos-rebuild has a reason to build things from source, and maybe I'm not RTFMing hard enough. But good god, I just need my system to update.
I guess our system's nixpkgs moves forward while the nixpkgs in our project flake.nix stays the same. So eventually the nixpkgs of the project is garbage collected, because it isn't pinned by the system, and it needs to redownload and rebuild?
I don't know.
I try to `nix flake update` my system configuration every 2-3 months.
Arguably, running Arch and not updating frequently is a pain, too.
You should manually create a garbage collection root (that pesky 'result' symlink, --out-link)
There are introspection tools for that.
> Build from source
That's not an issue unless you stay on master or use config flags (there is just a handful of them, but they are imortant, eg cudaSupport=true can easily give you 5-6 hours of build time, rocmSupport can easily give you a couple of days)
You can use hydra to observe the ebb and flow of broken packages getting fixed and pick good git hashes to pin to. Once I figured that out I was very happy and had a good time glancing through hydra to monitor this. Now I can make informed decisions about what hash I build from and still keep unstable train, upgrading more often then I was before.
Everything is a battle and nix battle is bit less hopeless than the imperative distro battle, but again, nix is bad but noone wants to build a better thingy.
I've seen countless attempts and I'm trying to make my own build system (not targeting linux distros though). The problem is extremely hard. Essentially you have to build a general purpose PL which is better (more introspectable/debuggable/extensible) than other GPPLs.
Nonetheless, in case those assumptions are off - GNU Guix exists, and is written in and extended in (Guile) Scheme. Guile is general purpose, has the excellent Andy Wingo powering a lot of improvements to the language the last years, and some people are very fond of it. Perhaps you'd be interested in investigating, if Nix's DSL wasn't to your liking.
Yes, the Scheme programming language doesn't solve any those problems I mention.
> improvements to the language
Despite of the language, if we speak about Linux distros, I need all those pesky blobs to get my desktop running. And, by the way, is there anything alike to flakes, so I can have an actually reproducible build, not a mess of channels?
You wrote a comment implying what's wrong with Nix is the language, and stating that there's unfortunately "nothing better" out there.
It is true in this universe that there is a software project called GNU Guix, which is 13 years old, whose literal origin idea is - what if we take the general idea behind Nix, but write it all afresh in a fully featured programming language, and make it extensible and configurable with that language.
There are a host of reasons you might not want to get into that personally, as I stated - but it really does exist, regardless of your personal feelings?
1) Nix language is ugly but it's pure. Scheme is not. To my taste it's even harder to reason about/debug/introspect because of lack of purity.
2) Guix itself lacks any nonfree software/blobs, which is essential to run a real desktop. That is a showstopper for newcomers and in my opinion that's what prevent the community growth, so there are much less packages, docs, etc.
2.1) There were channels to add nonfree blobs, but channels make system less reproducible and they were unofficial. Maybe I'm not up to date with the Guix developments, but as far as I'm aware there is no replacement for Flakes and a system with multiple channels is not reproducible and easy to break.
> Nix is bad because the language is very limited and extremely hard to debug and the module system is fragile and obscure
-- Limited :: here's a collection of Guile software https://sph.mn/foreign/guile-software.html. It's not Python, but it is a general purpose language used by lots of people for serious software work, with a nice library of lovely Scheme books for learning, too
-- Hard to debug :: Guile has built-in debugging support, and there are interfaces like Geiser in Emacs with extra tools, and also support for interacting with GDB
-- Module system :: Guile's module system was first implemented in 1996, I learned from browsing there. Anyway https://www.gnu.org/software/guile/manual/html_node/Modules....
So yes, if you don't like Lisps, or don't have time to learn new things, or don't like something else about Guix, by all means, steer clear. But the statement: "there's no Nix but with a general purpose, powerful programming language" is, I think, demonstrably false.
I don't like impure languages with weak type systems. Ironically, I mostly use Scala, which is an impure language but it has a very powerful type system, unlike any other.
You can even do this multiple times (don’t do it too many times though as it will increase evals times) and doing this you can run all kinds of different software at different versions.
However, after these last few years, I have a few gripes that I'd like to check with a guix user to see if they're represented in guix.
1) I can't easily pin a software I'm using to a specific version, it seems like everything has to be from a specific hash of nixpkgs
2) Relatedly, I can't upgrade firefox without upgrading every single piece of software installed on my computer
3) Dynamic linking doesn't work. I can get around it with steam-run, but that's hacky and I worry it won't be supported forever
I was more a happy Guix user and computing beginner, rather than an expert, I didn't do anything overly fancy. I was starting to dabble with more advanced topics, but had to leave that system unfortunately for unrelated reasons. Looking forward to a glorious return though, I must say, as soon as I can manage it.
Don't hesitate to try the official documentation with your problems, it's excellent. No answer there, try a full description of your issues on the mailing list, they're a great bunch too.
Search the help-guix archive here before emailing them -- https://lists.gnu.org/archive/html/help-guix/
Details on joining the help-guix mailing list, and/or where to send questions, here -- https://lists.gnu.org/mailman/listinfo/help-guix
> You can have that stuff on Nix if you need it.
In fact, I'd openly recommend such a solution: NixOS base, most applications via flatpak, and dev work in distrobox. Gives you 80% of the benefits for 20% of the work.
(Although I will also happily agree that Fedora's Silverblue or OpenSUSE MicroOS as the base is also pretty good; I think that's just a trade if you need to actually tweak the base layer)
No, you don't need NixOS on the desktop, but for certain people--like me!--it sure is nice. For example, I have this preoccupation with remembering how I configured something. I'd developed note-taking systems to remember how I'd set something up for the inevitable reinstallation (say a new computer)? Now? Download a few text files from github, type a command, and 10 minutes later I'm back in business.
It was a tough road to get comfortable, though I tell people that's because NixOS should be treated like an entirely different OS rather than as a distribution. But now? It's great. It solves a real problem for me. I definitely don't think it's for everyone, but this post draws an overly general conclusion.
Flatpak guarantees everything will work in most cases, and for other cases there's AppImage. Just need to get most devs to distribute AppImages. BoxBuddy with distrobox will solve _all_ edge cases where someone says "X works with Y in Z on my machine" so you replicate their machine in distrobox.
I know this is trading program size with convenience, but that's what Windows and macOS does too. It is better to be on some immutable linux distro rather than Windows in my opinion. We don't have to force the average person who just wants their computer to work to install (extreme example) Gentoo or whatever.
In my case I was using Flutter for an app and when a new version was released the version supported by the current NixOS version was old. So I had to search for a flake that supported the new version but it was abandoned some versions ago. So finally I had to search for a how to create a flake so the latest version would work.
Cue some hours of work until I had it working.
Compare this to Ubuntu where I execute "flutter upgrade" and I have it working in a minute.
For my dev machine if I need some kind of reproducibility I have Docker at hand. It's a 100% perfect value solution for this? No, but the cost to pay is much lower than Nix.
> However, this means that your projects end up populated with Nix files, which is particularly annoying when you want to submit PRs to upstream projects
This is a legitimate gripe, but it does have a mitigation, which is to add those files to your global gitignore. The trade-off you make here is that you have to explicitly add the files to repos that actually do need them, but that's a one and done cost.
> I'm now responsible for configuring this entire system. I am also responsible for updating this system.
On the contrary, I have felt much less responsible for configuring and updating my whole system since switching to NixOS. Most of my system is just whatever the upstream channels are doing. My whole system config is maybe 300 lines long, and most of that is a list of the programs I want installed, which is something that I've always wished I could have on other distros but can't.
On other distros I have felt much more responsible for controlling my whole system because random crap would break all the time and I had to learn how it was configured. That has never happened to me on NixOS.
Also, containers are totally available on NixOS and I use them all the time. I've avoided flatpak so far not because it's not available but because I want to try to do things the NixOS way and I haven't felt the need to move away from fully declarative.
None of which is to say that NixOS is right for you or anyone in particular, but so far my experience has been that there is a very steep initial learning curve that you do get over. Maybe someday I'll give Fedora Bluefin a shot, but in the meantime I'm definitely not regretting leaving behind the traditional distro model.
The installer is super easy to use; full disk encryption is baked in if you want it, the default desktop is fine, and the overwhelming majority of my config is just the apps I want installed by default.
I love this because it makes my desktop or laptop totally commodified: if I break my laptop I can get another one out of the closet and:
-- install nixos in 10 minutes
-- copy a couple of stanzas out of my nix config file into the new system one
-- untar a backup /home/user
-- `nix rebuild switch`
-- drink a beer
No installing a bunch of dev environment stuff. No greping internet for 'gpg pin-entry' or 'how to install docker' because I had to write that into a config in the first place. Basically I like nixos because I'm a lazy fucker and it makes me front end load the work so I only have to do it once.
Also, these days if I have trouble writing a nix thing I can usually just vibecode my way out of the problem. ¯\_(ツ)_/¯
Note: I’m not a Nix user - primarily Fedora and Debian, though I’ve used Bluefin a lot and used to use PopOS as a gaming desktop. (Rarely have any time for gaming these days…)
I ran nixos for a few years but I stopped when I started experimenting with AI. Dependencies were too out of date and running random small projects is riddled with extra pain, compared to other systems.
I also had problems during upgrades, caused by incompatible features and configurations. Instead of relying on the project documentation I had to rely on nix documentation. I have massive respect for nix maintainers (and I have even contributed) but maintaining any software for everyone is a monumental task. Distros maintainer have it hard enough.
Arch with AUR ended up being a simpler experience.
And I run rsync on my home directory to my local file server.
That's really all I need to recover. But I don't think I have had to recover a system in a very, very long time.
NixOS is also often a terrible choice for servers. I once had the misfortune to be part of a multi-user server running on NixOS. The idea made sense: users could all have their own versions of software installed. In practice, it was a disaster: NixOS seems to fundamentally assume that all users have root access.
Many ordinary processes ended up involving administrator involvement. Want to change your shell? Contact the administrator. Password? Contact the administrator. Add or remove an SSH key? Contact the administrator; hopefully you don’t do anything like use per-device keys. Run a user service? Contact the administrator, who might also need to do all the configuration for you. Numerous options that would be user-configurable in other distributions ended up in configuration.nix. And looking or asking for help online was an exercise in frustration: the idea that I wasn’t able to use sudo or edit configuration.nix seemed completely foreign.
Fortunately, we were able to switch everything over to Debian. This had the advantage of both actually working for normal user accounts, and still letting people use Nix, arguably with fewer problems.
I do expect someone may reply with some poorly (or un-) documented approach to handling these problems, perhaps using unstable and experimental features, but that tendency in itself was a considerable frustration in using Nix.
For complex packages like Steam, it's both possible and recommended to use FHS-compatible containers on NixOS. Still, I've seen people say things like, "All I do is set up containers-why not just use Docker instead of NixOS?" The thing is, if you dig deeper, tools like Docker or Flatpak are actually less powerful than Nix when it comes to container management.
I've been toying with an idea: using filesystem access tracing to replace the current approach of using random hashes for isolation. This could allow an FHS-style layout while preserving many of the guarantees of the Nix model. It would dramatically improve compatibility out-of-the-box, enable capabilities that aren't possible today, and reduce network and disk usage-since files could be modified in-place instead of being remade or redownloaded.
It's on my backlog, though. Starting a new distro doesn't seem particularly rewarding at the moment.
However, alongside the desire to “not care,” there's a strong curiosity to explore new operating systems, architectures, or philosophies. Recently, my inner child prevailed, leading me to purchase a second-hand notebook. I installed FreeBSD and configured Sway on it. Now, it sits among my other notebooks, at least one of which is set up by my inner adult for daily work.
My current project to move from my personal metapac-managed Arch distro to one based on Obarun Linux, to get away from systemd. Trying to bolt declarative package management onto RPM doesn't sound like a good time, and I'd do this with Debian before trying it with Red Hat.
Please educate yourself.
> Why bother reproducing something
Yes-yes, a very good question.
> Reproducible builds, also known as deterministic compilation, is a process of compiling software which ensures the resulting binary code can be reproduced. Source code compiled using deterministic compilation will always output the same binary.[1][2][3]
An operating system isn't a binary executable. You're gonna need to come at me with more than RTFM my bro.
Everything is wrong with this statement but no, please educate yourself.
> An operating system isn't a binary executable
In many cases it exactly is a binary executable, but in some cases one may say that it's just a binary blob containing a filesystem image. If you can build with any level of reproducibility, you are doing great.
What you lack isn't education, it's experience. Experience to determine what is a good use of time and what doesn't matter. Experience to determine how to approach implementing ideas like reproducibility without shooting yourself in the foot, as the author did. Experience you gain by screwing it up and learning the lesson. It sounds to me like you screw stuff up and don't learn the lesson, leaving you to develop hard-headed idealistic positions like this one that you rely on in lieu of said experience.
I can, I just did it.
> But you're chasing after a moving target
Please educate yourself. Read about bootstrap, hermeticity and reproducibility. Nix has a byte-to-byte reproducible core and that is very important and valuable.
> byte for byte reproducible core
Whoop de doo.
What about the other 9000 parts?
Unless you substantiate it, this is not a constructive response.
(To take a stab at a useful answer: Being able to perfectly reproduce a system is actually quite helpful when changing things, because the easier it is to roll back the safer it is to roll forward.)
The Arch system is flexible and is designed to account for this. It offers something better than reproducibility imo.
Er, yeah, that's why NixOS, like many distros, keeps public archives of packages and their source (see the discussion at https://news.ycombinator.com/item?id=36261492 )
Edit: found it: https://codeberg.org/amjoseph/sixos
Too new to rely on, sadly. Maybe in a few years.
Also I value packagers as the first third party who might review upstream code instead of nobody on package systems that allow upstream to package directly for all distros.
We already knows the Windows classic package idea failure, no need to reproduce it with extra layer of complexity and wasted resources...
Do people really not selectively add files to their commits? Sometimes I don't even commit the whole of a file's changes, thanks to `git add -p`.
Didn't have the energy to try to set up gaming & nvidia drivers & GUI on nix. If valve/steam is going for arch then I'll just go that way too
I haven't really learned nix os language right now and I have only installed bare minimal software like a librewolf, prism launcher, kde, I used to be a hyprland user and I might switch back in some days.
The biggest advantage to me feels nix-shell which is cross platform
I installed obs-studio from nix-shell, built a video, then used another nix-shell for ffmpeg and converted video to mp4 and uploaded it to github.
This is my general use case of obs and I very rarely make recordings and so this idea that I can install/try out software without having to worry about anything while still having no xdg issues unlike flatpak is a godsend.
Did I mention I started using nix-shell for places I would've used docker, like stirling-pdf one off cases
Most of the software on my computer needs to be rarely accessed and I love the sanity that nix provides knowing that It won't make my update times faster/if-any unlike archlinux and dependency hell is a problem I truly despise.
I have looked for better alternatives, maybe spack comes to mind, but nix-shell is still crazy good. And I can also some day use a functional programming to make it even more automated.
FWIW the way we use nix where I work is quite a bit lighter touch than the setups being discussed here. There are kind of three tiers of Nix implementations
- I want Nix to manage my entire os (NixOS)
- I want Nix to manage my user shell and dotfiles (home manager)
- I want Nix to manage per repository shells (nix + direnv with a flake.nix in each repo, or an .envrc that pulls a shared flake and extends it)
We use the latter and find it to be a good mix of keeping nix configuration simple but also enabling per environment shells that are reproducible
With Ubuntu 24.04 and vagrant virtual machines, you could have even less hassle than with Bluefin.
At work it is all about Windows and macOS, with a mix of legacy server VMs, Kubernetes or serverless on the server, we increasingly rely less on classical UNIX or Windows server workloads.
At home, it is all about Windows and Android.
So unless it is unavoidable, I don't need it anywhere, it is another Gentoo, at least for me.
Well, what software to use at home is your choice, unlike what software to use at work.