Perhaps dumb questions inbound ;)

I use Arch because I’m strapped for time and my system is always moving.

  • 2 minutes to install something? AUR probably has it.

  • Ten minutes of free time to look for a software that fits a new need? Try random AUR things (auditing PKGBUILDs is just twenty seconds or so).

  • If I need a tiny patch, I’ll just add a sed or patch file to the PKGBUILD. (Super easy, you barely learn any syntax cuz it’s intuitive shell.)

  • make && make install/meson blahblah usually just works.

  • Wiki does the thinking for me if I need something special (e.g. hw video acceleration)

Buuuut update surprises can be a pain (e.g. Pipewire explodes Saturday evening) and declarative rollbackable immutability sounds really freakin’ AWESOME, so I’m considering NixOS for my new laptop (old one’s webcam broke). So I ask:

  • How much can I grok in a week?
    • I need to know Nixlang, right? I have a ton of dotfiles and random homemade cpp commands in ~/.local/bin that I use daily
  • How quick is it to make a derivation?
    • I make install a lot, do I need to declare that due to non-FHS? Can I boilerplate the whole thing with someone else’s make install and ctrl+c ctrl+v? How does genAI fare? (Lemmy hates word guess bots, I know)
  • How quick is it to install something new and random?
    • Do I just use nix-shell if I need something asap? Do I need to make a derivation for all my programs? e.g. do I need to declare a Hyprland plugin I’m test-running?
  • How long do you research a new package for?
    • On Gentoo I always looked up USE flags (NOO my time); on Arch I just audit the PKGBUILD and test-run it (20 seconds); on Ubuntu I had to find the relevant PPA (2 minutes). What’s it like for Nix?
  • Can you set up dev environments quickly or do you need to write a ton of configs?
    • I hear python can be annoying. Do C++/Android Studio have header file/etc. issues?
  • What maintenance ouchies do you run into? How long to rectify?
  • Do I need to finagle on my own to have /boot encrypted?
    • I boot via: unencrypted EFI grub asks for LUKS password -> decrypt /boot, which then has a keyfile -> decrypt and mount btrfs root partition. But lots of guides don’t do it this way

Thanks for bearing with me ദ്ദി(。•̀ヮ<)~✩‧₊

  • moonpiedumplings@programming.dev
    link
    fedilink
    English
    arrow-up
    3
    ·
    edit-2
    2 months ago

    So, I use Arch, but I don’t use the AUR at all. Instead, I use nixpkgs to get stuff (admittedly only like 3 packages) not in the Arch repos.

    The main reason for this is the quality of AUR packages. Although I don’t really fear a malicious package, I do remember hearing about a package that moved a users /bin to /opt during the install phase.

    Something like that is literally impossible with Nix, due to the way that applications aren’t really installed to the system. But, nixpkgs also requires some level of vetting the package quality, which is also nice.

    I also use nix for managing all my development environments. For example, my blog github repo, has a few nix files at it’s root, and you should just be able to type nix-shell in folder, and then you will get an identical environment to me.

    declarative rollbackable immutability sounds really freakin’ AWESOME

    I have BTRFS snapshots set up, and with grub-btrfs, I can even boot from them and revert to an older kernel (my /boot is stored on BTRFS).

    However, I have given up on NixOS, for many reasons. The documentation is very poor, and it’s more complexity than it’s worth, to make my whole OS reproducible, rather than just my development environments. In addition to that, their are also issues with running certain apps that expect to see a normal FIlesystem Hierarchy, which nix does not provide. Although you can work around this with stuff like steam-run or creating a fake FHS using nix, I would rather not play that game.

    But, considering I installed some stuff in an Ubuntu 22 distrobox recently, because that was what VScode and Unity official provide repos for, maybe this doesn’t really matter. You can probably use distrobox on Nixos, but I’ve seen issues about GPU acceleration with distrobox (and other non-nix apps) as well.

    EDIT: I lied, I use the chaotic aur for some things.

    • fool@programming.devOP
      link
      fedilink
      arrow-up
      2
      ·
      2 months ago

      Thanks for the input!

      I’m nervous about faking FHS as well, especially for specialized stuff. I don’t know much about steam-run or its caveats – so I can’t debug it (Maybe it turns out to be really simple and solid? Who knows…)

      Thanks for mentioning the gpu accel issues in distrobox – I was considering using containerization to fight off any FHS issues but it seems I can’t jump the gun. I’ll probably just tighten dev envs by trickling in nix-shell usage; multiple versions of a package at once is an issue I’d def love to solve (in a way that’s more than just dockerfile)

      Interesting that this is the third comment suggesting just using btrfs snapshots to resist Arch update experiences. I have root and home on two flat btrfs subvols so it shouldn’t be that hard to implement. (yeah yeah “What backup?” is bad)

      Seems like the simplest way out is those two smallish changes. Wish I could transcend into declarativity but the thread’s nix survivor ratio is grim

      • moonpiedumplings@programming.dev
        link
        fedilink
        English
        arrow-up
        4
        ·
        2 months ago

        Wish I could transcend into declarativity but the thread’s nix survivor ratio is grim

        Yeah lol.

        I will say, that for my server, I decided to use kubernetes + fluxcd for declaratively. My entire kubernetes “state” is declared in a git repo, and this is the popular, industry standard for things like this, called GitOps. It makes it very easy to add an app, since it’s just adding a folder + some new config files. And unlike Nix, Kubernetes and Flux are very well documented with much tooling as well. Nix doesn’t really have a working LSP or good code autocomplete, but with kubernetes, I can just start typing in a yaml file and then hit tab and it spits out the template for me. Code autocompletion with kubernetes feels much more similar to the tooling of other, more mature tooling

        It’s not as declarative as nix though. There are things missing, like OCI containers could theoretically shift if you don’t rely on hashes and some other nitpicks. But declarativity is a spectrum, and I feel like, outside of scientific scenarios (think simulations where versioning, hardware, runtime etc being the same is very important), I think many non-nixos solutions are declarative enough.