• unknowing8343
    link
    fedilink
    English
    arrow-up
    6
    ·
    1 year ago

    There is a part of me that wants to try this, but I have one question.

    I believe this distribution allows you to have multiple versions of the same library to work with different programs at the same time, correct? Does this mean that each program downloads all its dependencies independently? If the answer is yes, I am staying with Arch. Too much bloat.

    • Atemu@lemmy.ml
      link
      fedilink
      English
      arrow-up
      11
      arrow-down
      1
      ·
      edit-2
      1 year ago

      this distribution allows you to have multiple versions of the same library to work with different programs at the same time, correct?

      That is correct. You can even have programs from entirely different releases running on the same system without conflicting with another; one with i.e. glibc 3.34 and one with 3.37 for example. Or even wilder setups with some packages using musl, other glibc and others yet being built statically.

      Does this mean that each program downloads all its dependencies independently?

      Each program references all of its (exact) dependencies. Nix then looks at the program and its references and builds a tree of dependencies.
      Then it tries to “realise” these dependencies (make them, well, real), possibly by substituting (downloading) them from a binary cache or automatically building them on your machine if they’re not available in any configured cache.

      However, if some package with the same exact version already exists in the Nix store, no action will be taken. Why should it, it’s already there.
      For example, if you were in an empty world and built an environment with hello and coreutils in it, they’d both depend on glibc. If both came from the same revision of Nixpkgs, chances are that they depend on the exact same version of glibc.
      What Nix would do here is fetch 1x glibc, 1x hello and 1x coreutils.

      Note however that you don’t need to manage any of this. You just say “I want hello and coreutils”. Nix takes care of getting the correct dependencies in place but they won’t be in your immediate environment. When you then say that your environment should no longer contain hello, it does that. At this point hello will still exist in the Nix store but it won’t be in your PATH any longer, so it’s not polluting any shared state; it just sits there on disk and the worst it could do is waste disk space. It’s not “installed” in the same sense as what it’d mean to have an unused dependencies installed on an FHS system.
      In order to reclaim disk space from unused Nix store paths, you can simply run a garbage collection. You don’t need to care about that one specific hello store path here though, you just say “remove all unused store paths for me, thanks” and Nix removes it along with all other unnecessary paths. In NixOS, you can even have that ran periodically for you.
      (Note that this is distinct from autoremove and the like; those “clean up” the shared state and free up disk space. In NixOS, these are separate processes and dependencies which you don’t explicitly declare are never in the shared state to begin with.)

      You see, while this could be seen as “bloat”, it has none of the negative consequences bloat has on other systems such as more packages for you to manually manage, more binaries in your PATH or weird interactions of other programs. It’s just easily managed disk space and disk space is honestly quite cheap.

    • Past Pollution@lemmy.world
      link
      fedilink
      English
      arrow-up
      7
      ·
      1 year ago

      Not exactly. If one package needs foo as a dependency, and another package also needs foo, it won’t download a second copy.

      But if another package needs, say, an older version of foo, it can download the older version of foo and you’ll have two foo packages.

      This is possible because Nix packages have hashes to differentiate between them, so any package can ask for the exact version of foo it needs instead of asking for foo in general and hoping the version it gets isn’t incompatible.

      Pretty sure this is also different from other containerized package managers, like Flatpaks or Snaps, which I believe throw all the dependencies in with their packages so each package has exactly what it needs in its container, and which is obviously going to be much more bloated than having shared dependencies. As far as I know Nix on the other hand doesn’t get any more bloated than other distros (keep in mind that regular distros like Arch will still sometimes have multiple versions of a package, for example Python).

    • Kryptkravler@beehaw.org
      link
      fedilink
      English
      arrow-up
      6
      ·
      1 year ago

      Yeah. I feel the same way. I remember NPM (nodejs) had this exact issue and you would end up having so many duplicate packages.

    • Corngood@lemmy.ml
      link
      fedilink
      English
      arrow-up
      3
      ·
      1 year ago

      Each version of a library (or any package) will only exist once, and things are garbage collected when not referenced.

    • lloram239@feddit.de
      link
      fedilink
      English
      arrow-up
      1
      ·
      1 year ago

      All dependencies and packages go into the /nix/store/ and are shared.

      However you still end up with a lot more bloat than typically, as an upgrade will leave all the old versions behind. This allows you to instantly up & downgrade in case something goes wrong. But it also means you need twice the space. You can of course free the space afterwards, but you still need it for the upgrade.

      Another factor that can lead to bloat is that Nix makes no use of binary compatibility, meaning dependencies are specified down to the individual bits, so a package doesn’t just depend on some thing named libfoo, but on a very specific libfoo with a known sha256 checksum.

      The official packages in nixpkgs will all use the same libfoo, but if you build your own software or install software from other places you can quickly end up with multiple libfoos around (NixOS can still dedup them a little via hardlinks, but still bloat). You can manually override dependencies, but that’s something you have to do manually, there is no automation as far as I can tell (nix build --override-input nixpkgs 'flake:nixpkgs' should get reasonably close)

      This also leads to bigger downloads, as an update to a core library can require updating a large number of upward dependencies.

      Overall I think it’s worth the trade-off, as it gives you much more flexibility, but a typical NixOS install will generally still end up bigger than regular distributions.