| A good summary of the pain points I have with Nix. I was introduced to it recently at DayJob because one of the engineers I work with is a huge advocate for it. I think there's a good use case for a small subset of Nix + direnv to manage system packages for repositories. That's essentially what we use at DayJob - but all it does is install system packages necessary for a containerized workflow. However, even that level of complexity, even if it's just a small flake.nix and .envrc, can be cumbersome to end users and we actively seek to hide away that complexity as much as possible because anyone not a Nix expert that is presented to work with Nix is entering a deep rabbit hole of complexity that is probably ultimately unrelated to the problem they are trying to solve. Part of this I think is a branding and documentation issue with Nix. As a counterexample: I work with Argo a lot and they handle this sort of situation better. There is ArgoCD, Argo Workflows, Argo Events, and Argo Rollouts. They are all under the Argo umbrella that do distinctive things with clearly demarcated roles. Unlike Nix, I never get confused when I'm searching for ArgoCD related docs. I don't have to worry that I'm going to happen on Argo Rollouts related stuff and somehow not understand that Argo Rollouts is not the piece of Argo that I'm trying to use right now. The lines between each of the technologies are clearly demarcated and there is no confusion about the best way to do XYZ in each of them. Not so with Nix - if I'm new to Nix and I'm looking up how to structure something in a flake, not only is there different setups/configs for a build tooling setup vs a package management setup, but even within a build tooling setup for instance there are probably 4-5 canonical different ways to configure things, all with a ton of decisions. Heck, if I'm new to Nix it is probably not even immediately obvious that I should be doing it with a flake instead of The Old Way, and I might even start implementing something using The Old Way before finding out buried in some Github issue from years ago that I'm Doing it Wrong Thing that makes Nix amazing: It's infinitely customizable and welcomes that philosophy Thing that prevents Nix from succeeding: It's infinitely customizable and welcomes that philosophy TL;DR Nix favors configuration over convention, oftentimes to its detriment Final footnote. I am writing this on a personal x86 machine managed with nix-darwin. It functions, but the amount of nondeterministic calls out to homebrew and kludges required for it to work essentially defeats the purpose of managing the machine declaratively in the first place. All of the following extremely popular softwares available in nixpkgs - 1Password, VSCode, Firefox, Docker Desktop - don't work out of the box on this machine and require either compiling the package yourself, jumping through some hoops to get it to work with MacOS code signing, or require you to just entirely ignore the Nix aspect and have Nix shell out to Homebrew. There are also quite a few binaries of packages that are simply unavailable in Nix. |