Kind of depends what you want to package. For projects that force you to provide dependencies yourself (e.g. most C or C++ projects), Nix packaging is very easy to use. Just slap a flake.nix together with the necessary dependencies, where to get the source from and how to build it.

Where Nix gets really difficult is with packages that reinvent their own packaging system and do dynamic downloads at compile or even runtime. Those really do not harmonize with Nix, as the Nix build process happens in isolation without network access and wants to have all dependencies specified beforehand, with checksum and all.

When it comes to languages with their own package manager it also gets a bit complicated, as while Nix does come with workarounds for all the common cases, there are generally multiple ways to do it, e.g. you can use mach-nix, pypi2nix, buildFHSUserEnv or buildPythonPackage to build Python packages and it’s not always obvious which is the best approach or which will even work.

Packages that softly depend on other packages via some kind of plugin mechanism are also tricky, due to Nix packages all being isolated in their own directories. Again, which workaround works best here can be tricky, some packages require specifying all the plugins at package build time others use environment variables or other means to locate plugins.

All that said, these issues are kind of fundamental when you want to have a proper reproducible packaging system and hard to avoid. I do prefer a system that forces some cleanliness from the ground up instead of adding ever more ugly patchwork on top, but I can understand why that can be at times very frustrating.

Well I guess it depends how deep you’re in the rabbit hole already, I think it’s relatively easy for me at this point to create a new package (I’m maintainer already for quite a few). But yeah … steep learning curve … Less so with Nix itself, though non-the-less, it’s a simple functional programming language with a new paradigm (derivations). But rather NixOS/nixpkgs Nix magic. For example there’s a dynamic dependently typed type-system built on top of untyped Nix in the NixOS module system that is spin up on evaluation time.

But I understand your point, at the beginning of my NixOS journey I have also rather created a “good enough” Dockerfile. Depending on the exact context I still do this nowadays (often because there’s an official well maintained docker image in comparison to a not so well maintained Nix one, and the context is too complex to maintain/develop/extend it myself). But if there’s a good solution in Nix I rather use that, and that is often less headache than setting up a service with e.g. docker-compose. I also use flakes mostly for a dev environment, if you’re a little bit deeper in it, you can spin up a relatively clean dev env in short time (I’m often copy pasting the ones I have written from different projects, and change the packages/dependencies).

As I said it has a steep learning curve and documentation is pretty much the nixpkgs repo itself (well after understanding the basics of Nix and NixOS at least, with the combination of the https://nixos.wiki mostly IMO). It also takes some time to get used to the quirks of NixOS (and understanding the necessary practical design decisions of these quirks).

But I have nowadays seldom trouble with switching the generations (i.e. nixos-rebuild switch), unless you’re updating flake inputs or (legacy) channels (where e.g. a new kernel might be used). In that case it makes sense to reboot into the new configuration. Also, obviously that can lead to short down-times (including just restarting a systemd service, if a service has changed in between the generations), if that is unacceptable, there obviously needs to be a more sophisticated solution, like kubernetes via e.g. kubnix. I’m not sure how much of that can be achieved with Ansible, as I haven’t used it that much because I disliked the “programming” capabilities of the Ansible yaml syntax (which feels kinda hacky IMHO).

But apart from NixOS, one can also just use Nix on a different system to e.g. deploy or create docker images (which can be really compact, as only the necessary dependencies for a package is packaged) that in turn could e.g. be managed with Ansible or something…