310 karma · joined August 1, 2019
Honestly I don't think so, in modern Linux namespaces it's perfectly possible to run them in isolated filesystems and to provide whatever they want at whatever paths they want without patches and workarounds. Instead Nix forces it's way up at the build level, so whatever artifact you get as a result of Nix derivation is suitable for running in a Nix system alone. I'd rather build a generic artifact that could be run in a traditional Linux system and let the user decide how it want to store it, instead of forcing the Nix way.
> I've never experienced a problem with it
There are problems with gcc-compiled binaries for example, and they are solved with a post-compilation patch. This is a workaround which does not scale and shouldn't be there in the first place, but that's mainly not gcc's problem in my eyes (although it could've done a better job too).
> The Nix store provides quite a nice separation between the "definition" side and the "building" side.
Somewhere deep inside it indeed does, there is no Nix the lang in final derivations. Nevertheless, I had too much trouble working at that level.
A number of problems that I would like to be fixed: - Post-build patches to insure correct shared library paths in artifacts due to requirement to use absolute /nix/store instead of using the traditional Linux filesystem. - Numerous wrappers around existing build tools to move dependencies into nix packages to generate stuff like Cargo.nix. - Not so good handling of content-addressable packages due to historic cruft. - Horrible-horrible way to compute runtime-dependencies of packages [1]. This may be okay as a heuristic to initialize a project, but in the end it must be tweakable manually.
I see a future where a package manager builds a fully-isolated environment with only and only required dependencies for my app, using CAS but not forcing it's style of the filesystem on me. Nix is a good step into that direction, but that's not a usable product in my view, just a research project.
[1]: https://nixos.org/guides/nix-pills/automatic-runtime-depende...
Edit: Forgot to add, Nix the language is also not good, but I'd like not to discuss that matter. This would be solved by a more modular architecture, where we can manage package with a one tool and build packages with a different tool. Until then, I would be forced to code in Nix, and that's problematic no matter how good the package management is.
> There is definitely a bit of a learning curve but the time investment is frequently over exaggerated
I'm not talking about the learning curve and its time investment, I'm talking about design problems. Nix's invasiveness is completely unnecessary in modern Linux, it makes its installation a very special case and requires lots of patches to just get stuff to work in nix. The fact that nix patches built binaries so that they point to correct shared libraries locations is a crutch which shouldn't be there in the first place.
It also tries to reimplement pretty much every package manager and build tool, even if they already work well and provide the reproducibility guarantees, including cargo, poetry, npm/yarn. This is a time investment, but it doesn't help me build software that is more robust and correct, that part is already handled for me. Instead, it just worsens the DX, as it forces me to use tools non-native to the ecosystem without first-class support for commonly used features.
There are three practical problems as a result: - huge image sizes with unused dependencies delivered as part of the artifact; - limited ability to share dependencies due to inheritance-based model of layers, instead of composition-based model of package managers; - non-reproducibility of docker images (not containers) due loosely specified build instructions.
Predicting future comments, nix mostly fixes these issues, but it has a bunch of issues of its own. Most importantly, nix is incredibly invasive in development process, adopting it requires heavy time investments. Containers also provide better isolation