The correct way to get almost anything installed is to write a Nix package, and install that. This is actually remarkably easy in comparison to other package managers - you write the package definition in a single file, normally just having to set the source code location and list its dependencies, and you can make it available to install with nix-env by adding it to a packageOverrides declaration in ~/.config/nixpkgs/config.nix.
The "stdenv" build system that nixpkgs uses will automatically install anything that can be configured via "./configure && make && make install" - cmake is supported just by listing that as a build dependency. Stuff written in scripting languages is more fiddly, but the nixpkgs documentation usually has examples. The downside is that a significant amount of the time, libraries written in scripting languages aren't packaged either, and you wind up packaging half a dozen libraries just to get some tool working. The <X>2nix tools can be useful in these cases to generate a package definition.
> The downside is that a significant amount of the time, libraries written in scripting languages aren't packaged either, and you wind up packaging half a dozen libraries just to get some tool working.
I've gone down this path a few times and days later when I still don't have the original thing I wanted working installed, then I sort of give up. I realize there are many benefits to Nix and I really enjoy them but I've switched to using Arch for practical reasons. I really, really miss the declarative reproducible style though.
I've used npm2nix and cargo2nix a couple of times with success - they generate package sets from the relevant sources.
I'm curious why not.
Aren't pip/cargo/whatever suitable for long-term use?
Consequently, couldn't nix's upgrading routines be modified to call out to pip/cargo/whatever to upgrade things best upgraded by pip/cargo/whatever?
I guess what I want to know is whether the architecture or the philosophy of nix interferes with nix's being modified in the way I just described. (Maybe it is just that nobody's gotten around to doing the implementation work.)
No. pip/cargo/etc are nondeterministic. Nix packages should not be. Where they are, things often break in bizarre ways.
In order to access the Internet, a Nix package must declare the hash of its output - given that pip can produce different output from run to run (it contains a constraint solver which depends on the versions available in the external package repository) this is impossible.
There's a similar tool for macOS called `nix-darwin`, as well: https://github.com/LnL7/nix-darwin
That won't work everywhere either as the location of 'env' is not a fixed standard either. Eg. In some *nix's its at /bin/env.
I'm not 100% sure on all of these, but every one of the dozens of Linux distros I've used has `/usr/bin/env`. macOS has it, and it looks like FreeBSD does, too, and so does NetBSD.
Who are the oddballs? Does it include any of the free Unices?
I would like to know if there is some common practice/standard around using '/usr/bin/env' in shebangs.
Could you please elaborate more?
There is also this function which may be necessary for certain software: https://nixos.org/nixpkgs/manual/#sec-fhs-environments
You can compile some software manually in place using, for example, `nix-shell -p libssh2 -p zlib` without having to write a derivation. However, without writing a derivation, it may cease to function after a garbage collection or an upgrade.
Writing a derivation isn't too bad, here's one for one of my repos: https://github.com/jbboehr/handlebars.c/blob/master/derivati...
Please correct me if I'm Nix-ing wrong.