rustPlatform.fetchCargoVendor: init

[TomaSajt]

Related: #327063

---

This PR adds an alternative method to vendor the dependencies of a cargo project.
It's currently implemented in python, but could be changed later, if need be.

I converted some packages to use this fetcher as an example.

If you can come up with a good name for this fetcher, please share it below.

If we use this fetcher we won't have to worry about cargo changing its vendoring implementation (since this is completely custom), thus we can replace all importCargoLock usages where it was only required because of the presence of git dependencies.

The main idea is that we try to create a minimal staging FOD.
We fetch the tarballs into $out/tarballs/${name}-${version}.tar.gz and we fetch the git repos to $out/git/${git_sha_rev}.

This is the most important part to get right, since we will only be able to minimally change this in the future.
We would still be able to add some extra behaviours in cases that currently just throw an error (e.g. fetching other registries).

---

The staging FOD will then be used to create the actual vendor dir.

Since this no longer requires network access, we can freely change the implementation of this part without worrying about that hash.

The cargo workspace inheritance logic was already reimplemented by a script written for importCargoLock, so I just reused that.

---

Since this is a two-step process and only the first step is a FOD, we won't have outputHash and friends on the result. If we want to override the outputHash, we would have to do so through the vendorStaging attribute.

TODO: it might be useful to add an easier way to override the outputHash of vendorStaging.

---

I added a commit that makes buildCargoPackage swap over to this new fetcher implementation. This is currently done through a boolean flag.
Should it maybe be an enum instead? e.g. cargoFetcher = "custom" vs cargoFetcher = "default";

---

Sidenote:
The current vendoring script only requires the Cargo.lock file and nothing else. This would mean that fetchCargoVendor may easily be changed to not take in src and friends, but only a path to the cargo lock, like ${src}/Cargo.lock. This is not IFD, since we never import the path from nix, only from a script. This would be similar to how it's done with fetchYarnDeps.

Having it as a normal mkDerivation, like it is now, we can also apply patches and use hooks, which might be preferred.

---

It would be possible to skip the staging phase and make the vendor dir the FOD, however, this would lock our output format to never be able to change again.

---

The current implementation uses nix-prefetch-git to fetch a reproducible git tree, as I wanted to make sure I didn't make any implementation mistakes. I wonder if there's a better method than this...

---

Nothing is stopping us from also compressing the final vendor directory into a tarball, just like with fetchCargoTarball.
Tarballs - obviously - take up less space, so ideally we would do this.
Though a better fetcher name might be needed in that case.
I'm keeping the output as a directory for now, until the testing is done.

---

I also worry about the fact that non-FOD derivations will rebuild whenever the inputs change. This would mean that whenever python needs to be rebuilt, every vendor directory created by this would also need to be rebuilt. The rebuilds of the vendor directories would not take much time, but the used up space would be a concern.
It would be great if we could tell hydra not to cache certain derivations.

Things done

• Built on platform(s)
  • x86_64-linux
  • aarch64-linux
  • x86_64-darwin
  • aarch64-darwin
• For non-Linux: Is sandboxing enabled in nix.conf? (See Nix manual)
  • sandbox = relaxed
  • sandbox = true
• Tested, as applicable:
  • NixOS test(s) (look inside nixos/tests)
  • and/or package tests
  • or, for functions and "core" functionality, tests in lib/tests or pkgs/test
  • made sure NixOS tests are linked to the relevant packages
• Tested compilation of all packages that depend on this change using nix-shell -p nixpkgs-review --run "nixpkgs-review rev HEAD". Note: all changes have to be committed, also see nixpkgs-review usage
• Tested basic functionality of all binary files (usually in ./result/bin/)
• 24.11 Release Notes (or backporting 23.11 and 24.05 Release notes)
  • (Package updates) Added a release notes entry if the change is major or breaking
  • (Module updates) Added a release notes entry if the change is significant
  • (Module addition) Added a release notes entry if adding a new NixOS module
• Fits CONTRIBUTING.md.

---

Add a 👍 reaction to pull requests you find important.

[TomaSajt]

@ofborg build granian popsicle

[TomaSajt]

I wonder why ofborg is failing...
Edit: looks like #331167 hit master, so I need config.toml instead of config.

[TomaSajt]

@ofborg build granian popsicle

[Atemu]

I wanted to do the minimum possible processing in the staging phase, to lower the possibility of messing something up in the implementation.

I see. That's reasonable but I also think unpacking a tarball is hard to get wrong. (If you remember the command)

importCargoLock should also "suffer" from the same storage problem, since it doesn't use its internal FODs directly either

Indeed. Given that it has more pressing issues, I don't think we should place much importance on it for sustainable use within Nixpkgs.

This means that fetchCargoVendor should not be worsening the storage situation by much, compared to importCargoLock

You are right of course. This is merely a low-hanging fruit optimisation beyond the status quo.

I'll open a separate issue to discuss optimisations for fetchCargoVendor.

[nixpkgs-ci]

Successfully created backport PR for release-24.11:

• [Backport release-24.11] rustPlatform.fetchCargoVendor: init #362288

[getchoo]

Making a backport as this is going to prevent a lot of other backports from going through, and shouldn't be a breaking change in any way. It can also lead to some nasty surprises on stable, as only setting useFetchCargoVendor = true won't cause an evaluation failure about fetchCargoVendor missing (found this in #362266)

[jian-lin]

I created #377558 to track the support of other registries.