On Wed, Mar 06, 2024 at 12:53:44PM -0500, Jason Conroy wrote: > On Wed, Mar 6, 2024 at 4:06 AM Efraim Flashner > wrote: > > > On Mon, Feb 26, 2024 at 09:24:29PM -0500, Jason Conroy wrote: > > > Hello Efraim, > > > > > > Thanks for investigating this - a Rust development workflow using only > > > Guix-native crates is something I've been waiting for! > > > > > > I was experimenting with your patches and it seems that they do pull in > > the > > > source crates for requested packages, but not their dependencies (example > > > below). Is there something I'm missing? > > > > When you say they pull in the source crates do you mean the sources of > > the other rust packages needed by rust-rand? I didn't test that, but I > > assumed it wouldn't. > > > > That's right, I'm interested in having `guix shell` populate the full set > of transitive dependencies in the registry, similar to what happens when > using `cargo build` in the conventional way (but instead using > Guix-packaged crates). I interpreted your message up-thread to mean that > you were trying to accomplish the same, but please correct me if not. The transitive dependencies getting pulled in automatically should work automatically if we ever finish the antioxidant-build-system. Until then I've been experimenting by manually listing the other crates I've needed but in theory we could try to make `guix shell --development` pull in the needed crates. I misread what you wrote as it was working. It's definitely something I want but wasn't ready to work on yet. > I took rust-rand as an example only because it does have some dependencies: > > $ cd $CARGO_PROJECT > $ cat Cargo.toml > [package] > name = "test_prog" > ... > [dependencies] > rand = "0.8.5" > > $ cargo build > Updating crates.io index > Downloaded cfg-if v1.0.0 > Downloaded rand_chacha v0.3.1 > Downloaded rand v0.8.5 > Downloaded ppv-lite86 v0.2.17 > Downloaded rand_core v0.6.4 > Downloaded getrandom v0.2.12 > Downloaded libc v0.2.153 > Downloaded 7 crates (932.0 KB) in 0.48s > Compiling libc v0.2.153 > Compiling cfg-if v1.0.0 > Compiling ppv-lite86 v0.2.17 > Compiling getrandom v0.2.12 > Compiling rand_core v0.6.4 > Compiling rand_chacha v0.3.1 > Compiling rand v0.8.5 > Compiling test_prog v0.1.0 (/home/...) > > > > Currently if you were to pull in rust-rand-0.8 and rust-rand-0.7 then > > you'd have both rand-0.*.crate files in the registry but only one of > > them would be listed in share/cargo/registry/index/ra/nd/rand. I need to > > adjust the generation of that file to combine multiple sources if they > > exist, and sort them (I'm not sure it's necessary, but wouldn't be > > surprised if we hit undefined behaviour if they were listed multiple > > times or out of order). > > > I'm somewhat new to rust, but it appears that outside of Guix, the > local-only development workflow looks like this: > > $ cd $CARGO_PROJECT > $ mkdir $VENDOR > $ cargo vendor $VENDOR > > After downloading and unpacking all of the crates into $VENDOR, this last > command instructs me to add the following in ~/cargo/config.toml. > Then, after opening a new guix shell without network access, I can confirm > that `cargo build` works fine with the vendored crates. > > [source.crates-io] > replace-with = "vendored-sources" > > [source.vendored-sources] > directory = "" I wanted local-registry over replace-with because IIRC replace-with won't fall back to downloading from crates.io if there's missing crates, while local-registry will check there first and then download any missing crates. The use-case I was looking at for that was adding a new dependency to a project and then not needing to re-create a shell or package the new crates before continuing on. The registry as setup in the patches is actually mostly correct, minus the multiple versions of a crate, it just needs to be fed the crates. > Getting back to your patch set: would it make sense to emulate this vendor > workflow instead of trying to construct a registry directly? Even assuming > that all details of the registry structure are stable and documented, the > layout of the vendor directory appears much simpler. And IIUC the code for > setting up vendored libraries already exists in cargo-build-system. > > I also need to figure out something with a > > config.toml to see if it's possible to generate one that could be > > included from another one, since you can't add 'local-registry = > > $GUIX_PROFILE/...' in a toml file. > > > > You've probably researched this more than I have, but it seems that this > use case is explicitly unsupported in the TOML language spec: > https://github.com/toml-lang/toml/issues/397 > > With that option off the table, I can't think of any elegant solutions. > Maybe a wrapper for the cargo binary that pre-processes cargo.toml and then > calls the real cargo? Thanks, I didn't see that one. So it looks like environment variables and INCLUDEs are off the table. An autoconf style macro that take a config.toml.in and spits out a config.toml with the correct directory would work, but from my understanding that's not really in line with any bit of how cargo and the rust ecosystem works. Another option would be to symlink $GUIX_ENVIRONMENT/share/cargo/registry to ./local-registry and then add the line 'local-registry = 'local-registry' in a config.toml. -- Efraim Flashner רנשלפ םירפא GPG key = A28B F40C 3E55 1372 662D 14F7 41AA E7DC CA3D 8351 Confidentiality cannot be guaranteed on emails sent or received unencrypted