A Quick Note About How To Bundle Rust Crates For An R Package

Recently, CRAN published a document titled “Using Rust in CRAN packages” and it refers to my R package, string2path, as an example of bundling all the source codes of the dependency Rust crates (I’m not sure if my name is still there at the point when you’re reading this). As I didn’t intend to develope the package as such a good example, let me explain about two possible pitfalls that you might not find easily just by reading the code.

Note that, I don’t talk about the case that requires downloading here.

My stance

First of all, while I’m hoping CRAN will success with Rust (so I sent some feedback about unclear points that everyone would wonder), I’d say, ultimately, CRAN is not a suitable place for using Rust. You should first consider R-universe to distribute your R pacakge using Rust. I believe R community needs a strong alternative to CRAN and R-univese can be. If you are interested in why I insist so, please read my blog post I wrote last year.

cargo vendor

Bundling, or “vendoring,” all dependency crates for your package can be easily done by cargo vendor. More specifically,

1. run cargo vendor and
2. copy the configuration shown in the output of cargo vendor to .cargo/config.toml to use that vendored dependency

You can try adding --offline option to cargo build to check if the build really uses the vendored ones.

While this mechanism is very simple, you can hit the following problems.

Path length limit on Windows

Vendored directories are so deep that you can easily hit with the path length limit on Windows. So, you probably cannot package the vendor directory as it is.

There can be some better solution, but I use tar command to avoid this failure; create a tar archive file and expand it in Makevars before cargo build. Here’s the command line that I actually use (ref):

# c.f. https://reproducible-builds.org/docs/archives/
tar \
  --sort=name \
  --mtime='1970-01-01 00:00:00Z' \
  --owner=0 \
  --group=0 \
  --numeric-owner \
  --xz \
  --create \
  --file=vendor.tar.xz \
  vendor

Another reason to use tar is that the size of these source codes. R package itself is compressed, but I found it’s more compact if I create another .xz file.

R CMD check considers .cargo as a “hidden directory”

As I described above, you have to configure .cargo/config.toml to use the vendored sources. But, if you include it naively, you’ll see this NOTE on R CMD check because .cargo starts with .:

Found the following hidden files and directories

So, you have to include .cargo/config.toml as a different name, and put it to the right place before cargo build. I do this like this (the file is named cargo_vendor_config.toml):

    # vendoring (Note: to avoid NOTE of "Found the following hidden files and
    # directories", .cargo needs to be created here)
    if [ "$(VENDORING)" = "yes" ]; then \
        $(TAR) --extract --xz -f ./rust/vendor.tar.xz -C ./rust && \
        mkdir -p ./rust/.cargo && \
        cp ./cargo_vendor_config.toml ./rust/.cargo/config.toml; \
    fi

    @BEFORE_CARGO_BUILD@ cd ./rust && cargo build --target=$(TARGET) --lib --release --offline

The full Makevars can be found here.