Default scaffolding with savvy::savvy_init().

[jibarozzo]

Addressing issue #5. Used the defaults from savvy::savvy_init() to create Rust scaffolding inside R package. Depending on what we decide with extendr or savvy we can merge this branch or create another with the default scaffolding from extendr. They provide similar structure and documentation.

[jdhoffa]

@jibarozzo i have not yet had the chance to dig into the differences between extendr and savvy

I will dig into that early next week, but in the mean time could you please summarize what factors lead you to decide on this approach?

[jibarozzo]

Initial differences

The scaffolding for both is about the same. The main difference I see is in error handling. Error handling with savvy uses Result instead of panic! . extendr doesn't seem to be progressing on that front, see here. I think having better error handling will benefit us in the long run.

Another main difference I see are the dependencies. extendr has many dependencies and relies on rxtendr and to provide helper functions to allow interfacing between R and Rust, meanwhile savvy (crate and R package are called the same) has barely any dependencies and the helper functions are readily available by installing the savvy R package, which installs the crate.

Caveats

• savvy prefers explicitness over ergonomics
• savvy provides limited amount of APIs and might not fit for complex usages

Some design ideas from the author (summarized here)

• Treating external SEXP and owned SEXP differently
• No implicit conversions

Final thoughts

The idea with extender is to expose Rust functions to R vs. savvy which geared toward working with R data structures inside Rust, rather than wrapping Rust code into an R-facing package. It depends on how we want it to handle the data.
The main decision to go with savvy was the simplicity and error handling. As I have dug deeper into the two frameworks I've learned that we would need to implement your own threading logic (e.g., using rayon in Rust), but integrating this back into R via savvy would be more cumbersome compared to using extendr. Given that the main objective is to re-implement functions using multi-threading savvy is not ideal. We could keep it if we are up to the challenge of a developing custom threading logic.

[jdhoffa]

Looks like there is CI failure on R CMD check for ubuntu and windows (however only warnings) but not sure how much that matters.

Approving

[jibarozzo]

I checked the warning a little bit closer. For the windows-latest (oldrel4) the logs indicate that R 4.0.5 is being used for the windows-latest (oldrel4) and it could be related to a corrupt .drectve file. I'll dig into this a more. Since the ubuntu-latest (devel) version test is the other one failing, I suspect the source for this is a new roll out of Ubuntu 24.04 and something to do with the the fact that R and Cargo were: Removed from the Ubuntu 24.04 image due to maintenance reasons. It suggests:

• Switch back to Ubuntu 22 by changing workflow YAML to use runs-on: ubuntu-22.04 We support two latest LTS Ubuntu versions, so Ubuntu 22 will still be maintained for the next 2 years.

For now, I'll merge and open an issue.