R Packages: There and Back Again

Alexander Senetcky

me@asenetcky.dev

2024-12-10

strata accepted into the CRAN

I learned some things along the way that I want to share.

• Not a nitty gritty look into package building

• High level overview of lessons learned

This presentation isn’t about strata but if you want to check it out, please do at: https://asenetcky.github.io/strata/

I’m not here to promote strata, and we won’t be covering it in depth. I will say that it is a package that helps teams withou fancy automation tooling to take the code they already have and place it in a framework that is easy to automate using the tools they already have on their computer.

What I wish I had known

It IS possible!

• There are excellent free resources out there

• Almost everything you need to know is in R Packages 2nd Edition

I just didn’t hear this enough, but it is possible and if the pros I mention later on in the presentation are critical to your projects, then I strongly encourage you to give package building a try.

Package Scope

• The R package can just be a small family of functions, it does not need to be the next tidyverse

• The R package just needs to be useful to you and your team

• It does not need to go up to the CRAN

I wish I had started years ago

If I had started honing this skill years ago, I would saved myself alot of headaches down the road.

What’s gained by making an R package?

Reproducibility

• Building a package bundles up all your code AND your dependencies

  • No more endless source() calls to helper scripts and functions

• Users can enforce strict versioning of dependencies

• R package build tools allow users to check, and ensure that the package it will run on just about any machine that is required

• Building a package forces users to write out that documentation we so often skip over

  • Makes it easier for others to use and understand your code

  • Makes you a better teammate

Portability

Building a package means that the code and documentation can be easily shared with others publicly and privately.

• No more emailing code around

• No more asking permission to access share drives or folders

• No more manually cutting and pasting code all over the place

Clarity and Organization

• The package structure forces you to organize your code in a way that is easy to understand and use

• The documentation lives alongside the code, making it easier to understand what the code does

• Build tools allow pieces of documentation to be used over and over again

  • Update once, and it updates everywhere

  • Consistent language throughout

Sharing R Packages

Getting an R Package requires alot of work and overheard that might not be worth it for a small project. However, users can still share their project with others it does not need to be on the CRAN.

Distributing your Package

Sharing your R package is easy to do with git and a repository hosting service.

• Github

• Gitlab

• Azure Devops

• Bitbucket

Sharing Publicly

You may remember Chuck showing off the pak package in a previous meeting. Once an R package is hosted in a remote repository, pak is just about all users need to install the package.

• pak::pak("<GITHUB USERNAME>/<PACKAGE NAME>")

• install the development version of strata:

  • pak::pak("asenetcky/strata")

Sharing Privately

• The remotes package is a great way to install packages from a private repository.

• For example we have a private CT DPH odp R package for internal use only that is hosted in our Azure Devops Project.

• Installing that would look something like:

  • remotes::install_git("<REPO CLONE URL>", git = "external)

One last note on Reproducibility

What we don’t have

• Most of us do not have access to the fanciest tools and services

• Most of us are not allowed to spool up a docker image, or server for development

What we do have

• Most of us have access to R

• Most of us have access to git (whether we realize it or not)

Pulling it all together

If users combine:

• R

• git

• Your own R Package

• renv package

Users will have a powerful combination of tools where reproducibility is paramount.