Using travis-ci in research compendia and student repos

In general I’m a big believer in the R package mechanism as a convenient way to package and distribute a reproducible compendium of code, data, notebooks and manuscripts. However, it’s often overkill, particularly since passing basic continuous integration checks (e.g. via devtools::use_travis()) means running a relatively strict R CMD check that will worry about things such as documentation norms and details that aren’t always appropriate.

What I most want is an easy way to confirm that any .Rmd notebooks in a repository can be successfully knit on an independent machine – i.e. that the compendium can access all the software / functions and data it needs. Sure, some additional checks might be necessary (though these are sometimes best formulated in possibly-hidden .Rmd chunks), but this would detect the main ills such as failing to load libraries or using user-specific absolute paths.

A light-weight solution proposed in this ropensci thread inspired me to write this mock-up compendium which does just that, using a DESCRIPTION file to indicate dependencies and other metadata, and a test.R script called by .travis.yml to rmarkdown::render all the .Rmds:

A minimal DESCRIPTION:

Package: compendium
Version: 0.1.0
Depends: tidyverse, rmarkdown

A Package name and version are required for devtools::install() to work. Hopefully this approach also helps transition users towards providing more metadata in DESCRIPTION files, such as Author, author role, and License information.

Here’s my test.R script:

f <- list.files(recursive = TRUE)
Rmds <- f[grepl(".Rmd$", f)]
lapply(Rmds, rmarkdown::render)

though one could clearly imagine permutations on this. (The thread discusses supporting make, or using a particular directory structure to decide what does and doesn’t get run.)

Getting this to exploit r-travis (which does a nice job of handling key things like installing and caching installs of packages) involves just a minor tweak to providing a different script: testing routine from the default of R CMD build, R CMD INSTALL and R CMD CHECK:

.travis.yml:

language: R
sudo: false
cache: packages
script: 
  - R -e "devtools::install()"
  - R -f test.R

We’ll see how this goes in practice.