Skip to content

Latest commit

 

History

History
55 lines (36 loc) · 2.44 KB

README.md

File metadata and controls

55 lines (36 loc) · 2.44 KB
Raw

How to make a package

There's not much to a package. Here are the steps to making one:

  1. Clone this repo.

  2. Add R functions and their roxygen2-style doccumentation to the R subdirectory. The man subdirectory has a README.md with an example of roxygen2 syntax that's adequate for most simple functions.

  3. Update the DESCRIPTION file with the package name, descriptions, author(s), today's date, and any needed Depends/Suggests/Imports or other comments

  4. Go to the package root directory (the one with the DESCRIPTION file) and run Rscript roxygen2::roxygenize(".") or whatever the RStudio equivalent is.

  5. When roxygenize runs, it should generate a NAMESPACE file in the root directory based on instructions in the R-file comments and it should also generate documentation in the man subdirectory based on the same comments.

  6. Update tests/testhat.R with the name of the package so that testthat loads the package before trying to test it. Writing and running tests is described in the R packages book,

    If you scroll down there are some

    Each individual test should run a small function and check that it has the correct output.

  7. Then you can go to the directory above the package root and run: R CMD INSTALL <package-root>

Good policies to follow:

  • Use the checkmate package for checking inputs, especially on user-facing functions. This package has a lovely vignette here: https://cran.r-project.org/web/packages/checkmate/vignettes/checkmate.html

  • Write unit tests when they add value, it's more often than you would think. They can add value when:

    • a function might fail silently and produce a cascade of failures that will require time to trace back to the failure point to debug.
    • a function might involve formulas that could easily be modified down the line to produce slightly incorrect results. Even checking function output for a single value can save a lot of time.
    • you discovered a bug in a function and you'd like to not have to trace that same bug again. This is called a "regression test".

  • Don't add dependencies unless you need them. Many packages have simple alternatives in base R so check with the group.

  • Use the styleR package to format your code. It can be run via usethis::use_tidy_style()