Sun 20 April 2014 | -- (permalink)
I spent a lot of last week documenting an R package. I'm still learning how to write good user-facing features (including documentation) in my software, so I asked Twitter for advice on two things: (1) what are some R packages with stellar documentation?, and (2) is there a documentation style guide?
I got some great responses! Here's the full Twitter conversation. And here are some of the cool things I learned over the course of the week:
The documentation for an R package is broken down into a few different parts:
- Documentation for each function in the package (this is pretty structured): Hadley Wickham's function docs page
- A few brief overview pages for the whole package (also fairly structured): Hadley Wickham's package docs page
- A package vignette, which is what I usually read when I'm learning to use a new package. The vignette is basically a "getting started" guide with lots of example code in it.
You can write almost all of this documentation into your source code files and use roxygen2 and devtools to automatically generate the properly-formatted, properly-organized, tex-like documentation files you'll need to publish a package.
I use Sublime Text + Terminal on a Mac, and these little tricks made the documentation process even easier with those tools:
- The RTools plugin for ST2 will automatically populate the roxygen documentation template above your function if you highlight the first line of the function and hit command-shift-alt-R. Super cool, props to Karthik for that. (Since RTools is designed to run code in
R.app, I also added the feature to the Sublime plugin I usually use for Terminal work).
- I added this line to my .bashrc file:
alias doc='R -q -e "devtools::document()"'. This means that when I'm in the package directory in the Terminal, I can just type "doc" to generate the documentation files.
- Since CRAN might get angry if your source files contain lines longer than 100 characters (thanks Scott), I added a vertical line at 100 characters in Sublime, so I can see all the lines that are too long. Here's how to do that.
My original tweet was really meant to get at documentation content: there's lots of good information out there on how to structure the documentation and which things need to be documented, but what should I actually write in the documentation files so that users will understand the function/package? Is there a standard (written or unwritten) about how R documentation files should read? An example of a well-documented package would also help answer the content question. Here's what the internet seemed to think about this:
- The function-specific documentation is not nearly as important as the vignette/web documentation (thanks Karl and Carson).
- Try not to document things in two different places (thanks Dave), unless a parameter description is getting extremely long (like, to the point of using
\itemize) (thanks Karl and Scott)
- There don't seem to be rules about declaring argument types in the
@paramfield, or always using the present tense, or nitty gritty things like that (please correct me if that's wrong)
- Try to give good/illustrative examples
- But really it's all a matter of personal preference. In the end, that meant my strategy was to go ahead and write the docs as I thought they should be, taking care to write a detailed vignette, and commit to taking user feedback super seriously.
People mentioned the following packages when asked for particularly well-documented ones:
- all the packages on Bioconductor, especially
GenomicRanges(vote from Vince)
mice+ descriptive paper, (vote from @statsepi)
lme4(vote from Valerio)
- anything by Roger, especially
filehash(vote from Karl)
R internet community
All this info from the Twitter conversation was super helpful to me. I even got a little code/doc review from Scott - how awesome is that!? Thanks, all, for your input. If you have more documentation suggestions, please share!