Sharing R code in a workgroup of Mac/Windows/Linux users

[This article was first published on R from Stata, and kindly contributed to R-bloggers]. (You can report issue about the content on this page here)
Want to share your content on R-bloggers? click here if you have a blog, or here if you don't.

In recent years there’s been a great deal of interest in, and work toward, creating more “reproducible” statistical code. I think this is a fantastic development. Looking at code I wrote in the late 90’s and early oughts, it’s clear how much a lot of my work from that period would have benefited from the coding habits I’ve developed over the last five or so years.

Working in R, a simple way to make code portable is to locate all data, functions, and scripts within a single directory/subdirectory which gets declared at the head of script and is followed by a series of relative paths in the lines to follow. This is standard process for projects you might find on Github.

But this is not always possible. Working in a corporate environment with protected/sensitive client data, I’m often forced to separate code from scripts, specifically, to keep certain sensitive data isolated on a particular server. I’m also not allowed to have any PII containing data in my git repositories. While I could exclude certain directories or file types from Git using, e.g., a .gitignore file, I’ve found it’s easier to just keep data in one place and scripts/functions in another. One more wrinkle: I work with Windows users, so simple references like “/data” will have different meanings if the Windows user is working from, say, the “D:\” drive.

I’ve tried and discarded a number of approaches to create a workflow that allows the same master code to run on a variety of computers which may have data stored in different places. Along the way I’ve refined my approach and have, in the last year or so, arrived at a process that allows me to work with others and with different types of OS’s in a reliable manner that requires little ongoing maintenance.

An added advantage the approach I’ve developed: if I want to move ALL of my main data sets, say from /data to /var/data, I can do this by changing one line of code in one file. After this change, all of my hundreds of scripts and markdown reports gracefully adapt to the change. They adapt because they all depend upon the same file to set up “the lay of the land” before any analysis. This requires a little extra work up front but, I find, saves a lot of headaches down the road — especially when working with other data scientists.

In short, my process is basically this:

1. The same (hopefully source controlled) file is sourced at the top of each analysis script. At the point of sourcing some parameters are declared which do the correct thing depending on whether the host is Linux, Mac, or Windows. Here’s an example:

## Load host-dependent directory environment
winos <- ifelse(grepl("windows",['sysname'],, 1, 0)
if(winos==1) source("C:/data/projects/scripts/R/functions/file_dir_params.R")
if(winos==0) source("~/projects/scripts/R/functions/file_dir_params.R")
rm(winos, host)

2. The next step is to build your version of the file/directory parameter file that was sourced in step 1 by a script (in this post I’ll call this file “file_dir_params.R”).

#--begin file_dir_params.R script--#

# Make a new environment:
fdirs <- new.env()

3. Now add this function to the “file_dir_params.R”, which is used to save a simple string indicating the host’s OS type:

# Function to standardize host OS name
get_os <- function(){
	sysinf <-
	if (!is.null(sysinf)){
		os <- sysinf['sysname']
		if (os == 'Darwin')
			os <- "osx"
	} else {
		os <- .Platform$OS.type
		if (grepl("^darwin", R.version$os))
			os <- "osx"
		if (grepl("linux-gnu", R.version$os))
			os <- "linux"
fdirs$computeros <- get_os()

4. With our new environment loaded and knowledge of the computer’s OS, we’re ready to build platform agnostic variables that point to the most important shared directories in your group:

## Declare the root project and data directory:

show  <- 1
build <- 1

if(grepl("windows", fdirs$computeros)==F){
	fdirs$prjdir <- "~/projects/"
	fdirs$prjdta <- "/your_data/"
	fdirs$prjdir <- "C:/projects/"
	fdirs$prjdta <- "D:/your_data/"

5. Depending on your setup you may have multiple project or data folders you want to declare. To keep my life simple on my development machine, I try and make all data a sub-directory of “prjdta” and all scripts a subdirectory of “prjdir”. Once you have these set using the platform agnostic pattern outlined in step 4, you’re ready to build out all your subdirectories. Because every sub-directory is a child of the root directories defined in step 4, make sure to always build new directory variables using one of the root directories (in my example, “prjdta” or “prjdir”). This is the key to making the code work across different platforms. It’s what allows the variable representing a folder to seamlessly shift between “C:/projects/some_folder” and “~/projects/some-folder” depending on the host which calls the script. One added and very useful bonus: if I move, say, my main data folder, I don’t have to rewrite 100 variable names in dozens of scripts. I just change the root data folder (in this example, “fdirs$prjdta”) once in one file and everything else takes care of itself in my batch environment. Here’s some examples:

# Add some child objects to the fdirs environment:
fdirs$dqrptsrc   <- paste0(fdirs$prjdta, "data_quality/source/")
if(show==1 & interactive()) cat("\ndqrptsrc =", fdirs$dqrptsrc)
if(make==1) system(paste0("mkdir -p ", fdirs$dqrptsrc))

# Define some colors for (say) GGPlot using your organization's color codes:
fdirs$com_ppt_orange <- "#FF6122"
fdirs$com_cr_blue    <- "#30812E"
fdirs$com_cr_red     <- "#D0212E"
fdirs$com_cr_green   <- "#8EB126"

6. Notice how, in the lines above, I reference the “show” and “build” flags we declared near the top of the “file_dir_params.R” script. What these do, respectively, is print the file location of the variable to standard output and build the folder if it does not exist. This can be useful but is not strictly necessary.

7. Our last step is to attach the “fdirs” environment (or safely reload it if it’s already attached):

# Attach the new environment (and safely reload if already attached):
while("fdirs" %in% search())

Final Thoughts

Attaching the environment as we did in step 7 is a great time saver because your can omit the “fdirs$” prefix when you reference variables in your scripts. For example, to load a file in a data folder I now just write:

x <- readRDS(paste0(dqrptsrc, "somefile.Rds"))

As opposed to:

x <- readRDS(paste0(fdirs$dqrptsrc, "somefile.Rds"))

But, as with everything, there’s a draw back: you’ll want to make sure the names of your variables don’t overlap with functions or other items you declare in a script. For this reason I use expressions like “prjdta” rather than “data”, and I avoid overly concise constructions that are used in a lot of example code, e.g. “x”, “y”, or “z”.

I hope some or all of the above is helpful to somebody, and feel free to drop me an email if you have any questions!

To leave a comment for the author, please follow the link and comment on their blog: R from Stata. offers daily e-mail updates about R news and tutorials about learning R and many other topics. Click here if you're looking to post or find an R/data-science job.
Want to share your content on R-bloggers? click here if you have a blog, or here if you don't.

Never miss an update!
Subscribe to R-bloggers to receive
e-mails with the latest R posts.
(You will not see this message again.)

Click here to close (This popup will not appear again)