Skip to content

Travis Hinkelman

Avoiding namespace pollution in R and Chez Scheme

I was reading a blog post that mentioned that Julia has "[w]eak conventions about namespace pollution" and it got me thinking about how I manage namespace pollution in R and Chez Scheme. The short answer is that I don't. I developed bad habits in R centered around writing overly terse code.

These habits emerged in a pre-RStudio, pre-dplyr world where I wasn't using autocomplete and disliked long data frame and function names. For example, I strongly preferred writing

> df$CS <- cumsum(df$Ttl)


> fish_counts$CountTotal_CumSum <- cumsum(fish_counts$CountTotal)

My R code was littered with acronyms that made sense when I was writing the code, but were hard to decipher when I came back to it later. Essentially, I was producing write-only code. Because I was usually writing short, one-off scripts that were infrequently revisited, I rarely felt the type of pain that this style can inflict.

I am a heavy user of 3rd-party packages for R, but my preference for terseness meant that I tended to put all of my library calls at the top of the script and re-arrange the order that packages were loaded to manage namespace conflicts [1]. Honestly, I only today learned that library has arguments (exclude and include.only) for managing the objects that are attached. Because it inevitably leads to more verbose code, I also rarely used :: (e.g., dplyr::filter) to unambigously specify which version of a function that I wanted to use.

I don't profess to have a strong understanding of namespace best practices. Nonetheless, my current understanding is that generally it is a good practice to only load the functions that you are using from a package. In interactive use, though, I don't think that you want to impede your programming flow by constraining the functions immediately available to you. ggplot2 comes to mind here as a case where you are better off loading the full package functionality.

When you are using only one function from a package, you might opt for :: if you only call that function once or twice and include.only if you call the function many times in a script. For me, MESS::auc is an example where I have typically used the :: operator because I want the reminder that auc is found in the MESS packge. Alternatively, I could use library(MESS, include.only = c("auc")).

The approach for importing libraries in Chez Scheme is similar to loading packages in R. To load all the functions in my chez-stats library, you use (import (chez-stats)). We can import only the mean and median procedures with

> (import (only (chez-stats) mean median))

We can also reduce the potential for namespace conflicts by importing a library with a prefix.

> (import (prefix (chez-stats) stats:))
> (stats:mean '(1 2 2 3 3 3 4 4 4 4))

My preliminary experience of using the prefix approach greatly improved my autocompletion experience in Emacs. Similarly, a discussion of function naming conventions in R packages was largely centered on how function prefixes pair nicely with autocompletion in making an R package friendly to new users.

[1] The conflicted package offers a stricter way to handle namespace conflicts but I have not tried it.