xfun::pkg_attach(): A Vectorized Version of library()

Yihui Xie 2018-09-04

Chances are you often need to use more than one package in R, so it is common to see that an R script starts with multiple calls to library(), the function that attaches a package to an R session, e.g.,


Repeat the same function call over and over again

The vectorized library()

With the function xfun::pkg_attach(), you can attach them through a character vector:

xfun::pkg_attach(c('ggplot2', 'dplyr', 'shiny'))

As I said in a post on “library() vs require()” in 2014, I believe it was a design mistake to use NSE (non-standard evaluation) in library(). It saves a little bit typing effort since you don’t need to type the quotes, but it also opens a can of worms, and introduces more non-standard things due to this non-standard thing (e.g., when you do code analysis, you have to make a special rule for the library() calls to make sure those “variables” in parentheses are not real variables but character strings of package names). The above xfun::pkg_attach() call is essentially a for-loop:

for (i in c('ggplot2', 'dplyr', 'shiny')) {
  library(i, character.only = TRUE)

Automatically install missing packages

It is also common to see code like this in R scripts (install a package if not installed or loadable):

if (!require('shiny')) install.packages('shiny')
if (!require('ggplot2')) install.packages('ggplot2')

The argument install = TRUE in xfun::pkg_attach() allows you to do the above task, e.g.,

xfun::pkg_attach(c('ggplot2', 'dplyr', 'shiny'), install = TRUE)

Or use the shorthand xfun::pkg_attach2(), which is essentially pkg_attach(..., install = TRUE):

xfun::pkg_attach2(c('ggplot2', 'dplyr', 'shiny'))

Some people may argue that you should never install software packages without the user’s permission. That is a good point. If your R script is meant to be run by other users, you need to be a little conservative, but some users may actually prefer convenience (i.e., don’t bother me with error messages about missing packages). Of course, if you are writing a script for yourself, you can always use pkg_attach2(). If you want to ask for permission, you certainly can, e.g.,

  c('ggplot2', 'dplyr', 'shiny'),
  install = readline('Install missing packages (y/n)? ') == 'y'

Suppress package startup messages

Some people don’t like package startup messages. I think those messages are fine if they only appear once or at most a couple of times to a new user. When you see the same messages over and over again, you will be bored, but you cannot shout at your screen that “I KNOW dplyr::filter() MASKS stats::filter()! PLEASE DON’T TELL ME THIS AGAIN!” when you library(tidyverse).

How to suppress package startup messages? As you could have guessed, the function is named suppressPackageStartupMessages(). Do you like it? I don’t, because it is too long and often calls for disproportional attention when you read the code (worse if you have to call it multiple times). Do I have a better name? No (so I’ll shut up on naming).

If you are attaching a package in knitr, you can use the chunk option message = FALSE to suppress package startup messages, e.g.,

```{r, message=FALSE}

Note that this will suppress any types of messages, including package startup messages. It might be a problem, but I often load packages in a separate code chunk in the beginning of a knitr document, so it is unlikely to suppress messages that I didn’t mean to suppress.

Since my colleague Winston was not satisfied by this solution, I added a message argument to xfun::pkg_attach() to suppress startup messages (and he didn’t like it, either, which was understandable).

xfun::pkg_attach('tidyverse', message = FALSE)

The default value of message in this function comes from the global option xfun.pkg_attach.message, so you may set it in your .Rprofile to always suppress startup messages, e.g.,

options(xfun.pkg_attach.message = FALSE)

I wish the library() function has a similar argument, so I don’t have to hack at it this way.

Will xfun::pkg_attach() become the new library()?

No, I don’t believe so, and I don’t mean so, either. I wrote this post just to illustrate one type of pain I mentioned a few days ago: the pain that is tiny yet comes frequently. To sum it up, the tiny pain here is:

  1. You cannot library(c('ggplot2', 'dplyr', 'shiny')).

  2. You have to test the availability of a package explicitly before deciding to install it, i.e., if (!require()) install.packages().

  3. You have to explicitly suppress the package startup messages every single time if you don’t want to see them, either through a long-named function or by a knitr chunk option.