Loading vs. attaching a package

R users often talk of loading a package when they use library(). But technical library() doesn’t load the package but attaches it. So what’s the difference?

If you’ve ever used the :: operator, e.g. dplyr::filter(), you have loaded a package. Loading a package does exactly what the name suggests: it loads all functions and datasets of a particular package. However, to access these functions and datasets you will still need to use :: every time you refer to something within the package.

When you use library() the package is attached to the search path. You can think of the search path as a queue of (literally) packages, each one filled with functions and datasets. Whenever you use a function without ::, R looks for a function with that name in the first package. If it doesn’t find it, R continues to look for it in the next package and so on until it finds the function.

Importantly, calling library(pkg) will place pkg at the beginning of the queue. That’s why after calling library(dplyr), filter() will be dplyr::filter() and not stats::filter() any longer.

args(filter) # stats::filter

## function (x, filter, method = c("convolution", "recursive"), 
##     sides = 2L, circular = FALSE, init = NULL) 
## NULL

library(dplyr)
args(filter) # dplyr::filter

## function (.data, ..., .preserve = FALSE) 
## NULL

This behavior can cause trouble, so if you know there’s a function with the same name in multiple packages I’d advice you to be explicit and use :: every time you use that function.