Syntax for loading vs. installing libraries.

Question

I've been an active R user for several years now, and something has always confused me. When installing packages (e.g. dplyr), one must specify the name of the package as a string i.e.

install.packages("dplyr")

Rather than

install.packages(dplyr)

This makes sense to me, since "dplyr" would pass as the name of the package and not as an object, which would be implied by dplyr without the quotation marks.

However, when we go to load the library, both the string and object version pass through and load the package. Both of the following correctly load the package:

library("dplyr")
detach("package:dplyr", unload=TRUE)
library(dplyr) 

There is no object named dplyr in my workspace, and I don't understand why these two base functions would have different syntax. I also don't understand why the version without the quotation marks wouldn't evaluate the object. For example

dplyr <- "mada"
install.packages(dplyr)
library(dplyr)

The above installs mada, but loads dplyr, even though dplyr is an object that evaluates to "mada". Note that I am working in RStudiov.0.99.467 on Mac OS 10.10.4, if that matters. Is there a reason for this, or is it simply that the functions work differently?

Answer 1

library function takes package name as first argument. This name actually refers to the name passed as package argument rather than any value associated to that name. library function internally converts the package argument to a character literal (except when character.only is set to TRUE).

So,

plyr <- "dplyr"
library(plyr)

will undergo

if(!character.only)
    package <- as.character(substitute(plyr))

to become

package <- "plyr"

install.packages takes a character vector as name(s) of packages to be installed.

dd <- "plyr"
install.packages(dd)

This would install plyr.

It is becauseinstall.packages don't do any such conversion with pkgs argument. So install.packages installs what pkgs refers to.

Answer 2

I don't understand why these two base functions would have different syntax.

It’s hard to speculate about the reasons for specific language design decisions. The decision to allow unquoted names for the library call has been criticised by many prominent R programmers as illogical, inconsistent and unnecessary, but here we are.

Besides the argument that it’s (marginally) more convenient not to type the quotes, another potential reason is similarity to other languages: for instance, in Python you import libraries by specifying their name like so: import lib_name, not import 'lib_name'. This encourages library writers to choose library names that are valid identifiers in the language (in Python, loaded libraries are objects that you need to refer to). This is less relevant in R, where you rarely refer to the library besides loading it.

I also don't understand why the version without the quotation marks wouldn't evaluate the object.

Because R allows non-standard evaluation. In particular, function arguments are only evaluated when they are referred to, not before:

f = function (arg) {}
f(stop('this won’t raise an error!'))

arg is never used, so never evaluated.

In the case of library, the argument isn’t evaluated either. Instead, it’s used in unevaluated form via substitute(package).

Answer 3

If you look through the source code for ?library you will see at line 230 click here :

if (!character.only) 
            package <- as.character(substitute(package))

The package name that the user supplies is coerced to character then concatenated:

pkgname <- paste("package", package, sep = ":")

This is what allows the input without quotes for library. install.packages does not have the same functionality.