library() vs require() in R | /en/2014/07/library-vs-require/

Guest *Kun Ren* @ 2014-07-27 13:26:41 originally posted:

Thanks for your explanation! By the way, do you think it is a good practice to use "foo::bar()", as I asked on stackoverflow (http://stackoverflow.com/questions/23232791/is-it-a-good-practice-to-call-functions-in-a-package-via)?

@renkun-ken

Guest *Jacob Westfall* @ 2014-07-27 18:38:28 originally posted:

I guess the moral is that when it comes to loading packages, "Try not! Do, or do not. There is no try."

Guest *duk* @ 2014-07-27 20:48:20 originally posted:

Nice post, thanks.

Guest *anthony damico* @ 2014-07-27 21:09:45 originally posted:

thank you

Guest *Guansong Wang* @ 2014-07-27 21:46:00 originally posted:

I agree it's pointless to use require() most of time. However, it's nice to be able to do something conditionally on the existence of some libraries, especially when I want to share the script/function.
For example, I use require() to load "multicore" (yes, quotations!). Then it will still run when someone wants to test it and doesn't have the packages installed.

Guest *Linlin Yan* @ 2014-07-27 23:21:04 originally posted:

+1 and I confess that I was making the same mistake! Thanks!

@yanlinlin82

Guest *Tal Galili* @ 2014-07-28 12:24:26 originally posted:

Dear Yihui - thank you for a GREAT post (as usual).

I admit that I didn't know (or forgot) that library gives an Error (when the package is not loaded) - so yay for learning something new!

One more (meaningless, and unimportant) reason I like (or liked) using require is that (for me) it is just faster to type (since all the letters for "require" are on the same keyboard row, as opposed to "library").

Yours,
Tal

Guest *Justace Clutter* @ 2014-07-28 13:15:19 originally posted:

One... I agree with everything here.

Now, let me just jump in here and say.... "Love the videos...." I was just sitting in my office chair here laughing. Great start to a Monday morning! Learned something new and got a smile out of it.

Guest *Aleksandr Blekh* @ 2014-07-28 13:49:57 originally posted:

Interesting post, thanks! However, in contrast with library(), require() returns a logical value, which makes it useful in certain situations. I find require() beneficial - in combination with library() - to guarantee the availability of the package of interest. Consider the following example:

if (!require(RCurl)) install.packages('RCurl')
library(RCurl)

Guest *Marcus Beck* @ 2014-07-28 13:51:33 originally posted:

What about using require to install a package since it returns a logical?

if(require(foo)) {
message('foo loaded correctly')
} else {
install.packages('foo')
}

This would assume that it's okay to install packages w/o consent though...

Guest *Norm Matloff* @ 2014-08-01 00:19:02 originally posted:

My impression is that somewhere, someone decided that require() is more appropriate within a function for documentation reasons, i.e. to show that this function needs the given package while other functions in the same program don't.

Personally, I think that most dicta in R, and in software development in general, are more follow-like-sheep orthodoxy than real "truths."

Guest *aliarsalankazmi* @ 2014-08-28 16:45:06 originally posted:

Thanks!

Guest *Charles* @ 2014-11-12 15:20:50 originally posted:

Hi, I have a problem seems. I should remove keep.source from function?

Error in require(MASS, quietly = TRUE, keep.source = FALSE) : unused argument (keep.source = FALSE)

Guest *Mike Williamson* @ 2014-12-17 18:17:45 originally posted:

Thank you for this, Yihui! Nice explanation.

You also captured the unfortunate reason why this confusion exists:

"There is one legitimate reason to use require(), though, and that is, 'require is a verb and library is a noun!' I completely agree. require should have been a very nice name to choose for the purpose of loading a package, but unfortunately... you know."

I would go one further and say that "require" means "to be necessary". Therefore, the action that it TRIES and will silently fail with a bool=False does not match the English definition of the word.

I think if library() were actually defined as require() and require() were defined as use() or something similar, these problems wouldn't exist. In fact, I see no problem with

if (!require('foo')) {
stop('The package foo was not installed')
}

precisely because it makes the code easier to read, both for R experts and newbies alike (and more similar to the try-except patterns in other languages). When these poor choices in the language structure are made, useRs make do with what we have.

However, this all pales in comparison to the much larger problem with libraries / packages: name collision by default. :( So, I guess we should pick & choose our battles.

Regardless, your discussion here was excellent! Xie Xie!

Guest *100 shades of red* @ 2015-02-12 16:29:48 originally posted:

require is better because library is the reserved keyword in other environments when you happen to use an interconnecting api between r and those

Guest *BroVic* @ 2016-05-15 00:37:46 originally posted:

Thank you for this insightful post, Yihui. What I initially thought was one of those quirks in R is actually something noteworthy!

Guest *Sree Krishna* @ 2016-06-22 14:09:39 originally posted:

This is enlightening. Thank you!

Guest *natus* @ 2017-01-05 21:51:28 originally posted:

library(stringi);detach("package:stringi",unload=TRUE);system.time({for(i in 1:50000){require(stringi);stri_trim("ntoton");stri_trim("ttotot");stri_trim("ntoton")}})
Le chargement a nécessité le package : stringi
utilisateur système écoulé
10.558 0.000 10.572
library(stringi);detach("package:stringi",unload=TRUE);system.time({for(i in 1:50000){library(stringi);stri_trim("ntoton");stri_trim("ttotot");stri_trim("ntoton")}})
utilisateur système écoulé
15.253 0.804 16.100
library(stringi);detach("package:stringi",unload=TRUE);system.time({for(i in 1:50000){stringi::stri_trim("ntoton");stringi::stri_trim("ttotot");stringi::stri_trim("ntoton")}})
utilisateur système écoulé
12.337 0.000 12.353
=> require is faster in such case. (by eg: using the parrallell/foreach package)

Guest *Sollano Rabelo Braga* @ 2017-01-26 13:37:51 originally posted:

This is awesome. Thank you!

Guest *Noah Fields* @ 2017-05-18 14:48:48 originally posted:

The name "require" sounds like it should do just that--require that the package be loaded or break down. The fact that it does the opposite is counter-intuitive. Perhaps this is the cause of people's confusion (and/or anger). They should have called it attempt().

Guest *BroVic* @ 2017-10-19 16:17:33 originally posted:

Great post (I've read this several times). For me, one of the quirkiest things about this topic is what happens when one calls library() without arguments. It's great, but one cannot help but wonder at the disconnect...

Guest *Liechi Zhang* @ 2018-01-12 08:28:15 originally posted:

Nice post. The behaviors of these functions were of no difference to me before, good to know them. I was also wondering why install.packages() requires "" around the pkg name but library() dose not, it is a surprise for me to find the answer here. This post also reminds me of another thing in R, why install.packages() accepts a vector of pkg names, but library() and require() will throw out an error when I try to load several pkgs at the same time? I know how to do that now using lapply(), just do not understand the logic of design here.

@liechi

Guest *Jose De Jesus Filho* @ 2018-01-12 19:08:57 originally posted:

I have read this post multiple times and still enjoy it.

Guest *Ido Hatam* @ 2019-05-15 23:02:53 originally posted:

Guess I'm late to the great post party, but this is a great post. Thanks.

Guest *Esty* @ 2019-11-18 19:41:43 originally posted:

VERY nice post, thank you