Good question. Drat is an R package which makes it really easy to provide R packages via a repository, and also makes it easy to use such repositories for package installation and upgrades.
The motivation for drat is to give package authors more control over how they make their packages available to users. It does this by enabling package authors to make releases on a GitHub repository (or any server, more on this below). drat allows anyone to create their own package repository – just like CRAN.
There are two key advantages of using drat over other methods of installing packages from GitHub. First is that a package installed from a drat repository will be supported by
update.packages(), so the user has easy methods for keeping up-to-date. A second advantage of using drat is that the package author can control what other people get when they download your package from GitHub. With other methods, users typically download a random install snapshot, which might have unexpected consequences for them.
drat itself is a package, so it has source code (on GitHub) and a package (on the main R repository network).
But wait: we also call the repositories created or used via drat “drat” repositories. So we have to distinguish between “drat the package” and “a drat repository” created or used by it.
Hope this clarified things a little. Drat really can be different things which reside in different places but it aims to be a helper package which helps creating and using R package repositories easier.
Glad you asked. We have written
useruniquely identifies a URL
user.github.io/drat/, and communicating a single variable is easier than communicating a full URL
Taken together, we have this a pretty good to default on GitHub for repositories. But read on …
Fear not, as drat also supports repositories on a local drive, or shared network drive, or actually just about anything where you can write to or read from.
We detail that in the corresponding vignettes.
install_github() is a fine tool and does what it sets out to do: grab a snapshot from GitHub and install it. This can be the HEAD (by default), or a tag, or a commit, or from a branch.
But we think that is not what R needs. R has become so very successful for many reasons, but (in the eyes of many) one key part of the success was repositories ensuring both easy installation and easy upgrades.
That second point is lost on
install_github(): it installs what we may call “orphans”. Packages that are disconnected from an upgrade path. (With the exception, of course, of a newer version of what you install appearing in a known repository so that
update.packages will see it.) We think that is a disservice to the users, and a repository can do better in a fundamental way than provide access to (somewhat random) commits. Of course, one could write new code building on top of
install_github() and adding the functionality. But then, why? R already has this functionality, and had it for decades: using repositories. So drat does not aim to replace
install_github(); it simply aims at something both different and possibly much simpler.
Moreover, drat puts the “release” option back into the hand of the package authors. By cutting a release tarball and placing it into a repository, we think a possibly more informed release snapshot is distributed than by pointing at any branch of repository.
And last but not least, drat (>= 0.0.4) and its repositories also support binary installations.
You bet. Jan Schulz provided a careful pull request to provide initial support upon which we have built. As of release 0.1.0, this should just work for both Windows and OS X.
Glad you asked! In fact, you can use this GitHub query to find several dozen of packages using
Additional_repositories to point to drat repos (still having drat in their name; other such as packages.ropensci.org may use domain nane such that the search may miss them). One example we used to point to here changed however. So for a more stable use case, one example from my packages: the RcppRedis package points via this line to the ghrr drat to (optionally) use the RcppMsgPack package (which is not on CRAN as it uses a MsgPack release newer than what it in Debian).
The miniCRAN package creates self-sufficient repositories by examining the dependency graph and downloading all dependent packages. As such, it is more of complement to drat than an alternative — and Word is that several people have in fact combined both.
The small helper script
getCommitMessageForDrat.sh implements a simple format I have suggested in the past and used:
It combined four elements: - the package name - the current version - the current git repository commit hash marker - the repository URL.
When both source and binary packages are uploaded, I often add
(win) to indicate the type of package committed.
This format has worked and could form the basis of a standard, but suggestions and refinements are certainly welcome.