Skip to content

Index

r2u: CRAN as Ubuntu Binaries

Key features

  • Full integration with apt as every binary resolves all its dependencies: No more installations (of pre-built archives) only to discover that a shared library is missing. No more surprises.

  • Full integration with apt so that an update of a system library cannot break an R package: if a (shared) library is used by a CRAN, the package manager knows and will not remove it. No more (R package) breakage from (system) library updates.

  • Installations are fast, automated and reversible thanks to package management layer.

  • Fast and well-connected mirror at r2u.stat.illinois.edu on the Internet2

  • Complete coverage with (currently) ~ 19,700 CRAN packages (and 230+ from BioConductor).

  • Complete support for both Ubuntu 20.04 ("focal") and Ubuntu 22.04 ("jammy").

  • Optional (but recommeded) bspm use automagically connects R functions like install.packages() to apt for access to binaries and dependencies.

Brief Demo

The gif below shows how one install.packages("tidyverse") command on an Ubuntu 20.04 system installs all packages and dependencies as binaries in 18 seconds (by passing the R package installation to apt using bspm).

This uses the Docker container referenced below, which has been set up with the five easy setup steps detailed here.

What is Covered ?

We generally support amd64 (i.e. standard Intel/AMD cpus) for the Ubuntu LTS release and the predecessor release. We use 'r-release' just like CRAN. So currently the 'focal' 20.04 LTS and 'jammy' 22.04 LTS releases are fully supported.

Support for other cpu architectures is certainly possible but somewhat unlikely due to a lack of (additional hardware) resources and time. Support for other distributions is possible but unlikely right now (due to a lack of resources and time). We hope to cover Debian at some point.

Current versions are R 4.2.1, and BioConductor release 3.15 packages are provided when required by CRAN packages.

What is Selected ?

Everything :)

Initially, we started from cran-logs by picking the N most-downloaded packages, along with their dependencies from BioConductor. (It should be noted that for example the first 100 packages already account for approximately half the total downloads: it is a very skewed distribution.) We iterated, and fairly soon arrived of full coverage of CRAN.

So we now cover

  • all CRAN packages (modulo at best handful of blacklisted ones) including all packages needing compilation
  • all BioConductor package that are implied by these (and build for us).

This currently resuls in 19772 and 19667 binary packages from CRAN in "focal" and "jammy", respectively, and 230 and 235 BioConductor packages, respectively, from the 3.15 release.

The sole exception are two packages we cannot build (as we do not have the required commercial software it accessess) plus less than a handful of 'odd build' that fail.

What is it Based On?

For the CRAN binaries we either repackage RSPM builds (where available) or build natively. All selected BioConductor 3.15 packages are built natively. For all of these, full dependency resolution and integration with the system is a key feature.

Everything is provided as .deb binary files with proper dependency resolution by using a proper apt repo which also has a signed Release file.

Usage and Setup

(Note that you could use this script add_cranapt_focal.sh or the variant for jammy to facilitate the setup but you may prefer to execute the steps outlined here by hand.)

First add the repository key so that apt knows it (this is optional but recommended)

apt install --yes --no-install-recommends wget      # to add the key
wget -q -O- https://eddelbuettel.github.io/r2u/assets/dirk_eddelbuettel_key.asc \
    | tee -a /etc/apt/trusted.gpg.d/cranapt_key.asc

Second, add the repository to the apt registry. You can use the original host

echo "deb [arch=amd64] https://dirk.eddelbuettel.com/cranapt focal main" \
    > /etc/apt/sources.list.d/cranapt.list
apt update

or use the mirror at the University of Illinois Urbana-Champaign:

echo "deb [arch=amd64] https://r2u.stat.illinois.edu/ubuntu focal main" \
    > /etc/apt/sources.list.d/cranapt.list
apt update

(In either example, replace focal with jammy for use with Ubuntu 22.04.)

Third, and optionally, if you do not yet have the current R version, run these two lines (or use the standard CRAN repo setup)

wget -q -O- https://cloud.r-project.org/bin/linux/ubuntu/marutter_pubkey.asc \
    | tee -a /etc/apt/trusted.gpg.d/cran_ubuntu_key.asc
echo "deb [arch=amd64] https://cloud.r-project.org/bin/linux/ubuntu focal-cran40/" \
    > /etc/apt/sources.list.d/cran-ubuntu.list

(Again, replace focal with jammy for use with Ubuntu 22.04.)

Fourth, add repository 'pinning' as apt might get confused by some older packages (in the Ubuntu distro) which accidentally appear with a higher version number. See the next section to ensure 'CRANapt' sorts highest.

After that the package are known (under their r-cran-* and r-bioc-* names). You can install them on the command-line using apt and apt-get, via aptitude as well as other front-ends.

Fifth, and also optional, install and enable the bspm package so that CRANapt and other packages become available via install.packages() and update.packages(). Note that you may need to install it directly from source via sudo Rscript -e 'install.packages("bspm")' to ensure it integrates correctly with the packaging system.

Pinning

Because we let apt (and related tools) pick the packages, we have to ensure that the CRANapt repo sorts higher than the default repo as (older) package builds in the distribution itself may appear (to apt) to be newer. A case in point was package gtable whose version in Ubuntu was 0.3.0+dfsg-1 which accidentally sorts higher than the rebuild we made under a newer and more consistent version number 0.3.0-1.ca2004.1. One possible fix is 'apt pinning'. Place a file /etc/apt/preferences.d/99cranapt with content

Package: *
Pin: release o=CRAN-Apt Project
Pin: release l=CRAN-Apt Packages
Pin-Priority: 700

which will now give packages from this repo a higher default priority of 700 overriding the standard value of 500.

Docker

There are also two Docker containers for Ubuntu 20.04 'focal' and 22.04 'jammy', respectively, at eddelbuettel/r2u that have the above, including pinning and bspm support, already set up.

Note that with recent builds of Docker (and possibly related to Ubuntu hosts) you may have to add the --security-opt seccomp=unconfined option to your Docker invocation to take advantage of bspm and the full system integration inside the container. This is also documented in the FAQ.

We also found that when building containers based off the r2u containers, we could not rely on the nice bspm integration as it requires superuser rights to pass off commands from install.packages() to apt. You can still use r2u containers as a base, but sadly have to turn off bspm and use just apt commands to install packages.

Try It

Use this link below (after possibly signing up for gitpod.io first)

Open in Gitpod

and run one of the three example scripts, or just start R in the terminal window.

The gif below display running one such example to install brms from binaries in a few seconds. Using this requires only (free) GitHub and GitPod accounts.

Support

Please file issues at the GitHub issues for r2u.

Known Issues

  • The littler package reflects build-time configuration, the RSPM binary is then expecting a different R location so it needs a binary rebuild. Added a 'force' flag, may need a list similar to the blacklist to always compiled.

  • A small number of packages do not build for lack required components; examples are ROracle and Rcplex. They, and their reverse dependencies, are are blacklisted and not built.

  • r2u is an apt repo, which via bspm becomes used "automagically" via standard R calls of install.packages() and alike. That last part is important: package installations that do not use install.packages() (such as the RStudio GUI, renv, rig, ...) do not benefit from install.packages() calling apt for you, and cannot take advantage of r2u.

Fixed Issues

  • [DONE] In May, some geospatial packages do not currently install on 20.04, adding the UbuntuGIS PPA as a base may help. This is not an issue on 22.04. It also no longer appears to be an issue with 20.04 now.

  • [DONE] The BioConductor release is still at 3.14 and should be upgraded to the now-current 3.15.

  • [DONE] Support for Ubuntu 22.04 has been added as well.

  • [DONE] Package nlme needed special treatment because the Debian/Ubuntu binary changed the 'hyphenated' version to a 'dot-separated' one leading to an inverted sorting order (see issue #7, as well as the FAQ, for more on this).

  • [DONE] Because Java is so special (and keeps its libraries 'elsewhere', the shared library dependence for package rJava was not fully specified. We added a special treatment for it.

  • [DONE] Both sf and terra needed a dependency adjustment on 'focal' for their most recent versions.

Author

Dirk Eddelbuettel

License

The repository-building code in this package is released under the GPL (>= 2).

All CRAN and BioConductor packages are released under their respective licenses.

Acknowledgment

This was made possible by the generous support of endless coffee thanks to my GitHub Sponsors.