Clean Built of Multiple R Versions on Ubuntu
Long time ago I built R-4.0.3 on Ubuntu system following these guides: link1, link2. Briefly, I added GPG Key to APT (Advanced Package Tool) and added CRAN repository and directly retrieved R by apt.
$ sudo apt-key adv --keyserver keyserver.ubuntu.com --recv-keys E298A3A825C0D65DFD57CBB651716619E084DAB9
$ sudo add-apt-repository 'deb https://cloud.r-project.org/bin/linux/ubuntu bionic-cran40/'
$ sudo apt update
$ sudo apt install r-base
Recently, I ran into an issue with cellassign when analyzing single cell datasets. I suspected that reverting to previous R versions might resolve the problem.
Plus, the GPG key added earlier is raising errors everytime I invoke sudo apt update, which is quite annoying. So I figured it might be better to re-build R from source.
Uninstall previously built R versions
First check GPG Keys added by previous R built and remove it.
$ sudo apt-key list
pub rsa4096 2019-06-11 [SC]
4A0C 1931 1880 3EB4 A561 E569 B3CF 35C3 15B5 5A9F
uid [ unknown] Launchpad PPA for cran
/etc/apt/trusted.gpg.d/ubuntu-keyring-2012-archive.gpg
$ sudo apt-key del "4A0C 1931 1880 3EB4 A561 E569 B3CF 35C3 15B5 5A9F"
Then remove cran repository from apt repository list:
$ sudo vim /etc/apt/sources.list # Manually edit, remove entries related to R or cran
Before uninstalling R, we might want to check what packages are currently installed and back them up, in case they’ll be needed later.
$ R
R version 4.0.3 (2020-10-10) -- "Bunny-Wunnies Freak Out"
Copyright (C) 2020 The R Foundation for Statistical Computing
Platform: x86_64-pc-linux-gnu (64-bit)
> .libPaths()
[1] "/home/luolab/R/x86_64-pc-linux-gnu-library/4.0"
[2] "/usr/local/lib/R/site-library"
[3] "/usr/lib/R/site-library" # This one is empty
[4] "/usr/lib/R/library"
In shell, backup the content listed in these paths.
$ cp -r /home/luolab/R/x86_64-pc-linux-gnu-library/ /media/luolab/4A9623FA9623E563/R_bak/home.luolab.R.x86_64-pc-linux-gnu-library # Moving to 120G SSD
$ cp -r /usr/local/lib/R/site-library/ /media/luolab/4A9623FA9623E563/R_bak/usr.local.lib.R.site-library
$ cp -r /usr/lib/R/library/ /media/luolab/4A9623FA9623E563/R_bak/usr.lib.R.library
Now actually uninstall R. In shell, do:
$ sudo apt-get --purge remove r-base-core
$ sudo apt-get autoremove
Make sure R is properly removed:
$ R
Command 'R' not found, but can be installed with:
sudo apt install r-base-core
Install R from precompiled binaries
I will use the method listed in this page for installation of R: Install R*
As a prerequisite, enable additional repositories for third-party or source packages:
$ sudo apt-get update
$ sudo apt-get install gdebi-core
Specify R version.
export R_VERSION=4.0.5
Download and install the desired version of R.
curl -O https://cdn.rstudio.com/r/ubuntu-1804/pkgs/r-${R_VERSION}_1_amd64.deb
sudo gdebi r-${R_VERSION}_1_amd64.deb
Verify R installation:
/opt/R/${R_VERSION}/bin/R --version
To ensure that R is available on the default system PATH variable, create symbolic links to the version of R just installed.
sudo ln -s /opt/R/${R_VERSION}/bin/R /usr/local/bin/R
sudo ln -s /opt/R/${R_VERSION}/bin/Rscript /usr/local/bin/Rscript
Next, time to install multiple versions of R. Repeat the above steps to specify, download, and install a different version of R alongside existing versions.
I re-configured with export R_VERSION=3.6.2 and executed the steps above to have R-3.6.2 installed.
Switching between R versions
Method 1
The tutorial has a note at the symlink step, saying:
This step only applies to the first installation of R on a given system. For subsequent installations, this section should be skipped.
I suspect that overriding the symlink with later installed R versions would allow switching between different R versions. So I did:
$ export R_VERSION=3.6.2
$ sudo rm /usr/local/bin/R /usr/local/bin/Rscript
$ sudo ln -s /opt/R/${R_VERSION}/bin/R /usr/local/bin/R # ${R_VERSION}=3.6.2
$ sudo ln -s /opt/R/${R_VERSION}/bin/Rscript /usr/local/bin/Rscript
After doing this, R indeed switched to 3.6.2, which is my desired version.
$ which R
/usr/local/bin/R
$ file /usr/local/bin/R
/usr/local/bin/R: symbolic link to /opt/R/3.6.2/bin/R
$ R --version
R version 3.6.2 (2020-02-29) -- "Holding the Windsock"
Copyright (C) 2020 The R Foundation for Statistical Computing
Platform: x86_64-pc-linux-gnu (64-bit)
Method 2
Another way of switching between different R versions is to get Rstudio recognize different R executables.
Referring to this post: Changing R versions for the RStudio Desktop IDE, we know that on Linux systems, Rstudio use the version of R pointed to by the output of which R.
To override which version of R is used, we set RSTUDIO_WHICH_R environment variable to the R executable that we want to run against. For example, in terminal:
$ export RSTUDIO_WHICH_R=/opt/R/3.6.2/bin/R
And within the same terminal, launch Rstudio by typing:
$ rstudio
Because the RSTUDIO_WHICH_R is a temporary variable, it is only available in the activated shell instance. Calling rstudio in a new terminal will still point to the default R executable, which in our case is /usr/local/bin/R, which ,by default, points to /opt/R/4.0.5/bin/R.
Now that we can work with multiple R versions, it’s time to build some R packages. Hopefully cellassign can be run properly this time.
cellassign depends on tensorflow which is another nasty built experience. I’ll write about it next time.
2021.12.3 Updates:
Previously in this post I installed R-3.6.3. Now I strongly discourage this practice. For configuration of cellassign, build R-3.6.2 instead of R-3.6.3, because some of the package dependencies (RcppAnnoy) has bad compatibility with R-3.6.3.
It turns out that cellassign still fails after reverting to R-3.6.3!
The error encountered with cellassign turns out to be a compatibility issue of R tensorflow. To avoid the error, build R tensorflow with devtools and explicitly specify a version. See this issue.
References:
Installing multiple versions of R on Linux*
R Installation and Administration
Using multiple versions of R with RStudio Workbench / RStudio Server Pro
Last modified on 2021-12-02