fedora_clang_rdsan

updated: 2023-12-02: I modified the Dockerfile_fedora_clang file so that the latest clang version is installed

• This repository includes slight modifications of the Dockerfiles from the rhub project to reproduce CRAN ASAN-UBSAN errors on Fedora Clang (latest version) using a Github Action
• In the Github Action an .R file is used to retrieve the latest version of the existing Fedora Docker images using the Docker API. Then, the highest numeric version (and not a character string such as "latest") will be used to build the Docker Image and check for potential CRAN ASAN-UBSAN Errors.
• Finally a docker image is pushed to my Docker Account that can be used to fix potential CRAN ASAN-UBSAN errors.
• This Github Action updates (currently) the mlampros/fedora_clang_san:latest docker image once per week.

Currently, as of August 2022, there is an issue when building an ASAN-UBSAN docker image using the latest Fedora version, which seems to be related to a change in 'glibc'

A temporary solution is to include one of the following two commands in the docker run command (runtime):

• --cap-add=SYS_PTRACE --security-opt seccomp=unconfined
• --security-opt seccomp=unconfined

References (related to the 'glibc' issue)

• https://stat.ethz.ch/pipermail/r-devel/2021-April/080679.html
• https://stat.ethz.ch/pipermail/r-sig-fedora/2021-March/000732.html
• https://bugzilla.redhat.com/show_bug.cgi?id=1900021#c3
• https://bugzilla.redhat.com/show_bug.cgi?id=1900021#c21
• https://bugzilla.redhat.com/show_bug.cgi?id=1900021#c42
• https://bugzilla.redhat.com/show_bug.cgi?id=1900021#c44
• r-hub/rhub#464

Reproducing the CRAN script (so that LLVM, Clang can be built from source) using a Github Action is not an option, because the workflow requires more disk space than available (even after using a Github Action step to reduce the disk space). Thus, by using the latest Fedora version an LLVM-Clang version close to the CRAN requirement is installed so that an R package can be checked for ASAN-UBSAN errors.

How to use the ASAN-UBSAN Docker Image

There are a couple of ways that the Docker image can be used:

1. From the command line (the user can bind a directory such as 'pkg_dir', that includes an R package, in my case 'OpenImageR')

• docker pull mlampros/fedora_clang_san:latest
• docker image ls
• docker run -it mlampros/fedora_clang_san:latest clang --version
• docker run --rm -ti --security-opt seccomp=unconfined -v /home/pkg_dir:/pkg_dir mlampros/fedora_clang_san:latest
• dnf install -y openssl-devel
• RDsan
• install.packages('devtools', dependencies = TRUE, repos = 'https://cloud.r-project.org/')
• install.packages(c( 'Rcpp', 'graphics', 'grDevices', 'grid', 'shiny', 'jpeg', 'png', 'tiff', 'R6', 'RcppArmadillo', 'testthat', 'knitr', 'rmarkdown', 'covr', 'remotes' ), repos = 'https://cloud.r-project.org/')
• setwd('/pkg_dir')
• system("RDsan CMD build OpenImageR")
• system("RDsan CMD check --as-cran OpenImageR_1.2.5.tar.gz")

2. From an R session using a saved bash file (see the example_R_Sys_requirements.sh file in this repo) in a specified by the user pkg_dir directory that includes the (R & system) requirements. Assuming 'docker' and 'R' are already installed in the Operating System a user can use the following by first openning a command line window (the 'pkg_dir' is a positional parameter when using '/bin/bash' in 'docker run')

• R (begin an R session)
• system("docker pull mlampros/fedora_clang_san:latest")
• system("docker image ls")
• system("docker run -it mlampros/fedora_clang_san:latest clang --version")
• system("docker run --security-opt seccomp=unconfined -v /home/pkg_dir:/pkg_dir mlampros/fedora_clang_san:latest /bin/bash /pkg_dir/example_R_Sys_requirements.sh 'pkg_dir'")

In both cases (1. and 2.) the output CRAN .Rcheck folder is saved in the (adjusted) by the user directory '/home/pkg_dir' so that the user can fix the errors and re-run the CRAN checks.

My preferred way is 1. because once the R packages are installed in the R session, the user has only to re-adjust the code (in 'pkg_dir') and re-run the 'R CMD build & check'. Whereas, the second case (when using the 'system()' command from within an R session) requires the re-installation of the R packages after a potential adjustment of the code.

Both cases (1. and 2.) were tested on Ubuntu 18.04 and Linux Mint 18.2

Similar Projects:

• https://github.com/r-hub/rhub-linux-builders (all OS's)
• https://github.com/rocker-org/r-devel-san-clang (OS: Debian)
• https://github.com/wch/r-debug (all OS's)