These rules are used for building R packages with Bazel. Although R has an excellent package management system, there is no continuous build and integration system for entire R package repositories. An advantage of using Bazel, over a custom solution of tracking the package dependency graph and triggering builds accordingly on each commit, is that R packages can be built and tested as part of one build system in multi-language monorepos.
These rules are mature for production use.
The following assumes that you are familiar with how to use Bazel in general.
To begin, you can add the following or equivalent to your WORKSPACE file:
load("@bazel_tools//tools/build_defs/repo:http.bzl", "http_archive")
# Change master to the git tag you want.
http_archive(
name = "rules_r",
strip_prefix = "rules_r-master",
urls = ["https://github.com/grailbio/rules_r/archive/master.tar.gz"],
)
load("@rules_r//R:dependencies.bzl", "r_register_toolchains", "r_rules_dependencies")
r_rules_dependencies()
r_register_toolchains()You can load the rules in your BUILD file like so:
load("@rules_r//R:defs.bzl",
"r_pkg", "r_library", "r_unit_test", "r_pkg_test")Advanced users can also set up Gazelle to maintain the BUILD files for the R packages in their repo automatically.
The following software must be installed on your system:
1. bazel (v5.0.0 or above)
2. R (4.1.2 or above; should be locatable using the `PATH` environment variable)
NOTE: After re-installing or upgrading R, please reset the registered
toolchain with bazel sync --configure to rebuild your packages with the new
installation.
NOTE: It is possible to use R from a bazel package instead of a system
installation. See the toolchain r-toolchain-nix in the tests directory as an
example.
For each package, you can also specify a different Makevars file that can be used to have finer control over native code compilation. The site-wide Makevars files are configured by default in the toolchains, and these define the compiler toolchain to use and the flags needed for these toolchains for reproducible builds.
For macOS, this setup will help you cover the requirements for a large number of packages:
brew install gcc pkg-config icu4c openssl
For Ubuntu, this (or equivalent for other Unix systems) helps:
apt-get install pkgconf libssl-dev libxml2-dev libcurl4-openssl-dev
For no interference from other packages during the build (possibly other
versions installed manually by the user), it is recommended that packages other
than those with recommended priority be installed in the directory pointed to
by R_LIBS_USER. The Bazel build process will then be able to hide all the
other packages from R by setting a different value for R_LIBS_USER.
When moving to Bazel for installing R packages on your system, we recommend cleaning up existing machines:
sudo Rscript \
-e 'options("repos"="https://cloud.r-project.org")' \
-e 'lib <- c(.Library, .Library.site)' \
-e 'non_base_pkgs <- installed.packages(lib.loc=lib, priority=c("recommended", "NA"))[, "Package"]' \
-e 'remove.packages(non_base_pkgs, lib=lib)'
# If not set up already, create the directory for R_LIBS_USER.
Rscript \
-e 'dir.create(Sys.getenv("R_LIBS_USER"), recursive=TRUE, showWarnings=FALSE)'
For more details on how R searches different paths for packages, see libPaths.
To depend on external packages from CRAN and other remote repos, you can define the
packages as a CSV with three columns -- Package, Version, and sha256. Then use
r_repository_list rule to define R repositories for each
package. For packages not in a CRAN like repo (e.g. github), you can use
r_repository rule directly. For packages on your local system
but outside your main repository, you will have to use local_repository with
a saved BUILD file. Same for VCS repositories.
load("@rules_r//R:repositories.bzl", "r_repository", "r_repository_list")
# R packages with non-standard sources.
r_repository(
name = "R_plotly",
sha256 = "24c848fa2cbb6aed6a59fa94f8c9b917de5b777d14919268e88bff6c4562ed29",
strip_prefix = "plotly-a60510e4bbce5c6bed34ef6439d7a48cb54cad0a",
urls = [
"https://github.com/ropensci/plotly/archive/a60510e4bbce5c6bed34ef6439d7a48cb54cad0a.tar.gz",
],
)
# R packages with standard sources.
# See below for an example of how to generate the CSV package_list.
r_repository_list(
name = "r_repositories_bzl",
build_file_overrides = "@myrepo//third-party/R:build_file_overrides.csv",
package_list = "@myrepo//third-party/R:packages.csv",
remote_repos = {
"BioCsoft": "https://bioconductor.org/packages/3.14/bioc",
"BioCann": "https://bioconductor.org/packages/3.14/data/annotation",
"BioCexp": "https://bioconductor.org/packages/3.14/data/experiment",
"CRAN": "https://cloud.r-project.org",
},
)
load("@r_repositories_bzl//:r_repositories.bzl", "r_repositories")
r_repositories()
The list of all external R packages configured this way can be obtained from your shell with
$ bazel query 'filter(":R_", //external:*)'
NOTE: Periods ('.') in the package names are replaced with underscores ('_') because bazel does not allow periods in repository names.
To generate and maintain a CSV file containing all your external dependencies
for use with r_repository_list, you can use the functions in the script
repo_management.R.
For example:
script="/path/to/rules_r/scripts/repo_management.R"
package_list_csv="/path/to/output/csv/file"
packages="comma-separated list of packages you want to add to the local cache"
bioc_version="bioc_version to use, e.g. 3.11"
# This will be the cache directory for a local copy of all the packages.
# The output CSV will always reflect the state of this directory.
local_r_repo="${HOME}/.cache/r-repo"
Rscript - <<EOF
source('${script}')
pkgs <- strsplit('${packages}', ',')[[1]]
# Set ForceDownload to TRUE when switching R or Bioc versions.
# options("ForceDownload" = TRUE)
# Keep in sync with r_repository_list in WORKSPACE.
options(repos = c(
BioCsoft = "https://bioconductor.org/packages/${bioc_version}/bioc",
BioCann = "https://bioconductor.org/packages/${bioc_version}/data/annotation",
BioCexp = "https://bioconductor.org/packages/${bioc_version}/data/experiment",
CRAN = "https://cloud.r-project.org")
)
addPackagesToRepo(pkgs, repo_dir = '${local_r_repo}')
packageList('${local_r_repo}', '${package_list_csv}')
EOFSome examples are available in the tests directory of this repo.
- See tests/packages/exampleA for a barebones R package.
- See tests/packages/exampleB for a barebones R package that depends on another package.
- See tests/packages/exampleC for an R package that depends on external R packages and user native code libraries.
- See tests/packages/exampleD for an R package that depends on external R packages which have been built with bazel provided native code libraries instead of a system requirement.
Also see Razel scripts that provide utility functions to generate BUILD files
and WORKSPACE rules.
Contributions are most welcome. Please submit a pull request giving the owners of this github repo access to your branch for minor style related edits, etc. We recommend opening an issue first to discuss the nature of your change before beginning work on it.
Please check open issues at the github repo.
r_pkg(srcs, pkg_name, deps, cc_deps, build_args, install_args, config_override,
roclets, roclets_deps, makevars, env_vars, inst_files, tools, build_tools,
metadata)Rule to install the package and its transitive dependencies in the Bazel sandbox, so it can be depended upon by other package builds.
The builds produced from this rule are tested to be byte-for-byte reproducible with the same R installation. For native code compilation, the compiler flags for reproducibility are defined in the default site Makevars file in the local toolchain. If using your own toolchain, ensure that your site Makevars file has the right flags.
| Implicit output targets | |
|---|---|
name.bin.tar.gz |
Binary archive of the package. |
name.tar.gz |
Source archive of the package. |
name.so |
Shared archive of package native code; empty file if package does not have native code. |
| Attributes | |
|---|---|
srcs |
Source files to be included for building the package. |
pkg_name |
Name of the package if different from the target name. |
deps |
R package dependencies of type `r_pkg` or `r_library`. |
cc_deps |
cc_library dependencies for this package. |
build_args |
Additional arguments to supply to R CMD build. Note that building vignettes is disabled by default to not require Tex installation for users. In order to build vignettes, override this attribute, and ensure that the relevant binaries are available in your system default PATH (usually /usr/bin and /usr/local/bin) |
install_args |
Additional arguments to supply to R CMD INSTALL. |
config_override |
Replace the package configure script with this file. |
roclets |
roclets to run before installing the package. If this is non-empty, then you must specify roclets_deps as the R package you want to use for running roclets. The runtime code will check if devtools is available and use `devtools::document`, failing which, it will check if roxygen2 is available and use `roxygen2::roxygenize`. |
roclets_deps |
roxygen2 or devtools dependency for running roclets. |
makevars |
Additional Makevars file supplied as R_MAKEVARS_USER. |
env_vars |
Extra environment variables to define for building the package. |
inst_files |
Files to be bundled with the package through the inst directory. The values of the dictionary will specify the package relative destination path. For example, '' will bundle the files to the top level directory, and 'mydir' will bundle all files into a directory mydir. |
tools |
Executables that code in this package will try to find in the system. |
build_tools |
Executables that native code compilation will try to find in the system. |
metadata |
Metadata key-value pairs to add to the DESCRIPTION file before building. When text is enclosed within `{}`, bazel volatile and stable status files will be used to substitute the text. Inclusion of these files in the build has consequences on local and remote caching. Also see `stamp`. |
stamp |
Same behavior as the stamp attribute in cc_binary rule. |
r_library(pkgs, library_path)Executable rule to install the given packages and all dependencies to a user provided or system default R library. Run the target with --help for usage information.
The rule used to provide a tar archive of the library as an implicit output.
That feature is now it's own rule -- r_library_tar. See documentation for
r_library_tar rule and example usage for
container_image rule.
| Attributes | |
|---|---|
pkgs |
Package (and dependencies) to install. |
library_path |
If different from system default, default library location for installation. For runtime overrides, use bazel run [target] -- -l [path]. |
r_unit_test(pkg, suggested_deps, env_vars, tools, data)Rule to keep all deps in the sandbox, and run the provided R test scripts.
When run with bazel coverage, this rule will also produce a coverage report
in Cobertura XML format. The coverage report will contain coverage for R code
in the package, and C/C++ code in the src directory of R packages.
| Attributes | |
|---|---|
pkg |
R package (of type r_pkg) to test. |
suggested_deps |
R package dependencies of type `r_pkg` or `r_library`. |
env_vars |
Extra environment variables to define before running the test. |
tools |
Executables to be made available to the test. |
data |
Data to be made available to the test. |
r_pkg_test(pkg, suggested_deps, check_args, env_vars, tools, data)Rule to keep all deps of the package in the sandbox, build a source archive of this package, and run R CMD check on the package source archive in the sandbox.
| Attributes | |
|---|---|
pkg |
R package (of type r_pkg) to test. |
suggested_deps |
R package dependencies of type `r_pkg` or `r_library`. |
check_args |
Additional arguments to supply to R CMD build. Note that building vignettes is disabled by default to not require Tex installation for users. In order to build vignettes, override this attribute, and ensure that the relevant binaries are available in your system default PATH (usually /usr/bin and /usr/local/bin) |
env_vars |
Extra environment variables to define before running the test. |
tools |
Executables to be made available to the test. |
data |
Data to be made available to the test. |
r_binary(name, src, deps, data, env_vars, tools, rscript_args, script_args)Build a wrapper shell script for running an executable which will have all the specified R packages available.
The target can be executed standalone, with bazel run, or called from other
executables if RUNFILES_DIR is exported in the environment with
the runfiles of the root executable.
| Attributes | |
|---|---|
src |
An Rscript interpreted file, or file with executable permissions. |
deps |
Dependencies of type |
data |
Files needed by this rule at runtime. |
env_vars |
Extra environment variables to define before running the binary. |
tools |
Executables to be made available to the binary. |
rscript_args |
If src file does not have executable permissions, arguments for the Rscript interpreter. We recommend using the shebang line and giving your script execute permissions instead of using this. |
script_args |
A list of arguments to pass to the src script. |
r_test(name, src, deps, data, env_vars, tools, rscript_args, script_args)This is identical to r_binary but is run as a test.
r_markdown(name, src, deps, data, env_vars, tools, rscript_args, script_args,
render_function="rmarkdown::render", input_argument="input", output_dir_argument="output_dir",
render_args)This rule renders an R markdown through generating a stub to call the render
function. The render function and the argument names for the function are
default set for rmarkdown::render but can be customized. Note that
render_args will need to be quoted appropriately if set. This rule can be
used wherever an r_binary rule can be used.
If arguments are given on the command line when running the target, flags of
the form --arg=value are passed as keyword arguments to the render
function. The values can be arbitrary R expressions, and strings will need to
be quoted. The last argument without the prefix -- will be the output
directory, else the output directory will be the default output
directory of the render function, typically the same directory as the input
file.
r_toolchain(r, rscript, version, args, makevars_site, env_vars, tools, files, system_state_file)Toolchain to specify the tools and environment for performing build actions. Also see r_register_toolchains for how to configure the default registered toolchains.
| Attributes | |
|---|---|
r |
Absolute path to R, or name of R executable; the search path will include the directories for tools attribute. |
rscript |
Absolute path to Rscript, or name of Rscript executable; the search path will include the directories for tools attribute. |
version |
If provided, ensure version of R matches this string in x.y form. This version check is performed in the `r_pkg` and `r_binary` (and by extension, `r_test` and `r_markdown`) rules. For stronger guarantees, perform this version check when generating the `system_state_file` (see attribute below). |
args |
Arguments to R and Rscript, in addition to `--slave --no-restore --no-init-file`. |
makevars_site |
Site-wide Makevars file. |
env_vars |
Environment variables for BUILD actions. |
tools |
Additional tools to make available in PATH. |
files |
Additional files available to the BUILD actions. |
system_state_file |
A file that captures your system state. Use it to rebuild all R packages whenever the contents of this file change. This is ideally generated by a repository_rule with `configure = True`, so that a call to `bazel sync --configure` resets this file. |
r_repository(urls, strip_prefix, type, sha256, build_file, rscript)Repository rule in place of new_http_archive that can run razel to generate
the BUILD file automatically. See section on
external packages and Razel scripts.
| Attributes | |
|---|---|
urls |
URLs from which the package source archive can be fetched. |
strip_prefix |
The prefix to strip from all file paths in the archive. |
type |
Type of the archive file (zip, tgz, etc.). |
sha256 |
sha256 checksum of the archive to verify. |
build_file |
Optional BUILD file for this repo. If not provided, one will be generated. |
razel_args |
Other arguments to supply to buildify function in razel. |
rscript |
Name, path or label (must start with `@` or `//`) of the interpreter to use for running the razel script. |
r_repository_list(package_list, build_file_overrides, remote_repos, other_args, rscript)Repository rule that will generate a bzl file containing a macro, to be called
as r_repositories(), for r_repository definitions for packages in
package_list CSV. See section on external packages.
| Attributes | |
|---|---|
package_list |
CSV containing packages with name, version and sha256; with a header. |
build_file_overrides |
CSV containing package name and BUILD file path; with a header. |
remote_repos |
Repos to use for fetching the archives. |
other_args |
Other arguments to supply to generateWorkspaceMacro function in razel. |
rscript |
Name, path or label (must start with `@` or `//`) of the interpreter to use for running the razel script. |
r_version |
If provided, ensure version of R matches this string in x.y form. |
load("@rules_r//R:dependencies.bzl", "r_rules_dependencies")
r_rules_dependencies()Repository rule that provides repository definitions for dependencies of the BUILD system. One such dependency is the site-wide Makevars file.
load("@rules_r//R:dependencies.bzl", "r_coverage_dependencies")
r_coverage_dependencies()
load("@r_coverage_deps_bzl//:r_repositories.bzl", coverage_deps = "r_repositories")
coverage_deps()Repository rule that provides repository definitions for dependencies in computing code coverage for unit tests. Not needed if users already have a repository definition for the covr package.
load("@rules_r//R:dependencies.bzl", "r_register_toolchains")
r_register_toolchains(r_home, strict, makevars_site, version, args, tools)Repository rule that generates and registers a platform independent toolchain of type r_toolchain based on the user's system and environment. If you want to register your own toolchain for specific platforms, register them before calling this function in your WORKSPACE file to give them preference.
NOTE: These toolchains read your system state and cache the findings for future runs. Whenever you install a new R version, or if you want to reset the toolchain for any reason, run:
bazel sync --configure| Attributes | |
|---|---|
r_home |
A path to `R_HOME` (as returned from `R RHOME`). If not specified, the rule looks for R and Rscript in `PATH`. The environment variable `BAZEL_R_HOME` takes precendence over this value. |
strict |
Fail if R is not found on the host system. |
makevars_site |
Generate a site-wide Makevars file. |
version |
version attribute value for r_toolchain. |
args |
args attribute value for r_toolchain. |
tools |
tools attribute value for r_toolchain. |