/rules_r

R rules for Bazel

Primary LanguageStarlarkApache License 2.0Apache-2.0

R Rules for Bazel Tests

General Information

Rules

Repository Rules

Container Rules

Overview

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. We use these rules internally at GRAIL to build 400+ R packages from CRAN and Bioconductor.

Getting started

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:

# Change master to the git tag you want.
http_archive(
    name = "com_grail_rules_r",
    strip_prefix = "rules_r-master",
    urls = ["https://github.com/grailbio/rules_r/archive/master.tar.gz"],
)

load("@com_grail_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("@com_grail_rules_r//R:defs.bzl",
     "r_pkg", "r_library", "r_unit_test", "r_pkg_test")

There are also convenience macros available which create multiple implicit targets:

load("@com_grail_rules_r//R:defs.bzl", "r_package")

and

load("@com_grail_rules_r//R:defs.bzl", "r_package_with_test")

Configuration

The following software must be installed on your system:

1. bazel (v3.0.0 or above)
2. R (3.4.3 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.

For each package, you can also specify a different Makevars file that can be used to have finer control over native code compilation. For macOS, the Makevars file used as default helps find gfortran. The site-wide Makevars files are configured by default in the toolchains.

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

Note

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.

External packages

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("@com_grail_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/3.11/bioc",
        "BioCann": "https://bioconductor.org/packages/3.11/data/annotation",
        "BioCexp": "https://bioconductor.org/packages/3.11/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/grail-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}')
EOF

Examples

Some examples are available in the tests directory of this repo.

Also see Razel scripts that provide utility functions to generate BUILD files and WORKSPACE rules.

Contributing

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.

Known Issues

Please check open issues at the github repo.

Rules

r_pkg

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, stamp)

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.

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

List of files, required

Source files to be included for building the package.

pkg_name

String; optional

Name of the package if different from the target name.

deps

List of labels; optional

R package dependencies of type `r_pkg`.

cc_deps

List of labels; optional

cc_library dependencies for this package.

build_args

List of strings; default ["--no-build-vignettes", "--no-manual"]

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

List of strings; optional

Additional arguments to supply to R CMD INSTALL.

config_override

File; optional

Replace the package configure script with this file.

roclets

List of strings; optional

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

List of labels; optional

roxygen2 or devtools dependency for running roclets.

makevars

File; optional

Additional Makevars file supplied as R_MAKEVARS_USER.

env_vars

Dictionary; optional

Extra environment variables to define for building the package.

inst_files

Label keyed Dictionary; optional

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

List of labels; optional

Executables that code in this package will try to find in the system.

build_tools

List of labels; optional

Executables that native code compilation will try to find in the system.

metadata

String keyed Dictionary; optional

Metadata key-value pairs to add to the DESCRIPTION file before building. Build status variables can be substituted when enclosed within `{}`.

stamp

Bool; default False

Include the stable status file when substituting values in the metadata. The volatile status file is always included.

r_library

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

List of labels, required

Package (and dependencies) to install.

library_path

String; optional

If different from system default, default library location for installation. For runtime overrides, use bazel run [target] -- -l [path].

r_unit_test

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

Label; required

R package (of type r_pkg) to test.

suggested_deps

List of labels; optional

R package dependencies of type `r_pkg`.

env_vars

Dictionary; optional

Extra environment variables to define before running the test.

tools

List of labels; optional

Executables to be made available to the test.

data

List of labels; optional

Data to be made available to the test.

r_pkg_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

Label; required

R package (of type r_pkg) to test.

suggested_deps

List of labels; optional

R package dependencies of type `r_pkg`.

check_args

List of strings; default ["--no-build-vignettes, "--no-manual"]

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

Dictionary; optional

Extra environment variables to define before running the test.

tools

List of labels; optional

Executables to be made available to the test.

data

List of labels; optional

Data to be made available to the test.

r_binary

r_binary(name, src, deps, data, env_vars, tools, rscript_args, script_args, stamp)

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

File; required

An Rscript interpreted file, or file with executable permissions.

deps

List of labels; optional

Dependencies of type r_binary, r_pkg, or r_library.

data

List of labels; optional

Files needed by this rule at runtime.

env_vars

Dictionary; optional

Extra environment variables to define before running the binary.

tools

List of labels; optional

Executables to be made available to the binary.

rscript_args

List of strings; optional

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

List of strings; optional

A list of arguments to pass to the src script.

stamp

Bool; default False

Include the stable status file in the runfiles of the binary. The volatile status file is always included.

r_test

r_test(name, src, deps, data, env_vars, tools, rscript_args, script_args, stamp)

This is identical to r_binary but is run as a test.

r_markdown

r_markdown(name, src, deps, data, env_vars, tools, rscript_args, script_args, stamp,
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_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

String, default R

Path to R.

rscript

String, default Rscript

Path to Rscript.

version

String; optional

If provided, ensure version of R matches this string in x.y form.

args

List of strings; default ["--no-save", "--no-site-file", "--no-environ"]

Arguments to R and Rscript, in addition to `--slave --no-restore --no-init-file`.

makevars_site

Label; optional

Site-wide Makevars file.

env_vars

Dictionary; optional

Environment variables for BUILD actions.

tools

List of labels; optional

Additional tools to make available in PATH.

files

List of labels; optional

Additional files available to the BUILD actions.

system_state_file

Label; optional

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.

Repository Rules

r_repository

r_repository(urls, strip_prefix, type, sha256, build_file)

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

List of strings; required

URLs from which the package source archive can be fetched.

strip_prefix

String; optional

The prefix to strip from all file paths in the archive.

type

String; optional

Type of the archive file (zip, tgz, etc.).

sha256

String; optional

sha256 checksum of the archive to verify.

build_file

File; optional

Optional BUILD file for this repo. If not provided, one will be generated.

razel_args

Dictionary; optional

Other arguments to supply to buildify function in razel.

r_repository_list

r_repository_list(package_list, build_file_overrides, remote_repos, other_args)

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

File; required

CSV containing packages with name, version and sha256; with a header.

build_file_overrides

File; optional

CSV containing package name and BUILD file path; with a header.

remote_repos

Dictionary; optional

Repos to use for fetching the archives.

other_args

Dictionary; optional

Other arguments to supply to generateWorkspaceMacro function in razel.

r_rules_dependencies

load("@com_grail_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 for macOS.

r_coverage_dependencies

load("@com_grail_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.

r_register_toolchains

load("@com_grail_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

String, optional

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

Bool; default True

Fail if R is not found on the host system.

makevars_site

Bool; default True

Generate a site-wide Makevars file.

version

String; optional

version attribute value for r_toolchain.

args

List of strings; default ["--no-save", "--no-site-file", "--no-environ"]

args attribute value for r_toolchain.

tools

List of strings; optional

tools attribute value for r_toolchain.