This function calculates the test coverage for a development package on the
path. By default it runs only the package tests, but it can also run
vignette and example code.
package_coverage(
path = ".",
type = c("tests", "vignettes", "examples", "all", "none"),
combine_types = TRUE,
relative_path = TRUE,
quiet = TRUE,
clean = TRUE,
line_exclusions = NULL,
function_exclusions = NULL,
code = character(),
install_path = temp_file("R_LIBS"),
...,
allow_failures = getOption("covr.allow_failures", FALSE),
show_code_output = FALSE
)A coverage object. If combine_types = FALSE and more than one
type is run, a list of class coverages with one coverage object for
each type, named by type.
file path to the package, or to an installed package.
run the package ‘tests’, ‘vignettes’, ‘examples’, ‘all’, or ‘none’. The default is ‘tests’.
If TRUE (the default) the coverage for all types
is simply summed into one coverage object. If FALSE separate objects
are used for each type of coverage.
whether to output the paths as relative or absolute paths. If a string, it is interpreted as a root path and all paths will be relative to that root.
whether to load and compile the package quietly, useful for debugging errors.
whether to clean temporary output files after running, mainly
useful for debugging errors. If coverage fails, or failures are reported as
warnings with allow_failures = TRUE, install_path is kept even when
clean = TRUE, so the test output can be inspected.
a named list of files with the lines to exclude from each file.
a vector of regular expressions matching function
names to exclude. Example print\\\. to match print methods.
A character vector of additional test code to run.
The path the instrumented package will be installed to
and tests run in. By default it is a path in the R sessions temporary
directory. It can sometimes be useful to set this (along with clean = FALSE) to help debug test failures. It is not removed if coverage fails.
Additional arguments passed to tools::testInstalledPackage().
Use outDir to save the output of the tests and examples to a directory
which is kept after coverage finishes; by default it is written to
install_path.
If TRUE, failures or errors when running tests,
vignettes, examples or code are reported as warnings (with class
covr_warning) and coverage is still returned. Coverage may be incomplete
in this case, as code which would have run after the failure is not
tracked. The default, FALSE, throws an error (with class covr_error)
on the first failure. The default can be changed with the
covr.allow_failures option, e.g. options(covr.allow_failures = TRUE).
If TRUE, the output of code is shown on the
console while it runs, e.g. to follow it in the logs of a CI job. The
default, FALSE, only saves it to a file in install_path, which is shown
if code fails.
This function uses tools::testInstalledPackage() to run the
code, if you would like to test your package in another way you can set
type = "none" and pass the code to run as a character vector to the
code parameter.
Coverage is also recorded in other R processes that load the package, such
as callr sessions, future or parallel cluster workers, or
Shiny apps run by shinytest2. Each process saves its coverage when it
exits normally, so the coverage is lost if a process is killed (e.g. with
SIGKILL or SIGTERM) or is still running when the tests finish, and
package_coverage() warns (with class covr_unsaved_trace). Stop these
processes gracefully and wait for them to exit, e.g. call $stop() on a
shinytest2 AppDriver, which waits longer for the app to exit when
run by covr. See vignette("shiny", package = "covr") for measuring the
coverage of Shiny apps.
path can also be an installed package, e.g.
system.file(package = "pkg"), to compute coverage without the package
source. The package must have been installed with its tests and source
references, e.g. with
remotes::install_github("user/pkg", INSTALL_opts = c("--install-tests", "--with-keep.source")). File names are those of the source it was
installed from, and compiled code isn't covered.
The coverage of Rust code, built by cargo into a static library that is
linked into the package, is measured with LLVM's source-based coverage.
This needs the llvm tools of the Rust toolchain
(rustup component add llvm-tools), or for Rust installed by a Linux
distribution, those of the LLVM version rustc uses (e.g.
apt-get install llvm-18); without them package_coverage()
warns and reports no coverage for the Rust code. covr instruments the
crates of the package's cargo workspace by having cargo run rustc through a
wrapper (with RUSTC_WORKSPACE_WRAPPER) that adds -C instrument-coverage,
so it works however the package sets RUSTFLAGS, and dependencies are
built as usual. Vendored crates, including the C code they compile, and
code generated in cargo's target directory aren't included. The code that
macros such as #[extendr] and #[savvy] generate is counted on the
macro's line, so the line isn't covered if parts of the generated code
never run, e.g. the handling of arguments that can't be converted. The
coverage of Rust code isn't measured on Windows, as rustup's Rust for the
x86_64-pc-windows-gnu target that Rtools uses has no profiler runtime.
Parallelized code using parallel's mcparallel() needs to
use a patched parallel:::mcexit. This is done automatically if the
package depends on parallel, but can also be explicitly set using the
global option covr.fix_parallel_mcexit or the environment variable
COVR_FIX_PARALLEL_MCEXIT.
exclusions() For details on excluding parts of the
package from the coverage calculations.