Tarpaulin is a code coverage reporting tool for the Cargo build system, named for a waterproof cloth used to cover cargo on a ship.
Currently, Tarpaulin provides working line coverage, and while fairly reliable, may still contain minor inaccuracies in the results. A lot of work has been done to get it working on a wide range of projects, but unique combinations of packages and build features can cause issues, so please report anything you find that's wrong. Also, check out our roadmap for planned features.
On Linux, Tarpaulin's default tracing backend is still Ptrace and will only work
on x86_64 processors. This can be changed to the llvm coverage instrumentation
with --engine llvm
. For Mac and Windows, this is the default collection
method.
It can also be run in Docker, which is useful for when you don't use Linux but want to run it locally, e.g. during development. See below for how to do that.
Below is the help-text for a thorough explanation of the flags and features available:
Cargo-Tarpaulin is a tool to determine code coverage achieved via tests
Usage: cargo tarpaulin [OPTIONS] [-- <ARGS>...]
Arguments:
[ARGS]... Arguments to be passed to the test executables can be used to filter or skip certain tests
Options:
--print-rust-flags Print the RUSTFLAGS options that tarpaulin will compile your program with and exit
--print-rustdoc-flags Print the RUSTDOCFLAGS options that tarpaulin will compile any doctests with and exit
--color <WHEN> Coloring: auto, always, never [possible values: Auto, Always, Never]
--debug Show debug output - this is used for diagnosing issues with tarpaulin
-v, --verbose Show extra output
--dump-traces Log tracing events and save to a json file. Also, enabled when --debug is used
--stderr Print tarpaulin logs to stderr instead - test output will still be printed to stdout
--run-types <TYPE> Type of the coverage run [possible values: Tests, Doctests, Benchmarks, Examples, Lib, Bins, AllTargets]
--benches Test all benches
--doc Test only this library's documentation
--all-targets Test all targets (excluding doctests)
--lib Test only this package's library unit tests
--bins Test all binaries
--examples Test all examples
--tests Test all tests
--config <FILE> Path to a toml file specifying a list of options this will override any other options set
--ignore-config Ignore any project config files
--bin [<NAME>...] Test only the specified binary
--example [<NAME>...] Test only the specified example
--test [<NAME>...] Test only the specified test target
--bench [<NAME>...] Test only the specified bench target
--no-fail-fast Run all tests regardless of failure
--profile <NAME> Build artefacts with the specified profile
--ignore-tests Ignore lines of test functions when collecting coverage (default)
--no-dead-code Stops tarpaulin from building projects with -Clink-dead-code
--include-tests Include lines of test functions when collecting coverage
--ignore-panics Ignore panic macros in tests
--count Counts the number of hits during coverage
-i, --ignored Run ignored tests as well
-l, --line Line coverage
--skip-clean The opposite of --force-clean
--force-clean Adds a clean stage to work around cargo bugs that may affect coverage results
--fail-under <PERCENTAGE> Sets a percentage threshold for failure ranging from 0-100, if coverage is below exit with a non-zero code
-b, --branch Branch coverage: NOT IMPLEMENTED
-f, --forward Forwards unexpected signals to test. This is now the default behaviour
--coveralls <KEY> Coveralls key, either the repo token, or if you're using travis use $TRAVIS_JOB_ID and specify travis-{ci|pro} in --ciserver
--report-uri <URI> URI to send report to, only used if the option --coveralls is used
--no-default-features Do not include default features
--features [<FEATURES>...] Features to be included in the target project
--all-features Build all available features
--all Alias for --workspace (deprecated)
--workspace Test all packages in the workspace
-p, --packages [<PACKAGE>...] Package id specifications for which package should be build. See cargo help pkgid for more info
-e, --exclude [<PACKAGE>...] Package id specifications to exclude from coverage. See cargo help pkgid for more info
--exclude-files [<FILE>...] Exclude given files from coverage results has * wildcard
-t, --timeout <SECONDS> Integer for the maximum time in seconds without response from test before timeout (default is 1 minute)
--post-test-delay <SECONDS> Delay after test to collect coverage profiles
--follow-exec Follow executed processes capturing coverage information if they're part of your project
--release Build in release mode
--no-run Compile tests but don't run coverage
--implicit-test-threads 'Don't supply an explicit `--test-threads` argument to test executable. By default tarpaulin will infer the default rustc would pick if not ran via tarpaulin and set it
--locked Do not update Cargo.lock
--frozen Do not update Cargo.lock or any caches
--target <TRIPLE> Compilation target triple
--target-dir <DIR> Directory for all generated artifacts
--offline Run without accessing the network
--avoid-cfg-tarpaulin Remove --cfg=tarpaulin from the RUSTFLAG
-j, --jobs <N> Number of parallel jobs, defaults to # of CPUs
--rustflags <FLAGS> Rustflags to add when building project (can also be set via RUSTFLAGS env var)
--objects [<objects>...] Other object files to load which contain information for llvm coverage - must have been compiled with llvm coverage instrumentation (ignored for ptrace)
-Z [<FEATURES>...] List of unstable nightly only flags
-o, --out [<FMT>...] Output format of coverage report [possible values: Json, Stdout, Xml, Html, Lcov]
--engine <ENGINE> Coverage tracing backend to use [possible values: Auto, Ptrace, Llvm]
--output-dir <PATH> Specify a custom directory to write report files
--command <CMD> cargo subcommand to run. So far only test and build are supported [possible values: Test, Build]
-r, --root <DIR> Calculates relative paths to root directory. If --manifest-path isn't specified it will look for a Cargo.toml in root
--manifest-path <PATH> Path to Cargo.toml
--ciserver <SERVICE> CI server being used, if unspecified tarpaulin may automatically infer for coveralls uploads
--fail-immediately Option to fail immediately after a single test fails
-h, --help Print help
-V, --version Print version
If your tests or application make use of unix signals they may not work with
ptrace instrumentation in Tarpaulin. This is because Tarpaulin relies on the
sigtrap signal to catch when the instrumentation points are hit. The
--forward
option results in forwarding the signals from process stops not
caused by SIGSTOP, SIGSEGV or SIGILL to the test binary.
Despite generally being far more accurate there are some nuances with the LLVM coverage instrumentation.
- If a test has a non-zero exit code coverage data isn't returned
- Some areas of thread unsafety
- Unable to handle fork and similar syscalls (one process will overwrite another's profraw file)
In these cases coverage results may differ a lot between ptrace and llvm and llvm
coverage may be a worse choice. Things like doc tests with the should_panic
attribute or --no-fail-fast
won't report any coverage because of non-zero
exit codes and if you use these and want coverage data from them you should
avoid the llvm coverage backend.
Below is a list of features currently implemented. As Tarpaulin loads binary files into memory and parses the debugging information, different setups could lead to coverage not working. In this instance, please raise an issue detailing your setup and an example project and I'll attempt to fix it (please link us to a repo and the commit containing your project and paste the verbose output).
- Line coverage
- Full compatibility with cargo test CLI arguments
- Uploading coverage to https://coveralls.io or https://codecov.io
- HTML report generation and other coverage report types
- Coverage of tests, doctests, benchmarks and examples possible
- Excluding irrelevant files from coverage
- Config file for mutually exclusive coverage settings (see
Config file
section for details)
Issues, feature requests and pull requests are always welcome! For a guide on how to approach bugs found in Tarpaulin and add features please check CONTRIBUTING. If you're having any troubles also look to our TROUBLESHOOTING
Rust 1.23 introduced a regression in the compiler affecting Tarpaulin's accuracy. If you see missing lines or files, check your compiler version.
Tarpaulin is a command-line program, you install it into your development environment with cargo install:
cargo install cargo-tarpaulin
When using the Nix package manager, the nixpkgs.cargo-tarpaulin
package can be used.
This ensures that Tarpaulin will be built with the same rust version as the rest of your packages.
You can also use cargo-binstall:
cargo binstall cargo-tarpaulin
When Tarpaulin runs your tests it strives to run them in the same environment as if they were run via cargo test. To achieve this it sets the following environment variables when executing the test binaries:
- RUST_BACKTRACE - When --verbose flag is used
- CARGO_MANIFEST_DIR - Path to Cargo.toml From --root | --manifest-path or guessed from the current or parent directory
- CARGO_PKG_NAME - From Cargo.toml
- CARGO_PKG_AUTHORS - From Cargo.toml
- CARGO_PKG_VERSION - From Cargo.toml
- LLVM_PROFILE_FILE - Used for LLVM coverage
For Tarpaulin to construct the Cargo environment correctly, Tarpaulin needs to find Cargo.toml by either:
- Using --root or --manifest-path or
- By invoking Cargo from the current working directory within the project holding Cargo.toml manifest or
- By invoking Cargo from a sub-directory within the project
If Cargo does not find any Cargo.toml from using either of the above methods the run will error "cargo metadata" and exit.
Several RFCs are open in rust-lang to expose more of these directly in order to avoid the issues arising out of this.
To get detailed help on available arguments when running Tarpaulin call:
cargo tarpaulin --help
Currently, no options are required, if no root directory is defined Tarpaulin will run in the current working directory.
Below is a Tarpaulin run utilising one of our example projects. This is a relatively simple project to test and if you check the test, you can see the output correctly reports the lines the test hits.
cargo tarpaulin
Jan 30 21:43:33.715 INFO cargo_tarpaulin::config: Creating config
Jan 30 21:43:33.908 INFO cargo_tarpaulin: Running Tarpaulin
Jan 30 21:43:33.908 INFO cargo_tarpaulin: Building project
Jan 30 21:43:33.908 INFO cargo_tarpaulin::cargo: Cleaning project
Compiling simple_project v0.1.0 (/home/daniel/personal/tarpaulin/tests/data/simple_project)
Finished test [unoptimized + debuginfo] target(s) in 0.51s
Jan 30 21:43:34.631 INFO cargo_tarpaulin::process_handling::linux: Launching test
Jan 30 21:43:34.631 INFO cargo_tarpaulin::process_handling: running /home/daniel/personal/tarpaulin/tests/data/simple_project/target/debug/deps/simple_project-417a21905eb8be09
running 1 test
test tests::bad_test ... ok
test result: ok. 1 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out; finished in 0.02s
Jan 30 21:43:35.563 INFO cargo_tarpaulin::report: Coverage Results:
|| Uncovered Lines:
|| src/lib.rs: 6
|| src/unused.rs: 4-6
|| Tested/Total Lines:
|| src/lib.rs: 3/4
|| src/unused.rs: 0/3
||
42.86% coverage, 3/7 lines covered
Tarpaulin can also report the change in coverage for each file between runs. If the tests were updated in the previous example to cover all the lines we would expect the following output.
cargo tarpaulin
Jan 30 21:45:37.611 INFO cargo_tarpaulin::config: Creating config
Jan 30 21:45:37.623 INFO cargo_tarpaulin: Running Tarpaulin
Jan 30 21:45:37.623 INFO cargo_tarpaulin: Building project
Jan 30 21:45:37.623 INFO cargo_tarpaulin::cargo: Cleaning project
Compiling simple_project v0.1.0 (/home/daniel/personal/tarpaulin/tests/data/simple_project)
Finished test [unoptimized + debuginfo] target(s) in 0.40s
Jan 30 21:45:38.085 INFO cargo_tarpaulin::process_handling::linux: Launching test
Jan 30 21:45:38.085 INFO cargo_tarpaulin::process_handling: running /home/daniel/personal/tarpaulin/tests/data/simple_project/target/debug/deps/simple_project-417a21905eb8be09
running 2 tests
test unused::blah ... ok
test tests::bad_test ... ok
test result: ok. 2 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out; finished in 0.02s
Jan 30 21:45:38.990 INFO cargo_tarpaulin::report: Coverage Results:
|| Uncovered Lines:
|| src/lib.rs: 6
|| Tested/Total Lines:
|| src/lib.rs: 3/4 +0.00%
|| src/unused.rs: 3/3 +100.00%
||
85.71% coverage, 6/7 lines covered, +42.86% change in coverage
Hint: if using coveralls.io with travis-ci run with the options
--ciserver travis-ci --coveralls $TRAVIS_JOB_ID
. The coveralls.io repo-token
is mainly designed for private repos and it won't generate a badge for the
coverage results submitted (although you can still see them on the coveralls
web interface). For an example of a project using Tarpaulin, you can check out
my crate keygraph-rs.
Tarpaulin allows you to ignore modules or functions using attributes. Below is an example of ignoring the main function in a project:
#[cfg(not(tarpaulin_include))]
fn main() {
println!("I won't be included in results");
}
// Also supports the nightly rustc `coverage(off)` attribute.
#[coverage(off)]
fn not_included() {
}
Unfortunately, due to the unexpected cfg warnings cargo now emits you will
likely want to add the recommended lints to your Cargo.toml
, or utilise any
existing build scripts. If you're using a nightly compiler then making use of
unstable coverage attributes may be preferable.
However, the skip attribute only allows you to exclude code from coverage
it doesn't change the code present in the binaries or what tests are run.
Because of this, --cfg=tarpaulin
is used when building your project for
Tarpaulin allowing you to also conditionally include/exclude code from
compilation entirely. For example to have a test that isn't included in
the test binaries when built with Tarpaulin and cannot be run just do:
#[test]
#[cfg(not(tarpaulin))]
fn big_test_not_for_tarpaulin() {
// Something that would be very slow in tarpaulin or not work
}
If you still want the test included in the binary and ignored by default you can use:
#[test]
#[cfg_attr(tarpaulin, ignore)]
fn ignored_by_tarpaulin() {
}
There is also nightly support for using tool attributes with Tarpaulin for skip. For example:
#![feature(register_tool)]
#![register_tool(tarpaulin)]
#[tarpaulin::skip]
fn main() {
println!("I won't be in coverage stats");
}
As Tarpaulin changes the RUSTFLAGS
when building tests sometimes rebuilds of
test binaries can't be avoided. There is also a --clean
and --skip-clean
argument, the default has been changed at times to avoid issues with incremental
compilation when changing RUSTFLAGS
. If you aim to reduce the amount of
unnecessary recompilation attempting to add the --skip-clean
flag should be
the first step. After that you can either:
- Use
cargo tarpaulin --print-rust-flags
and use those flags for dev and coverage - Use
--target-dir
when running Tarpaulin and have a coverage build and dev build
Tarpaulin aims to be easy to add to your CI workflow. With well-tested support for Travis-CI it also supports sending CI specific meta-data to coveralls.io for Circle, Semaphore, Jenkins and Codeship (though only Jenkins has been tested).
You can also use Tarpaulin on Azure, check out crate-ci/azure-pipelines for an example config.
The expected most common use case is launching coverage via a CI service to upload to a site like codecov or coveralls. Given the built-in support and ubiquity of travis-ci it seems prudent to document the required steps here for new users. To follow these steps you'll first need a travis-ci and a project setup for your coverage reporting site of choice.
We recommend taking the minimal rust .travis.yml, installing the libssl-dev
dependency Tarpaulin has and then running Tarpaulin with the version of
rustc you require. Tarpaulin is installed in before_cache
to allow it to be cached
and prevent having to reinstall every Travis run. You can also replace cargo test
with a verbose run of Tarpaulin to see the test results as well as coverage output.
Tarpaulin is run after success as there are still some unstable features which could
cause coverage runs to fail. If you don't rely on any of these features you can
alternatively replace cargo test
with a call to cargo tarpaulin
.
For codecov.io you'll need to export CODECOV_TOKEN
there are instructions on this in
the settings of your codecov project.
language: rust
# tarpaulin has only been tested on bionic and trusty other distros may have issues
dist: bionic
addons:
apt:
packages:
- libssl-dev
cache: cargo
rust:
- stable
- beta
- nightly
matrix:
allow_failures:
- rust: nightly
before_script: |
if [[ "$TRAVIS_RUST_VERSION" == stable ]]; then
cargo install cargo-tarpaulin
fi
script:
- cargo clean
- cargo build
- cargo test
after_success: |
if [[ "$TRAVIS_RUST_VERSION" == stable ]]; then
# Uncomment the following line for coveralls.io
# cargo tarpaulin --ciserver travis-ci --coveralls $TRAVIS_JOB_ID
# Uncomment the following two lines create and upload a report for codecov.io
# cargo tarpaulin --out xml
# bash <(curl -s https://codecov.io/bash)
fi
If you rely on certain nightly features you may need to change the before_script
to
before_cache
to force Tarpaulin to reinstall each time. However, if it can be avoided it
will speed up your CI runs.
Alternatively, there are the prebuilt docker images or you can use cargo-binstall.
The prebuilt binary is built using github actions ubuntu:latest image, because of this it doesn't work on xenial or trusty, but it works on bionic. You should still keep the rest of the recommended travis settings.
File .github/workflows/coverage.yml
Example how to run coverage within docker
with seccomp
in GitHub Actions and push the result
to <codecov.io>.
name: coverage
on: [push]
jobs:
test:
name: coverage
runs-on: ubuntu-latest
container:
image: xd009642/tarpaulin:develop-nightly
options: --security-opt seccomp=unconfined
steps:
- name: Checkout repository
uses: actions/checkout@v2
- name: Generate code coverage
run: |
cargo +nightly tarpaulin --verbose --all-features --workspace --timeout 120 --out xml
- name: Upload to codecov.io
uses: codecov/codecov-action@v2
with:
# token: ${{secrets.CODECOV_TOKEN}} # not required for public repos
fail_ci_if_error: true
To run Tarpaulin on CircleCI you need to run Tarpaulin in docker and set the machine flag to true as shown below:
jobs:
coverage:
machine: true
steps:
- checkout
- run:
name: Coverage with docker
command: docker run --security-opt seccomp=unconfined -v "${PWD}:/volume" xd009642/tarpaulin
To get the coverage results showing up in your Gitlab pipelines add the following regex to the Test coverage
section in the gitlab job definition in .gitlab-ci.yml
:
job: ...
coverage: '/^\d+.\d+% coverage/'
Gitlab can show coverage information in the diff of a merge request. For that, use
job: ...
artifacts:
reports:
coverage_report:
coverage_format: cobertura
path: cobertura.xml
and generate a cobertura.xml
as described under Pycobertura.
For installation add cargo install cargo-tarpaulin -f
to the script section.
Tarpaulin has builds deployed to docker-hub, to run Tarpaulin on any system that has Docker, run this in your project directory:
docker run --security-opt seccomp=unconfined -v "${PWD}:/volume" xd009642/tarpaulin
This builds your project inside Docker and runs Tarpaulin without any arguments. There are also tags available for the latest version on the develop branch in stable or nightly. And versions after 0.5.6 will have the latest release built with the rust stable and nightly compilers. To get the latest development version built with rustc-nightly run the following:
docker run --security-opt seccomp=unconfined -v "${PWD}:/volume" xd009642/tarpaulin:develop-nightly
Note that the build might fail if the Docker image doesn't contain any necessary dependencies. In that case, you can install dependencies before, like this:
docker run --security-opt seccomp=unconfined -v "${PWD}:/volume" xd009642/tarpaulin sh -c "apt-get install xxx && cargo tarpaulin"
Alternatively, taking the seccomp json and setting all seccomp actions
for the personality
syscall to SCMP_ACT_ALLOW
to avoid removing all
the seccomp policies for Docker.
Tarpaulin has a config file setting where multiple coverage setups can be
encoded in a toml file. This can be provided by an argument,
by the environment variable CARGO_TARPAULIN_CONFIG_FILE
or if a
.tarpaulin.toml
or tarpaulin.toml
is present in the same directory as
the projects manifest or in the root directory that will be used unless
--ignore-config
is passed. Below is an example file:
[feature_a_coverage]
features = "feature_a"
[feature_a_and_b_coverage]
features = "feature_a feature_b"
release = true
[report]
coveralls = "coveralls_key"
out = ["Html", "Xml"]
Here we'd create three configurations, one would run your tests with
feature_a
enabled, and the other with the tests built-in release and
both feature_a
and feature_b
enabled. The last configuration uses a reserved
configuration name report
and this doesn't result in a coverage run but
affects the report output. This is a reserved feature name and any non-reporting
based options chosen will not affect the output of Tarpaulin.
For reference on available keys and their types refer to the CLI help text
at the start of the readme or src/config/mod.rs
for the concrete types
if anything is unclear. For arguments to be passed into the test binary that
follow --
in Tarpaulin use args
in the toml file.
Setting the field config
will not affect the run as it won't be parsed
for additional configuration.
For the flags --lib
, --examples
, --benches
, --tests
, --all-targets
,
--doc
, --bins
use the run-types
entry in the config file.
There are some tools available which can extend Tarpaulin functionality for other potential user needs.
Normally, Tarpaulin can't report on code coverage within the code for a
procedural macro. You'll need to add a test that expands the macro at run-time
to get those stats. The
runtime-macros
crate was made for
this purpose, and its documentation describes how to use it with Tarpaulin.
pycobertura
is a python library
for working with cobertura reports. It offers a report diffing tool as well as
its own report implementations.
To generate a cobertura.xml
simply run the following Tarpaulin command:
cargo tarpaulin --out xml
Then install pycobertura
with pip and execute the desired command.
As Tarpaulin doesn't allow you to change the name of the generated cobertura report be mindful of this if diffing reports between multiple commits.
- Branch coverage for tests
- Condition coverage for tests
- MCDC coverage reports
- LLVM coverage support
- Support for embedded targets
- OSX support
- Windows support
Tarpaulin is currently licensed under the terms of both the MIT license and the Apache License (Version 2.0). See LICENSE-MIT and LICENSE-APACHE for more details.