Guillaume Desmottes
March 24, 2021
Reading time:
Earlier this year, the Rust compiler gained support for LLVM source-base code coverage. This feature is called source-base because it operates on AST and preprocessor information directly, producing more precise coverage data compared to the traditional gcov coverage technique.
GitLab provides built-in integration of coverage information allowing for example reviewers to check if a MR is changing tested code or if it's increasing or decreasing the total coverage of the project. In this post we'll explain how to setup a CI job in a Rust project to feed source-base coverage information to GitLab.
The frst step is to add a new job to your CI pipeline, which will take care of generating the coverage reports. See the GitLab documentation if your project does not have any CI setup yet.
LLVM source-base code coverage instrumentation is currently only available in Rust nightly
so our job will use an image providing this version of the compiler. It also needs the llvm-tools-preview
component.
In order to generate the code coverage information, called raw profile, we need to set the environment variable RUSTFLAGS="-Zinstrument-coverage"
. By default, the profile is saved to a file called default.profraw
. This will be a problem if we have multiple tests as each one will override the profile of the previous one. To avoid this, we'll also define LLVM_PROFILE_FILE with a generic pattern so each test will save its profile to its own file: LLVM_PROFILE_FILE="coverage-%p-%m.profraw"
.
Once we have setup these variables we just need to run the tests as usual. Here is how a coverage CI job would look like:
coverage: image: "rustdocker/rust:nightly" stage: extras variables: RUSTFLAGS: "-Zinstrument-coverage" LLVM_PROFILE_FILE: "coverage-%p-%m.profraw" script: - rustup component add llvm-tools-preview - cargo test
Now that the Rust compiler has generated the coverage profiles we can generate a report. This is done using grcov from Mozilla which is installed using cargo install
.
We can then use it to generate a html
report that we'll export as a job artifact. Here is an example of such report for the zbus crate.
coverage: image: "rustdocker/rust:nightly" stage: extras variables: RUSTFLAGS: "-Zinstrument-coverage" LLVM_PROFILE_FILE: "coverage-%p-%m.profraw" script: - rustup component add llvm-tools-preview - cargo test # generate html report - cargo install grcov - grcov . --binary-path ./target/debug/ -s . -t html --branch --ignore-not-existing --ignore "*cargo*" -o ./coverage/ artifacts: paths: - 'coverage'
We want our coverage report to cover only the code of our current crate, not its dependencies. This is achieved by passing the --ignore "*cargo*"
flag to grcov to exclude code from the cargo
registry. Depending of your exact setup you may have to adjust it. I'd suggest to first generate a report without any --ignore
and then tweak it to exclude all the code not from your crate.
The html
report we just generated is very handy for manually checking which part of the code is covered by tests but is not usable directly by GitLab. To do so we need to feed it a cobertura report which, at the time of writing, is not yet supported by grcov.
Fortunately, it's possible to generate a lcov
report and then convert it to cobertura as suggested in the ticket.
Passing this report to GitLab as a reports
artifact will enable Test Coverage Visualization allowing reviewers to easily check if a MR is changing tested or untested code.
Here is the updated job:
coverage: image: "rustdocker/rust:nightly" stage: extras variables: RUSTFLAGS: "-Zinstrument-coverage" LLVM_PROFILE_FILE: "coverage-%p-%m.profraw" script: - rustup component add llvm-tools-preview - cargo test # generate html report - cargo install grcov - grcov . --binary-path ./target/debug/ -s . -t html --branch --ignore-not-existing --ignore "*cargo*" -o ./coverage/ # generate cobertura report for gitlab integration - pip3 install lcov_cobertura - grcov . --binary-path ./target/debug/ -s . -t lcov --branch --ignore-not-existing --ignore "*cargo*" -o coverage.lcov - python3 /usr/local/lib/python3.5/dist-packages/lcov_cobertura.py coverage.lcov artifacts: paths: - 'coverage' reports: cobertura: coverage.xml
Finally we want to tell GitLab the number of covered lines of code and the total number of lines so it can compute the coverage rate.
As grcov cannot easily produce this information yet we'll use the lcov
tool:
# output coverage summary for gitlab parsing - apt-get update && apt-get install -y lcov - lcov --summary coverage.lcov
The last step is to tell GitLab how to extract those numbers from the job logs. This is done in the CI/CD settings page, Test coverage parsing section, by setting this regular expression:
\s*lines\.*:\s*([\d\.]+%)
CI pipelines and merge requests will now display the coverage rate of the branch:
The flexibility of grcov and GitLab allowed our coverage
job to provide:
html
report as an artifact;We had to work around some current grcov limitations by using external tools to convert and parse the reports. Once grcov will have gained support for these formats the whole process will become much more straightforward.
Here is the final job, you can also check the CI of zbus and gstreamer-rs for real-life examples.
coverage: image: "rustdocker/rust:nightly" stage: extras variables: RUSTFLAGS: "-Zinstrument-coverage" LLVM_PROFILE_FILE: "coverage-%p-%m.profraw" script: - rustup component add llvm-tools-preview - cargo test # generate html report - cargo install grcov - grcov . --binary-path ./target/debug/ -s . -t html --branch --ignore-not-existing --ignore "*cargo*" -o ./coverage/ # generate cobertura report for gitlab integration - pip3 install lcov_cobertura - grcov . --binary-path ./target/debug/ -s . -t lcov --branch --ignore-not-existing --ignore "*cargo*" -o coverage.lcov - python3 /usr/local/lib/python3.5/dist-packages/lcov_cobertura.py coverage.lcov # output coverage summary for gitlab parsing - apt-get update && apt-get install -y lcov - lcov --summary coverage.lcov artifacts: paths: - 'coverage' reports: cobertura: coverage.xml
Note that this job will build and download the reporting tools at each run. A future improvement for projects running a lot of coverage reports would be to build a seperate docker container with all the tooling preinstalled, as we did in zbus
.
19/12/2024
In the world of deep learning optimization, two powerful tools stand out: torch.compile, PyTorch’s just-in-time (JIT) compiler, and NVIDIA’s…
08/10/2024
Having multiple developers work on pre-merge testing distributes the process and ensures that every contribution is rigorously tested before…
15/08/2024
After rigorous debugging, a new unit testing framework was added to the backend compiler for NVK. This is a walkthrough of the steps taken…
01/08/2024
We're reflecting on the steps taken as we continually seek to improve Linux kernel integration. This will include more detail about the…
27/06/2024
With each board running a mainline-first Linux software stack and tested in a CI loop with the LAVA test framework, the Farm showcased Collabora's…
26/06/2024
WirePlumber 0.5 arrived recently with many new and essential features including the Smart Filter Policy, enabling audio filters to automatically…
Comments (2)
Ben Widawsky:
Dec 14, 2023 at 08:53 PM
grcov does now support cobertura supports (landed shortly after this post actually)
Reply to this comment
Reply to this comment
Mark Filion:
Dec 15, 2023 at 02:09 PM
Very true Ben, thanks for catching that!
Reply to this comment
Reply to this comment
Add a Comment