Code Coverage in Gerrit

Tests are critical because they find bugs and regressions, enforce better designs and make code easier to maintain. Code coverage helps you ensure your tests are thorough.

Chromium CLs can show a line-by-line breakdown of test coverage. You can use it to ensure you only submit well-tested code.

To see code coverage for a Chromium CL, trigger a CQ dry run, and once the builds finish and code coverage data is processed successfully, look at the change view to see absolute and incremental code coverage percentages:

code_coverage_percentages

Absolute coverage percentage is the percentage of lines covered by tests out of all the lines in the file, while incremental coverage percentage only accounts for newly added or modified lines. Both these coverage metrics, are further classified into Unit Tests coverage(coverage from just unit tests) and All Tests coverage(covererd by all tests running in CQ, including unit tests). All Tests coverage is a superset of Unit Tests coverage.

To further dig into specific lines that are not covered by tests, look at the right column of the side by side diff view, and specifically notice the background color of each line number, where a light orange color indicates missing coverage and a light blue color indicates existing coverage. Moreover hovering over the line number shows an informative tooltip with “Not covered by tests” or “Covered by tests” text respectively. It only shows All Tests Coverage right now

code_coverage_annotations

Code coverage data is shared between patchsets that are commit-message-edit or trivial-rebase away, however, if a newly uploaded patchset has non-trivial code change, a new CQ dry run must be triggered before coverage data shows up again.

The code coverage tool supports coverage for C/C++, JAVA and Javascript on all major platforms(Linux, MacOS, Windows, Android, iOS and ChromeOS)

CLs Blocked Due to Low Coverage

For some teams in Chrome, we have turned on a coverage check, which blocks a CL from submission if the incremental coverage is below a preset threshold(default = 70%). CLs with insufficient test coverage have a CodeCoverage-1 label added to them, which prevents them from being submitted. Also, a descriptive message is added to the CL, notifying developer of why the CL was blocked, and how to resolve it. low_coverage_message

Once the tests are added, another run of coverage builders (through CQ+1 or CQ+2) changes the label to CodeCoverage+1, allowing CLs to proceed with submission.

Devs can also choose to bypass this block, in case they think they are being unfairly punished. They can do so by adding a Low-Coverage-Reason: reason footer to the change description. The reason string should mention the category the bypass reason belongs to. For e.g. Low-Coverage-Reason: TRIVIAL_CHANGE This change contains only minor cosmetic changes. (TRIVIAL_CHANGE is the category)

Available category choices are:

TRIVIAL_CHANGE: CL contains mostly minor changes e.g. renaming, file moves, logging statements, simple interface definitions etc.
TESTS_ARE_DISABLED: Corresponding tests exist, but they are currently disabled.
TESTS_IN_SEPARATE_CL: Developer plan to write tests in a separate CL (Should not be exercised often as per best practices)
HARD_TO_TEST: The code under consideration is hard to test. For e.g. Interfaces with system, real hardware etc.
COVERAGE_UNDERREPORTED: To be used when the developer thinks that tests exist, but corresponding coverage is missing.
LARGE_SCALE_REFACTOR: The current change is part of a large scale refactor. Should explain why the refactor shouldn't have tests.
EXPERIMENTAL_CODE: The current code is experimental and unlikely to be released to users.
OTHER: None of the above categories are the right fit

In case the developer doesn't specify the coverage category as prescribed, a warning will be shown in the UI, with details on how to fix impropery_formatted_coverage_footer

Contacts

Reporting problems

For any breakage report and feature requests, please file a bug.

Mailing list

For questions and general discussions, please join code-coverage group.

FAQ

Why is coverage not shown even though the try job finished successfully?

There are several possible reasons:

A particular source file/test may not be available on a particular project or platform. As of now, only chromium/src project and Linux platform is supported.
There is a bug in the pipeline. Please file a bug to report the breakage.

How does it work?

Please refer to code_coverage.md for how code coverage works in Chromium in general, and specifically, for per-CL coverage in Gerrit, the clang_code_coverage_wrapper is used to compile and instrument ONLY the source files that are affected by the CL for the sake of performance and a chromium-coverage Gerrit plugin is used to display code coverage information in Gerrit.