Testing & coverage¶

Testing is mandatory and drives development (TDD): new behaviour gets a test alongside it, not after the fact.

Unit tests (tests/unit) cover each behaviour in isolation, including the Numbers, Letters, and Conundrum game logic in countdown::solver.
Integration tests (tests/integration) exercise the file-backed dictionary and full solve pipelines together.
App tests (tests/app) exercise the Qt-facing Solver facade with Qt Test.
QML tests (tests/qml) drive the real Countdown QML module (Numbers/Letters/Conundrum/Settings pages) headlessly with Qt Quick Test, exercising the QML → solver → library round trip end to end.

The Catch2 suites are registered with CTest via catch_discover_tests, so each TEST_CASE is an individually runnable CTest; the Qt Test/Quick Test suites each register as a single CTest entry covering every case in that binary.

ctest --preset windows-msvc-debug

Coverage¶

A dedicated linux-gcc-coverage preset builds everything with --coverage (gcov) instrumentation:

$ cmake --preset linux-gcc-coverage -DCMAKE_PREFIX_PATH=<qt>/6.8.3/gcc_64
$ cmake --build --preset linux-gcc-coverage
$ QT_QPA_PLATFORM=offscreen ctest --preset linux-gcc-coverage
$ gcovr --root . --filter 'src/.*' --exclude 'src/app/main\.cpp' \
    --exclude-throw-branches --exclude-unreachable-branches \
    --gcov-ignore-errors=no_working_dir_found \
    --object-directory build/linux-gcc-coverage --print-summary

If your default gcov doesn't match the compiler that built the coverage data (e.g. building with g++-15 but gcov resolves to an older default), add --gcov-executable gcov-<N> — a version mismatch doesn't error, it just silently reports 0% coverage.

CI runs this on every push/PR (Coverage (Linux, GCC) in ci.yml, pinning --gcov-executable gcov-15 to match the g++-15 it builds with) and fails the job below 80% line or branch coverage.