SQLiteCpp CI Workflows

Common steps

• Checkout code.
• Initialize submodules: git submodule update --init --recursive.
• Configure build directory (build or builddir).
• Build and run tests with verbose output.

GitHub Actions (CMake)

• Matrix across Windows (MSVC/MinGW), Ubuntu, macOS.
• Keep the latest-MSVC Windows build in Debug mode.
• Complement it with two Visual Studio 2022 Release builds for Win32/x86: one shared and one static. Do not recreate older or broader Visual Studio matrices.
• CMake config includes:
  • -DBUILD_SHARED_LIBS=ON for standard configurations; VS2022 Win32/x86 Release covers both ON and OFF.
  • -DSQLITECPP_BUILD_TESTS=ON
  • -DSQLITECPP_BUILD_EXAMPLES=ON
  • -DSQLITECPP_RUN_CPPCHECK=OFF
  • -DSQLITECPP_RUN_CPPLINT=OFF
• Tests: ctest --verbose --output-on-failure.
• Keep explicit custom assertion handler coverage in Debug and Release. The SQLITE_ENABLE_ASSERT_HANDLER CMake option defaults to OFF, so default builds of tests and examples do not compile or link the application-provided handlers. The dedicated static-library job enables this option and builds the unit tests and both examples to check their handler definitions.

GitHub Actions (Meson)

• Use pipx install meson ninja.
• Set CC, CXX, and optional linkers.
• Setup with tests/examples and sqlite3 fallback: meson setup builddir -DSQLITECPP_BUILD_TESTS=true -DSQLITECPP_BUILD_EXAMPLES=true --force-fallback-for=sqlite3.
• Build: meson compile -C builddir.
• Test: meson test -C builddir.

Coverage (Coveralls)

The Coverage workflow (.github/workflows/coverage.yml) builds with -DSQLITECPP_USE_GCOV=ON (GCC only), runs ctest (both UnitTests and Example1Run), captures with lcov while excluding /usr, googletest, sqlite3, examples and tests, and uploads to Coveralls.

To find which lines are still uncovered without rebuilding locally, query the Coveralls JSON API for the master repo (the build id comes from the repo summary):

# Overall percentage and the latest build id
curl -s https://coveralls.io/github/SRombauts/SQLiteCpp.json   # covered_percent, badge, build id

# Per-file missed counts for a build (source_files is a JSON-encoded string inside the payload)
curl -s "https://coveralls.io/builds/<BUILD_ID>/source_files.json?per_page=100"

# Exact missed line numbers for one file (array of hit counts; null/0 entries are uncovered)
curl -s "https://coveralls.io/builds/<BUILD_ID>/source.json?filename=src/Statement.cpp"

Before giving up on a miss, work out the root cause; some look like artifacts but are fixable:

• Multi-line pack expansion ((void)std::initializer_list<int>{ ... }; split across lines for variadic bind/execute_many). When the per-element call can throw, gcov puts the post-call edge on the standalone initializer_list line and reports it uncovered even though the calls run. Collapse the expansion onto a single line so gcov attributes it to the tested call expression.
• A guard that looks unreachable may still be reachable through a public constructor. The Column "Statement was destroyed" check is shadowed by checkRow() on the normal path, but Column's constructor and Statement::TStatementPtr are public, so a test can construct a Column from a null pointer directly. Check the public surface before assuming dead code.

Genuinely not worth chasing:

• Closing-brace lines that gcov marks uncovered as exception-unwinding landing pads. These are compiler-version specific (a local gcc may not even reproduce what CI's gcc reports), and removing them means contorting the function. Leave them.
• Platform-specific success paths, e.g. the normal return of loadExtension, which the only portable test cannot reach without a real loadable extension binary.

Cover the genuinely reachable lines and leave the rest.