adv-test

name	adv-test
description	Build and run the Adventure headless unit-test suite (doctest via CTest). Use after changing anything in adventure_lib (Lua engine, metrics, systems) to check for regressions before committing.

adv-test — build + run the test suite

The tests link adventure_lib and never open a window, so they run anywhere.

Steps

# build (configure first if needed: cmake --preset mingw-debug)
cmake --build build/mingw-debug

# run via CTest
ctest --test-dir build/mingw-debug --output-on-failure

For richer output, run the test exe directly (doctest reporter):

./build/mingw-debug/adventure_tests.exe            # summary
./build/mingw-debug/adventure_tests.exe -s         # list successful assertions too
./build/mingw-debug/adventure_tests.exe -tc="watchdog*"   # filter by test-case name

Performance gate (benchmarks)

The adventure_bench target is a headless micro-benchmark harness that returns nonzero if any case exceeds its ms budget. It's registered as a CTest test only for optimized builds (Debug is too noisy), so run it in Release:

cmake --preset mingw-release && cmake --build build/mingw-release
ctest --test-dir build/mingw-release -C Release --output-on-failure   # tests + bench gate
./build/mingw-release/adventure_bench.exe                             # raw numbers

Add a bench case for any new hot-path code: ADV_BENCH(name, iters, budgetMs) { ...one iteration... } in a bench/*.cpp, listed in the adventure_bench target. Budgets are ceilings with headroom, not targets — they catch regressions.

To profile the real render loop headlessly: ADVENTURE_PROFILE=300 ./adventure.exe writes profile.csv (frame avg/p50/p95/max, RSS, Lua bytes, per-section ms).

Sanitizers (ASan + UBSan)

CI runs the headless tests under AddressSanitizer + UndefinedBehaviorSanitizer (Linux/gcc) — the net that catches use-after-free / UB our unit tests can't. To reproduce locally on Linux/WSL:

cmake -S . -B build/asan -G Ninja -DCMAKE_BUILD_TYPE=Debug \
  -DCMAKE_C_FLAGS="-fsanitize=address,undefined -fno-sanitize-recover=all" \
  -DCMAKE_CXX_FLAGS="-fsanitize=address,undefined -fno-sanitize-recover=all" \
  -DCMAKE_EXE_LINKER_FLAGS="-fsanitize=address,undefined"
cmake --build build/asan --target adventure_tests
ASAN_OPTIONS=detect_leaks=0 ctest --test-dir build/asan -R adventure_tests --output-on-failure

The Lua sandbox is a security boundary — tests/test_sandbox.cpp adversarially tries to escape/exhaust it; keep it green when touching src/lua.

Notes

• MinGW binaries statically link the GCC/libstdc++/winpthread runtimes; if a run ever fails with 0xc0000139 (entrypoint not found), a wrong runtime DLL is on PATH — the static link should prevent it.
• Negative-path tests intentionally print Lua errors (watchdog kill, syntax error, error('boom')); a green summary (Status: SUCCESS!) is what matters.
• Add tests next to the code: put .cpp in tests/, list it in adventure_tests in CMakeLists.txt. Keep logic testable by injecting time/IO (see Metrics::setClock).