Testing and Verification

Choose checks according to the boundary changed. Record commands actually run, their result, and any intentional omission.

Minimum workflow

1. Configure a clean or compatible build directory.
2. Build the affected target.
3. Run focused unit or integration tests.
4. Run applicable architecture/red-line checks.
5. For runtime behavior, capture a bounded smoke run and verify a meaningful success condition rather than accepting a timeout alone.

Change-specific checks

Change	Required verification
Public API or package ABI	Compile consumers; inspect exported package entry when relevant; test compatibility paths.
Generator/schema/template	Run generator tests, regenerate maintained examples, and require byte-equal output unless fixtures intentionally change.
Transport or zero-copy path	Build supported feature combinations; test unsupported combinations fail clearly; verify ownership/copy semantics.
Extension	Build and load the extension with representative configuration; run focused RPC or messaging smoke.
Cross compilation	Validate the target prefix, configure with host paths isolated, build target artifacts, and inspect architecture/dependencies.
Performance-sensitive path	Use the public benchmark methodology with pinned environment metadata and repeated runs; compare like-for-like modes.

Test quality rules

• Add a regression test with a non-trivial bug fix.
• Do not use skip/only markers, delete assertions, or broaden accepted exit codes to conceal a failure.
• Keep fixtures deterministic and review generated changes as code.
• A functional receive marker does not establish latency or zero-copy behavior.
• Historical test counts and measurements are context, not acceptance criteria for a new checkout.

Common commands

cmake --preset minimal
cmake --build build -j"$(nproc)"
ctest --test-dir build --output-on-failure
python3 -m pytest src/tools/ibmw_gen/tests
bash scripts/check_red_lines.sh

Not every build exposes every command or target. Use the preset and focused test target appropriate to the feature, then report exactly what ran. For benchmark work, follow Transport Benchmark.