| name | build-and-test |
| description | Build Rodin and run the right test executables. Use when compiling the library, running unit/manufactured/benchmark tests, or verifying a change. |
Build and test Rodin
Build
git submodule update --init --recursive
cmake -S . -B build -DCMAKE_BUILD_TYPE=RelWithDebInfo
cmake --build build -j
A configured build/ tree usually already exists — prefer building just the
target you need:
cmake --build build -j --target RodinVariationalH1Test
Target names equal executable names. An ASAN tree may exist at build-asan/
(use it for memory bugs, e.g. the flaky PETSc metricDot crash — see
doc/agents/petsc.md).
Test
ctest is NOT wired up (ctest -N lists 0 tests). Run GoogleTest executables
directly; the path mirrors the source tree:
build/tests/unit/Rodin/<Module>/Rodin<Module><Component>Test
build/tests/unit/Rodin/Variational/RodinVariationalH1Test
build/tests/unit/Rodin/Solver/RodinNewtonSolverTest
build/tests/unit/Rodin/Math/RodinMathMatrixTest
Use --gtest_filter=<Suite>.<Name> to narrow, --gtest_list_tests to
enumerate.
Suites:
tests/unit — fast, per-module; always run the executable(s) covering the
code you touched.tests/manufactured — convergence and targeted regressions. Mandatory for
assembly/solver changes (PETSc assembly regressions live in
tests/manufactured/Rodin/PETSc/TargetedAssemblyTest.cpp).tests/benchmarks — Google Benchmark, target RodinBenchmarks; use
--benchmark_filter=<regex>. Used as the measurement oracle for
performance claims.
Pass criteria
- The affected unit test executable exits 0.
- Derivative/term changes: the FD-consistency tests still pass (and at P2,
not only P1, for Adaptation terms).
- Performance changes: identical numerics (iteration counts, final
energies/fits) vs a baseline run, plus the benchmark delta.
- PETSc assembly changes: remember CI runs PETSc 3.19 — local green is not
the gate (doc/agents/petsc.md).
