Geom Module Unit Tests

Tests for the mesh geometry infrastructure in src/Geom/. All C++ tests use doctest. MPI-aware tests are registered with CTest at multiple process counts.

Building and Running

# Build all Geom C++ test executables
cmake --build build -t geom_unit_tests -j8

# Run every Geom CTest
ctest --test-dir build -R geom_ --output-on-failure

# Run a single suite
ctest --test-dir build -R geom_elements --output-on-failure

Target Summary

CMake target

CTest name

Source file

Timeout

geom_test_elements

geom_elements

test_Elements.cpp

120 s

geom_test_quadrature

geom_quadrature

test_Quadrature.cpp

120 s

geom_test_mesh_index_conversion

geom_mesh_index_conversion_np{1,2,4}

test_MeshIndexConversion.cpp

120 s

geom_test_mesh_pipeline

geom_mesh_pipeline_np{1,2,4}

test_MeshPipeline.cpp

120 s

geom_test_mesh_distributed_read

geom_mesh_distributed_read_np{1,2,4}

test_MeshDistributedRead.cpp

120 s

geom_test_mesh_connectivity

geom_mesh_connectivity_np{1,2,4}

test_MeshConnectivity.cpp

120 s

geom_test_mesh_connectivity_ghost

geom_mesh_connectivity_ghost_np{1,2,4}

test_MeshConnectivity_Ghost.cpp

120 s

geom_test_mesh_connectivity_interpolate

geom_mesh_connectivity_interpolate_np{1,2,4}

test_MeshConnectivity_Interpolate.cpp

120 s

geom_test_mesh_reorder

geom_mesh_reorder_np{1,2,4}

test_MeshReorder.cpp

120 s

Element Types (test_Elements.cpp)

See also: test_Elements.cpp

Serial-only tests for element geometry definitions, node counts, and shape function evaluation.

Quadrature Rules (test_Quadrature.cpp)

See also: test_Quadrature.cpp

Serial-only tests for Gaussian quadrature points and weights on standard elements (triangles, quads, tetrahedra, hexahedra, prisms, pyramids).

Mesh Index Conversion (test_MeshIndexConversion.cpp)

See also: test_MeshIndexConversion.cpp

MPI-parallel tests verifying local-to-global and global-to-local index conversions on partitioned meshes.

Mesh Pipeline (test_MeshPipeline.cpp)

See also: test_MeshPipeline.cpp

MPI-parallel end-to-end tests for the mesh construction pipeline: reading, partitioning, ghost creation, and boundary extraction.

Distributed Mesh I/O (test_MeshDistributedRead.cpp)

See also: test_MeshDistributedRead.cpp

MPI-parallel tests for reading CGNS mesh files in parallel and redistributing across different partition counts.

MeshConnectivity DSL (test_MeshConnectivity.cpp)

See also: test_MeshConnectivity.cpp

MPI-parallel tests for the MeshConnectivity standalone DSL operations: Inverse, Compose, ComposeFiltered, Interpolate, and adjacency registry management. Validates cone/support inversion, compose with predicates, and entity-kind registration.

Ghost Tree Evaluation (test_MeshConnectivity_Ghost.cpp)

See also: test_MeshConnectivity_Ghost.cpp

MPI-parallel tests for evaluateGhostTree, GhostSpec, CompiledGhostTree, and multi-layer ghost cell support. Tests include single-layer and multi-layer ghost chains, scratch-pull between BFS levels, and 2D tiled synthetic grids with analytical ghost formulas.

Interpolation (test_MeshConnectivity_Interpolate.cpp)

See also: test_MeshConnectivity_Interpolate.cpp

MPI-parallel tests for InterpolateGlobal — distributed sub-entity extraction with global deduplication and periodic-aware matching.

Mesh Reordering (test_MeshReorder.cpp)

See also: test_MeshReorder.cpp

MPI-parallel tests for the distributed entity reordering framework: ReorderPlan, ReorderRegistry, ReorderInput, and UnstructuredMesh::ReorderEntities. See Distributed Reorder Design for the architecture, and PermutationTransfer Tests for the underlying MPI primitive.

Classification and registry

  • classifyAdj basic classification — exhaustively covers all five AdjAction outcomes (SKIP, RELOCATE, REMAP, RELOCATE_REMAP, SELF) across cell-only, node-only, both-reordered, and intra-level adjacency scenarios.

  • ReorderRegistry register and query — registerAdj, registerCompanion, registerGlobalMapping, and getGlobalMapping with both hits and misses. Verifies callback storage and retrieval.

Synthetic plan application

  • ReorderPlan::apply cell-only local permutation — synthetic cell/node arrays with identity cell partition; validates that the registered RELOCATE callback is invoked and data is preserved.

  • ReorderPlan::apply node-only remap — node reorder with cells unchanged; validates REMAP path and companion relocate for the node-parallel coords analog.

  • ReorderPlan::apply RELOCATE_REMAP (both source and target reordered) — explicitly exercises the fourth AdjAction case: cell2node with both Cell and Node in the reorder set. Asserts classifyAdj returns RELOCATE_REMAP and confirms data integrity after both phases.

Real-mesh ReorderEntities

  • Cell-only local on UniformSquare_10.cgns — identity partition, full rebuild pipeline (RecoverNode2Cell → BuildGhost → Global2Local); verifies counts and entry validity.

  • Cell-only with face destruction — destroys faces via destroyKinds={EntityKind::Face}, then rebuilds from scratch via InterpolateFace.

  • Cell distributed round-robin with follow — non-identity partition (i % nRanks); Node and Bnd automatically follow Cell via the default follow policy. Validates global count preservation, entry validity, and post-reorder mesh rebuild.

  • Node-only local — node reorder only, cells stay; verifies coord preservation under identity partition and mesh rebuild success.

Expected-value verification

  • PermutationTransfer + buildLookup: reverse permutation value tracking — uses fromLocalPermutation with a non-identity permutation and verifies that each old global maps to the exact expected new global via lookup.resolve(). Also verifies that row data moves to the permuted slot.

  • Distributed value tracking cross-rank — covered in PermutationTransfer Tests.

Framework extension

  • ReorderEntities with external companion array (solver-like) — creates an external ArrayAdjacencyPair<3> tagged with per-cell DOF patterns (column pattern (g*10+0, g*10+1, g*10+2)), registers it as a companion via registerCompanion on the registry returned by buildReorderRegistry, applies the plan, and verifies that all three columns of each row travel together (DOF layout preserved across distribution). Demonstrates the pattern for solver arrays that must participate in mesh reordering.

Registry invariants

  • buildReorderRegistry populates pullSets — validates that pull-set entries are off-rank, within [0, globalSize), sorted, and unique. Confirms all expected adjacency and companion entries are registered by UnstructuredMesh::buildReorderRegistry.