Overview

The DNDSR project uses the doctest (v2.4.11) framework for C++ unit testing. Test sources live under test/cpp/DNDS/ and are built only when DNDS_BUILD_TESTS=ON.

Every MPI-aware test is registered with CTest at four process counts (np = 1, 2, 4, 8) so that correctness under parallel execution is verified automatically.

Building and Running

# Configure with tests
cmake -B build -DDNDS_BUILD_TESTS=ON
 
# Build all test executables at once
cmake --build build -t dnds_unit_tests -j8
 
# Run every registered test (serial + MPI np=1,2,4)
ctest --test-dir build -R dnds_ --output-on-failure
 
# Run a single test suite
ctest --test-dir build -R dnds_mpi_np2 --output-on-failure
 
# Run a single executable directly
mpirun -np 4 ./build/test/cpp/dnds_test_array_transformer

Naming Conventions

CMake target	CTest names	Source file
`dnds_test_array`	`dnds_array`	test_Array.cpp
`dnds_test_mpi`	`dnds_mpi_np{1,2,4}`	test_MPI.cpp
`dnds_test_array_transformer`	`dnds_array_transformer_np{1,2,4}`	test_ArrayTransformer.cpp
`dnds_test_array_derived`	`dnds_array_derived_np{1,2,4}`	test_ArrayDerived.cpp
`dnds_test_array_dof`	`dnds_array_dof_np{1,2,4}`	test_ArrayDOF.cpp
`dnds_test_index_mapping`	`dnds_index_mapping_np{1,2,4}`	test_IndexMapping.cpp
`dnds_test_serializer`	`dnds_serializer_np{1,2,4}`	test_Serializer.cpp

Note on POSIX <tt>index()</tt> Ambiguity

Because doctest includes <cstring> which transitively pulls in the POSIX index() function from <strings.h>, the bare name index is ambiguous when using namespace DNDS; is active. All test files therefore qualify the DNDS type aliases as DNDS::index, DNDS::real, and DNDS::rowsize in declarations.

Array Tests (test_Array.cpp)

See also: test_Array.cpp

Serial-only tests covering every DNDS::Array data layout:

TABLE_StaticFixed — Array<real, 3>: compile-time fixed row size. Validates Size(), RowSize(), element read/write via operator() and operator[], DataSize(), DataSizeBytes().
TABLE_Fixed — Array<real, DynamicSize>: runtime-uniform row size.
TABLE_StaticMax — Array<real, NonUniformSize, 4>: per-row variable sizes up to a compile-time maximum.
TABLE_Max — Array<real, NonUniformSize, DynamicSize>: per-row variable sizes up to a runtime maximum.
CSR — Array<real, NonUniformSize>: compressed sparse row. Tests lambda-based Resize, Compress/Decompress round-trips, ResizeRow in decompressed mode.
Clone and copy — clone(), CopyData(), copy constructor, SwapData() independence checks.
Edge cases — zero-size arrays, single-element arrays, CSR rows of length zero.
Serialization — JSON round-trip for TABLE_Fixed, CSR, and TABLE_StaticFixed via DNDS::Serializer::SerializerJSON.
Signature — GetArraySignature(), ParseArraySignatureTuple(), ArraySignatureIsCompatible().
Hash — identical arrays produce equal hashes; modified arrays differ.

MPI Wrapper Tests (test_MPI.cpp)

See also: test_MPI.cpp

Run at np = 1, 2, 4, 8. All expected values are formulated in terms of mpi.size so they hold for any rank count.

MPIInfo basics — setWorld(), field validity, equality.
Allreduce — MPI_SUM and MPI_MAX for real and index; AllreduceOneReal, AllreduceOneIndex.
Scan — inclusive prefix sum.
Allgather — single and multi-element.
Bcast — from rank 0 and from last rank.
Barrier — returns MPI_SUCCESS.
BasicType_To_MPIIntType — scalar types, std::array, Eigen::Matrix.
CommStrategy — get/set HIndexed/InSituPack.
Alltoall — single and multi-element exchange.

ArrayTransformer Tests (test_ArrayTransformer.cpp)

See also: test_ArrayTransformer.cpp

Exercises ghost (halo) communication at np = 1, 2, 4, 8.

ParArray basics — createGlobalMapping(), globalSize(), AssertConsistent().
Pull – TABLE_StaticFixed — 100 elements/rank, pulls first 5 from each remote rank, verifies ghost data.
Pull – TABLE_Fixed — complete replication pull.
Pull – CSR — variable row sizes, verifies both data and row sizes in ghosts.
Pull – std::array elements — std::array<real,9> as the element type.
Persistent pull — initPersistentPull / startPersistentPull / waitPersistentPull, then modify father and re-pull.
BorrowGGIndexing — second array shares the ghost mapping of the first, independently pulls correct data.
Push — write to son, pushOnce(), verify father received values.

ArrayDerived Tests (test_ArrayDerived.cpp)

See also: test_ArrayDerived.cpp

Covers every derived array type at np = 1, 2, 4, 8.

ArrayAdjacency — basics (resize, row resize, compress, operator[], rowPtr()), ghost communication, clone, fixed-size variant.
ArrayEigenVector — basics (static and dynamic sizes), ghost communication.
ArrayEigenMatrix — static sizes, dynamic sizes, NonUniform rows, ghost communication.
ArrayEigenMatrixBatch — InitializeWriteRow, BatchSize, operator(), ghost communication.
ArrayEigenUniMatrixBatch — static and dynamic sizes, ResizeBatch.

ArrayDOF Tests (test_ArrayDOF.cpp)

See also: test_ArrayDOF.cpp

Tests every vector-space operation on DNDS::ArrayDof at np = 1, 2, 4, 8. All MPI-global reductions (norm2, dot, min, max, sum, componentWiseNorm1) use rank-aware expected values.

setConstant (scalar and matrix)
operator+= (scalar, array, matrix)
operator-=, operator*= (scalar, element-wise, matrix)
operator/=
addTo
norm2, norm2(other) (L2 distance)
dot
min, max, sum
componentWiseNorm1, componentWiseNorm1(other)
operator= (value copy), clone (deep copy)
scalar-array multiply (ArrayDof<N,1> *= ArrayDof<1,1>)
identity: dot(x, x) == norm2(x)^2

IndexMapping Tests (test_IndexMapping.cpp)

See also: test_IndexMapping.cpp

Tests global/local index mapping at np = 1, 2, 4, 8.

GlobalOffsetsMapping — uniform distribution, non-uniform distribution, search() for first/last/middle/out-of-range indices.
OffsetAscendIndexMapping — pull-based construction, searchInMain, searchInGhost, searchInAllGhost, search, search_indexAppend, operator() reverse mapping, empty ghost set.

Serializer Tests (test_Serializer.cpp)

See also: test_Serializer.cpp

Round-trip write/read verification at np = 1, 2, 4, 8.

SerializerJSON scalar — WriteInt/ReadInt, WriteIndex/ReadIndex, WriteReal/ReadReal, WriteString/ReadString.
SerializerJSON vector — WriteRealVector, WriteIndexVector, WriteRowsizeVector.
SerializerJSON uint8 — with and without base64 codec.
SerializerJSON paths — CreatePath, GoToPath, GetCurrentPath, ListCurrentPath.
SerializerJSON pointer dedup — shared pointer written twice resolves to same object on read.
SerializerH5 scalar — WriteInt/ReadInt, WriteIndex/ReadIndex, WriteReal/ReadReal, WriteString/ReadString (HDF5 attributes).
SerializerH5 vector — WriteRealVector, WriteIndexVector with explicit ArrayGlobalOffset; read with ArrayGlobalOffset_Unknown auto-detection from the rank_offsets companion dataset.
SerializerH5 distributed vector — non-uniform per-rank sizes, read with both ArrayGlobalOffset_Unknown and explicit offset.
SerializerH5 uint8 — two-pass read (nullptr size query, then actual read with offset from pass 1).
SerializerH5 paths — GoToPath, GetCurrentPath, ListCurrentPath (groups materialized by writing content).
SerializerH5 string — fixed-length HDF5 string attributes.

Table of Contents