In AddBoundaryElements, use GetSubMeshEdgeFromParent(parentFaceIdx), not
GetSubMeshFaceFromParent(parentFaceIdx) for 2D meshes. This fixes an assert
failure/segfault when constructing a submesh with internal boundaries in 2D.
Fixes#4729
- Rename RemoveBoundaryElementToEdge to DeleteBoundaryElementToEdge
- Remove unneeded member variables `from` and `attributes` from NCSubMesh and ParNCSubMesh
- Rewrapping to 80
- Forward some base methods to public access
- Restore some old implementation methods in terms of new versions
- Doyxgen and comment fixes
- Unused variables
- Memory leak
- Initialization order
- Remove std::vector usage
- Fix char index into arrays
- Early exit if an empty ncsubmesh is constructed
- Array list initalizer was hardcoded to int
- Remove anonynmous namespace from ncmesh_tables.hpp, use static and
constexpr instead
- Remove unneeded table includes in ncsubmesh.cpp and pncsubmesh.cpp
- Move trivial type assertion of Array to class body from constructor
- Fix warning about use of abs over std::abs in batched linalg
- Add parent hashtable accessors to bypass access controls in parent
classes.
- Change loop condition to avoid need for ncmesh_tables.hpp include
- Change [ParSubMesh] to [SubMesh] for Catch2 category, the parallel
is implicit in [Parallel] label
- Missing include in mesh_test_utils.cpp
- Fix bug for array access with empty ncmesh
- Fix to permute children if a grandchild discovers a different face node ordering. Running out of tests
- Fix bug for triangle faces where central face parent nodes were not being identified correctly.
- Fix bug in ParentFaceNodes where a central triangle face would discover parent nodes with a different orientation to the surrounding faces.
- Fix bug where check was on parent_nodes rather than face_nodes for a second node hit
Many small features and a lot of really quite dirty code, print statements etc. This comes from squashing a large number of commits together.
A few different strategies were tried and failed (partially building the NC structure, having ranks have different NC etc.) before arriving at
the final reverse tree traversal algorithm.
Veselin Dobrev