moose
moose copied to clipboard
Libmesh submodule update
trafficstars
This shouldn't be merged yet, since we don't want to make @milljm 's job in #26848 even harder; plus there's another libMesh PR that I'd be happy to see snuck in under the wire if we really take too long ... but I'm anxious like to see what downstream CI makes of the changes that are here so far; it's entirely possible that we have users who are reading QuadShell4 meshes (because that's a Cubit default) and turning them second-order, and any golded results based on that will break when "full second-order" becomes actually-full second-order - see #27987 for the fix to such a failure in the Navier-Stokes module.
Summary of changes:
- Support for
QuadShell9elements, and fixes for refinement ofQuadShell8elements. This is a behavior-changing update:QuadShell4elements, when elevated to second order, now becomeQuadShell9(with a full biquadratic Lagrange space) rather thanQuadShell8(with a "Serendipity element" space). - Fixes for libMesh
make distandmake distcheckwith new submodules and new contributed packages. - Support and unit tests for non-square EigenSparseMatrix matrices
- Support and unit tests for reading sparse matrices from gzipped-Matlab format files.
- Utility for converting sparse matrix files from PETSc binary formats to Matlab format.
- Improved speed for reading sparse matrices.
- Refactoring of some solution transfer code; exposure of
MeshfreeInterpolationFunctionclass to users. - Support and unit tests in TIMPI for
set_unionofmultisetandmultimapcontainers, includingunordered_versions. - Refactored Nedelec finite element code for simplification and generality.
- Fixes and testing for more IsoGeometricAnalysis use cases on distributed meshes.
- Additional fparser headers installed, to enable user subclassing
- Assorted bug fixes: configure now supports preexisting
-Werrorin compiler flags (with tested compilers), more standard treatment of PETSc error codes. - PetscVector subvectors can now be created by supplying only local subvector indices.
- Incoming vectors to
local_variable_indices()are now cleared before filling. - PerfLog profiling of matrix reads and mesh constraint initialization.
- Refactoring and optional debugging features for mesh constraint code.
- Added
RBConstruction::is_serial_training_type()and EIM updates to reduced basis code. - Added a DofMap API for users to request full matrix sparsity pattern retention even with linear algebra packages that do not need it themselves.