Allow all Petsc matrix types to proceed from a common base class

[lindsayad]

And critically, always allow retrieving a PetscMatrix context from a PETSc mat. Eventually we may be able to remove the PetscMatrix from Mat constructors entirely. That would be good

This change was strongly suggested by the need to convert LinearSolver to be able to handle StaticCondensation

[moosebuild]

Job Coverage on 84661f7 wanted to post the following:

Coverage

	24bb5a	#3914 84661f
	Total	Total	+/-	New
Rate	63.29%	63.28%	-0.01%	66.76%
Hits	71626	71667	+41	239
Misses	41544	41590	+46	119

Diff coverage report

Full coverage report

Warnings

• New new line coverage rate 66.76% is less than the suggested 90.0%

This comment will be updated on new commits.

[moosebuild]

Job Test MOOSE clang on 84661f7 : invalidated by @lindsayad

[moosebuild]

Job Test MOOSE GCC min on 84661f7 : invalidated by @lindsayad

[moosebuild]

Job Test MOOSE debug on 84661f7 : invalidated by @lindsayad

[moosebuild]

Job Test MOOSE min clang on 84661f7 : invalidated by @lindsayad

[jwpeterson]

This PR caused a build failure on the libmesh devel -> master merge for the --disable-exceptions build.

In file included from /opt/petsc/include/petscsys.h:1100,
                 from /opt/petsc/include/petscbag.h:3,
                 from /opt/petsc/include/petsc.h:6,
                 from ./include/libmesh/petsc_macro.h:56,
                 from ./include/libmesh/petsc_matrix_base.h:29,
                 from ../src/numerics/petsc_matrix_base.C:25:
../src/numerics/petsc_matrix_base.C: In static member function 'static libMesh::PetscMatrixBase<T>* libMesh::PetscMatrixBase<T>::get_context(Mat)':
./include/libmesh/petsc_solver_exception.h:96:42: error: 'this' is unavailable for static member functions
 #define LIBMESH_CHKERR(ierr) CHKERRABORT(this->comm().get(), ierr);
                                          ^~~~
/opt/petsc/include/petscerror.h:713:25: note: in definition of macro 'PetscCallAbort'
         (void)MPI_Abort(comm, (PetscMPIInt)ierr_petsc_call_abort_); \
                         ^~~~
./include/libmesh/petsc_solver_exception.h:96:30: note: in expansion of macro 'CHKERRABORT'
 #define LIBMESH_CHKERR(ierr) CHKERRABORT(this->comm().get(), ierr);
                              ^~~~~~~~~~~
./include/libmesh/petsc_solver_exception.h:129:5: note: in expansion of macro 'LIBMESH_CHKERR'
     LIBMESH_CHKERR(libmesh_petsc_call_ierr);                                                       \
     ^~~~~~~~~~~~~~
../src/numerics/petsc_matrix_base.C:119:3: note: in expansion of macro 'LibmeshPetscCall'
   LibmeshPetscCall(PetscObjectQuery((PetscObject)mat, "PetscMatrixCtx", (PetscObject *)&container));
   ^~~~~~~~~~~~~~~~
./include/libmesh/petsc_solver_exception.h:96:42: error: 'this' is unavailable for static member functions
 #define LIBMESH_CHKERR(ierr) CHKERRABORT(this->comm().get(), ierr);
                                          ^~~~
/opt/petsc/include/petscerror.h:713:25: note: in definition of macro 'PetscCallAbort'
         (void)MPI_Abort(comm, (PetscMPIInt)ierr_petsc_call_abort_); \
                         ^~~~
./include/libmesh/petsc_solver_exception.h:96:30: note: in expansion of macro 'CHKERRABORT'
 #define LIBMESH_CHKERR(ierr) CHKERRABORT(this->comm().get(), ierr);
                              ^~~~~~~~~~~
./include/libmesh/petsc_solver_exception.h:129:5: note: in expansion of macro 'LIBMESH_CHKERR'
     LIBMESH_CHKERR(libmesh_petsc_call_ierr);                                                       \
     ^~~~~~~~~~~~~~
../src/numerics/petsc_matrix_base.C:123:3: note: in expansion of macro 'LibmeshPetscCall'
   LibmeshPetscCall(PetscContainerGetPointer(container, &ctx));
   ^~~~~~~~~~~~~~~~

[lindsayad]

beautiful failed code

[jwpeterson]

Yeah, the LIBMESH_CHKERR macro uses this when exceptions are disabled, so it can't be used in static functions, which I guess means that LibmeshPetscCall can't be called in static functions...

Maybe we could add a version of these exceptionless macros that calls CHKERRABORT with the communicator from the Mat that is passed in to get_context()?

[lindsayad]

it could be good to have such a macro in the future. For now, I think we have a good change in #3925