Avoid const return type in functions

[masterleinad]

const doesn't do anything when we are returning values.

[bangerth]

Could type be a reference/pointer? And if so, would that make the pointer const, or the pointed to object const?

[masterleinad]

The const doesn't matter in that case either, see https://godbolt.org/z/8jesMca3a. Returning a constant pointer is the same as returning a non-constant pointer and a pointer to a constant type is still returned as such.

[bangerth]

Several of the testers fail, at least one of them with this error message:

/home/runner/work/dealii/dealii/source/base/mpi.cc:681:5: error: ambiguating new declaration of ‘const std::vector<unsigned int> dealii::Utilities::MPI::mpi_processes_within_communicator(MPI_Comm, MPI_Comm)’
  681 |     mpi_processes_within_communicator(const MPI_Comm, const MPI_Comm)
      |     ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
In file included from /home/runner/work/dealii/dealii/source/base/mpi.cc:19:
/home/runner/work/dealii/dealii/include/deal.II/base/mpi.h:158:5: note: old declaration ‘std::vector<unsigned int> dealii::Utilities::MPI::mpi_processes_within_communicator(MPI_Comm, MPI_Comm)’
  158 |     mpi_processes_within_communicator(const MPI_Comm comm_large,
      |     ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

[bangerth]

Still fails:

/jenkins/workspace/dealii-serial_PR-15671/tests/multigrid/mg_data_out_01.cc: In instantiation of ‘void print(dealii::DataOut<dim>&, const dealii::Triangulation<dim, dim>&) [with int dim = 2]’:
/jenkins/workspace/dealii-serial_PR-15671/tests/multigrid/mg_data_out_01.cc:62:8:   required from ‘void do_test() [with int dim = 2]’
/jenkins/workspace/dealii-serial_PR-15671/tests/multigrid/mg_data_out_01.cc:95:14:   required from here
/jenkins/workspace/dealii-serial_PR-15671/tests/multigrid/mg_data_out_01.cc:34:9: error: cannot bind non-const lvalue reference of type ‘std::pair<std::function<dealii::TriaIterator<dealii::CellAccessor<2, 2> >(const dealii::Triangulation<2, 2>&)>, std::function<dealii::TriaIterator<dealii::CellAccessor<2, 2> >(const dealii::Triangulation<2, 2>&, const dealii::TriaIterator<dealii::CellAccessor<2, 2> >&)> >&’ to an rvalue of type ‘std::pair<std::function<dealii::TriaIterator<dealii::CellAccessor<2, 2> >(const dealii::Triangulation<2, 2>&)>, std::function<dealii::TriaIterator<dealii::CellAccessor<2, 2> >(const dealii::Triangulation<2, 2>&, const dealii::TriaIterator<dealii::CellAccessor<2, 2> >&)> >’
   34 |   auto &p = data_out.get_cell_selection();
      |         ^
/jenkins/workspace/dealii-serial_PR-15671/tests/multigrid/mg_data_out_01.cc: In instantiation of ‘void print(dealii::DataOut<dim>&, const dealii::Triangulation<dim, dim>&) [with int dim = 3]’:
/jenkins/workspace/dealii-serial_PR-15671/tests/multigrid/mg_data_out_01.cc:62:8:   required from ‘void do_test() [with int dim = 3]’
/jenkins/workspace/dealii-serial_PR-15671/tests/multigrid/mg_data_out_01.cc:96:14:   required from here
/jenkins/workspace/dealii-serial_PR-15671/tests/multigrid/mg_data_out_01.cc:34:9: error: cannot bind non-const lvalue reference of type ‘std::pair<std::function<dealii::TriaIterator<dealii::CellAccessor<3, 3> >(const dealii::Triangulation<3, 3>&)>, std::function<dealii::TriaIterator<dealii::CellAccessor<3, 3> >(const dealii::Triangulation<3, 3>&, const dealii::TriaIterator<dealii::CellAccessor<3, 3> >&)> >&’ to an rvalue of type ‘std::pair<std::function<dealii::TriaIterator<dealii::CellAccessor<3, 3> >(const dealii::Triangulation<3, 3>&)>, std::function<dealii::TriaIterator<dealii::CellAccessor<3, 3> >(const dealii::Triangulation<3, 3>&, const dealii::TriaIterator<dealii::CellAccessor<3, 3> >&)> >’
make[3]: *** [CMakeFiles/multigrid.mg_data_out_01.debug.dir/build.make:76: CMakeFiles/multigrid.mg_data_out_01.debug.dir/mg_data_out_01.cc.o] Error 1
make[2]: *** [CMakeFiles/Makefile2:1902: CMakeFiles/multigrid.mg_data_out_01.debug.dir/all] Error 2
make[1]: *** [CMakeFiles/Makefile2:2143: CMakeFiles/multigrid.mg_data_out_01.debug.test.dir/rule] Error 2

[bangerth]

That is a pretty fascinating failure, by the way. It shows that the const on an rvalue return object really does matter in some contexts!

[bangerth]

In short, this does not compile:

const int f();

int &i = f();

The error in the test probably points to a case where we shouldn't store a reference to begin with.

[masterleinad]

That is a pretty fascinating failure, by the way. It shows that the const on an rvalue return object really does matter in some contexts!

Yes, also see https://godbolt.org/z/r5nhc6Yde. auto& becomes const Foo& when the return type is const.

[bangerth]

I wonder whether we want to reconsider this. It breaks ASPECT in ~4 places: geodynamics/aspect#5208