hpc-maths / samurai Goto Github PK

The periodic boundary conditions are defined in the mesh. But is it what we want?

Shouldn't periodic boundary conditions be a boundary condition like any other?

The fmt version was fixed to 0.7 in the conda environment because the CI failed with the 0.8 on Mac Os.

When we try to build a field from a field_expression, we don't check whether all the meshes in the fields making up this expression are compatible.
And, to build the mesh of the output field, we take the first find in the expression.

see https://github.com/hpc-maths/samurai/blob/master/include/samurai/field.hpp#L546

What happened?

It's not possible to name a template parameter as I want when I use INIT_BC macro.

Input code

template <class Field_t>
struct Dirichlet : public Bc<Field_t>
{
    INIT_BC(Dirichlet)

    void apply(Field_t& f, const cell_t& cell_out, const cell_t& cell_in,
               const value_t& value) const override
    {
        f[cell_out] = 2 * value - f[cell_in];
    }
};

What expected?

Give template parameter Field (or Field_t in this example) to INIT_BC macro.

#define INIT_BC(NAME, FIELD)                              \
    using base_t  = samurai::Bc<FIELD>;            \
    using cell_t  = typename base_t::cell_t;       \
    using value_t = typename base_t::value_t;      \
    using base_t::Bc;                              \
                                                   \
    std::unique_ptr<base_t> clone() const override \
    {                                              \
        return std::make_unique<NAME>(*this);      \
    }

INIT_BC(Dirichlet, Field_t)

What is your operating system?

Mac OS

How did you install our software?

from source

Software version

0.x.x

Relevant log output

No response

Code of Conduct

• I agree to follow this project's Code of Conduct

By default, the function update_ghost updates all the ghosts defined in the mesh (AMR or MRA). When you compute the new solution using a time scheme, for example, only the ghosts used by the spatial scheme need to be updated. The ghosts used to adapt the mesh are not needed.

Then, the cost can be considerably reduced if you only update the right ghosts.

When we add a boundary condition using for example

auto u = samurai::make_field<double, 4>("u", mesh);
samurai::make_bc<samurai::dirichlet>(u, 0., 0., 0., 0.);

we defined the boundary condition for all components of the field.

But sometimes, we don't want to split our field and we want to have several boundary conditions depending of the field component.

The BCclass must then be updated to take this case into account.

What happened?

IT's not possible to use update_bc function with UniformMeshsince the periodicity is not implemented for this kind of mesh.

Input code

static constexpr std::size_t dim = 1;
using config                     = UniformConfig<dim>;
auto mesh                        = UniformMesh<config>({{0}, {1}}, 4);
auto u                           = make_field<double, 1>("u", mesh);

make_bc<Dirichlet>(u);
update_bc(u);

What expected?

Have the same behaviour of MRMesh

What is your operating system?

Mac OS

How did you install our software?

from source

Software version

No response

Relevant log output

No response

Code of Conduct

• I agree to follow this project's Code of Conduct

cppcheck gives some errors that are disabled in samurai. To find them, you can search the keyword cppcheck-suppress.

You have to check if these errors are false positives or not.