MeshlessMedusa/include/medusa/bits/domains/GeneralFill.hpp

#ifndef MEDUSA_BITS_DOMAINS_GENERALFILL_HPP_
#define MEDUSA_BITS_DOMAINS_GENERALFILL_HPP_

/**
 * @file
 * Implementation of the general node placing algorithm.
 */

#include "GeneralFill_fwd.hpp"
#include "discretization_helpers.hpp"

#include <medusa/bits/utils/assert.hpp>
#include <medusa/bits/types/Range.hpp>
#include <random>
#include <medusa/bits/utils/randutils.hpp>
#include <medusa/bits/spatial_search/KDTreeMutable.hpp>
#include <medusa/bits/spatial_search/KDTree.hpp>

namespace mm
{

    template <typename vec_t>
    GeneralFill<vec_t>::GeneralFill() : seed_(get_seed()) {}

    /// @cond
    template <typename vec_t>
    template <typename func_t>
    void GeneralFill<vec_t>::operator()(DomainDiscretization<vec_t> &domain, const func_t &h,
                                        int type) const
    {
        KDTreeMutable<vec_t> tree;
        operator()(domain, h, tree, type);
    }

    template <typename vec_t>
    template <typename func_t, typename search_structure_t>
    void GeneralFill<vec_t>::operator()(DomainDiscretization<vec_t> &domain, const func_t &h,
                                        search_structure_t &search, int type) const
    {
        KDTree<vec_t> contains_search;
        domain.makeDiscreteContainsStructure(contains_search);
        auto contains_function = [&](const vec_t point)
        {
            return domain.contains(point, contains_search);
        };
        operator()(domain, h, search, contains_function, type);
    }

    template <typename vec_t>
    template <typename func_t, typename search_structure_t, typename contains_function_t>
    void GeneralFill<vec_t>::operator()(DomainDiscretization<vec_t> &domain, const func_t &h,
                                        search_structure_t &search,
                                        contains_function_t &contains_function, int type) const
    {
        if (type == 0)
            type = 1;
        std::mt19937 gen(seed_);

        std::cout << "Filling the domain interior...\n";
        int cur_node = 0;
        int end_node = domain.size();
        if (end_node == 0)
        { // If domain is empty, pick a random node inside it.
            vec_t lo_bound, hi_bound, random_node;
            std::tie(lo_bound, hi_bound) = domain.shape().bbox();
            std::vector<std::uniform_real_distribution<scalar_t>> distributions;
            for (int j = 0; j < dim; ++j)
                distributions.emplace_back(lo_bound[j], hi_bound[j]);
            int count = 0;
            do
            {
                for (int j = 0; j < dim; ++j)
                {
                    random_node[j] = distributions[j](gen);
                }
                if (++count >= 10000)
                {
                    std::string message = "No suitable node in domain could be found after 10000 tries."
                                          " This might happen if domain volume is very small compared "
                                          "to the volume of the bounding box. Try manually supplying "
                                          "the initial point.";
                    throw std::runtime_error(message);
                }
            } while (!contains_function(random_node));
            // std::cout << "新起始点：" << random_node[0] << ',' << random_node[1] << ',' << random_node[2] << '\n';
            domain.addInternalNode(random_node, type);
            end_node = 1;
        }

        // Main algorithm loop.
        search.insert(domain.positions());
        while ((cur_node < end_node) && (end_node < max_points))
        {
            // std::cout << cur_node << " / " << end0_node << std::endl;
            vec_t p = domain.pos(cur_node);
            // std::cout << "新队头：" << p[0] << ',' << p[1] << ',' << p[2] << '\n';
            scalar_t r = h(p);
            assert_msg(r > 0, "Nodal spacing radius should be > 0, got %g.", r);
            // std::cout << end_node << " : " << max_points << '\n';

            auto candidates = discretization_helpers::SphereDiscretization<scalar_t, dim>::construct(r, n_samples, gen);
            // filter candidates regarding the domain and proximity criteria
            for (const auto &f : candidates)
            {
                vec_t node = p + f; // Shift center to point `p`.
                if (!contains_function(node))
                    continue; // If node is not in the domain.
                if (search.existsPointInSphere(node, r * zeta))
                    continue; // If node is not far enough
                // std::cout << "新候选点：" << node[0] << ',' << node[1] << ',' << node[2] << '\n';
                domain.addInternalNode(node, type); // from other nodes.
                search.insert(node);
                end_node++;
            }
            cur_node++;
        }
        if (end_node >= max_points)
        {
            std::cerr << "Maximum number of points reached, fill may be incomplete." << std::endl;
        }
        std::cout << "Filling the domain interior done!\n";
    }

    template <typename vec_t>
    GeneralFill<vec_t> &GeneralFill<vec_t>::proximityTolerance(scalar_t zeta)
    {
        assert_msg((0 < zeta && zeta < 1), "Zeta must be between 0 and 1, got %f.", zeta);
        this->zeta = zeta;
        return *this;
    }
    /// @endcond

} // namespace mm

#endif // MEDUSA_BITS_DOMAINS_GENERALFILL_HPP_