QIDISlicer/src/libslic3r/PrintObject.cpp

#include "AABBTreeLines.hpp"
#include "BridgeDetector.hpp"
#include "ExPolygon.hpp"
#include "Exception.hpp"
#include "Flow.hpp"
#include "GCode/ExtrusionProcessor.hpp"
#include "KDTreeIndirect.hpp"
#include "Line.hpp"
#include "Point.hpp"
#include "Polygon.hpp"
#include "Polyline.hpp"
#include "Print.hpp"
#include "BoundingBox.hpp"
#include "ClipperUtils.hpp"
#include "ElephantFootCompensation.hpp"
#include "Geometry.hpp"
#include "I18N.hpp"
#include "Layer.hpp"
#include "MutablePolygon.hpp"
#include "PrintBase.hpp"
#include "PrintConfig.hpp"
#include "Support/SupportMaterial.hpp"
#include "Support/TreeSupport.hpp"
#include "Surface.hpp"
#include "Slicing.hpp"
#include "SurfaceCollection.hpp"
#include "Tesselate.hpp"
#include "TriangleMeshSlicer.hpp"
#include "Utils.hpp"
#include "Fill/FillAdaptive.hpp"
#include "Fill/FillLightning.hpp"
#include "Format/STL.hpp"
#include "Support/SupportMaterial.hpp"
#include "SupportSpotsGenerator.hpp"
#include "TriangleSelectorWrapper.hpp"
#include "format.hpp"
#include "libslic3r.h"

#include <algorithm>
#include <cmath>
#include <cstddef>
#include <cstdint>
#include <float.h>
#include <functional>
#include <limits>
#include <map>
#include <oneapi/tbb/blocked_range.h>
#include <oneapi/tbb/concurrent_vector.h>
#include <oneapi/tbb/parallel_for.h>
#include <string>
#include <string_view>
#include <tuple>
#include <unordered_map>
#include <unordered_set>
#include <utility>

#include <boost/log/trivial.hpp>

#include <tbb/parallel_for.h>
#include <vector>

using namespace std::literals;

// #define PRINT_OBJECT_TIMING

#ifdef PRINT_OBJECT_TIMING
    // time limit for one ClipperLib operation (union / diff / offset), in ms
    #define PRINT_OBJECT_TIME_LIMIT_DEFAULT 50
    #include <boost/current_function.hpp>
    #include "Timer.hpp"
    #define PRINT_OBJECT_TIME_LIMIT_SECONDS(limit) Timing::TimeLimitAlarm time_limit_alarm(uint64_t(limit) * 1000000000l, BOOST_CURRENT_FUNCTION)
    #define PRINT_OBJECT_TIME_LIMIT_MILLIS(limit) Timing::TimeLimitAlarm time_limit_alarm(uint64_t(limit) * 1000000l, BOOST_CURRENT_FUNCTION)
#else
    #define PRINT_OBJECT_TIME_LIMIT_SECONDS(limit) do {} while(false)
    #define PRINT_OBJECT_TIME_LIMIT_MILLIS(limit) do {} while(false)
#endif // PRINT_OBJECT_TIMING

#ifdef SLIC3R_DEBUG_SLICE_PROCESSING
#define SLIC3R_DEBUG
#endif

// #define SLIC3R_DEBUG

// Make assert active if SLIC3R_DEBUG
#ifdef SLIC3R_DEBUG
    #undef NDEBUG
    #define DEBUG
    #define _DEBUG
    #include "SVG.hpp"
    #undef assert 
    #include <cassert>
#endif

    #include "SVG.hpp"

namespace Slic3r {

// Constructor is called from the main thread, therefore all Model / ModelObject / ModelIntance data are valid.
PrintObject::PrintObject(Print* print, ModelObject* model_object, const Transform3d& trafo, PrintInstances&& instances) :
    PrintObjectBaseWithState(print, model_object),
    m_trafo(trafo)
{
    // Compute centering offet to be applied to our meshes so that we work with smaller coordinates
    // requiring less bits to represent Clipper coordinates.

	// Snug bounding box of a rotated and scaled object by the 1st instantion, without the instance translation applied.
	// All the instances share the transformation matrix with the exception of translation in XY and rotation by Z,
	// therefore a bounding box from 1st instance of a ModelObject is good enough for calculating the object center,
	// snug height and an approximate bounding box in XY.
    BoundingBoxf3  bbox        = model_object->raw_bounding_box();
    Vec3d 		   bbox_center = bbox.center();
	// We may need to rotate the bbox / bbox_center from the original instance to the current instance.
    double z_diff = Geometry::rotation_diff_z(model_object->instances.front()->get_matrix(), instances.front().model_instance->get_matrix());
    if (std::abs(z_diff) > EPSILON) {
		auto z_rot  = Eigen::AngleAxisd(z_diff, Vec3d::UnitZ());
		bbox 		= bbox.transformed(Transform3d(z_rot));
		bbox_center = (z_rot * bbox_center).eval();
	}

    // Center of the transformed mesh (without translation).
    m_center_offset = Point::new_scale(bbox_center.x(), bbox_center.y());
    // Size of the transformed mesh. This bounding may not be snug in XY plane, but it is snug in Z.
    m_size = (bbox.size() * (1. / SCALING_FACTOR)).cast<coord_t>();
    m_size.z() = coord_t(model_object->max_z() * (1. / SCALING_FACTOR));

    this->set_instances(std::move(instances));
}

PrintBase::ApplyStatus PrintObject::set_instances(PrintInstances &&instances)
{
    for (PrintInstance &i : instances)
    	// Add the center offset, which will be subtracted from the mesh when slicing.
    	i.shift += m_center_offset;
    // Invalidate and set copies.
    PrintBase::ApplyStatus status = PrintBase::APPLY_STATUS_UNCHANGED;
    bool equal_length = instances.size() == m_instances.size();
    bool equal = equal_length && std::equal(instances.begin(), instances.end(), m_instances.begin(), 
    	[](const PrintInstance& lhs, const PrintInstance& rhs) { return lhs.model_instance == rhs.model_instance && lhs.shift == rhs.shift; });
    if (! equal) {
        status = PrintBase::APPLY_STATUS_CHANGED;
        if (m_print->invalidate_steps({ psSkirtBrim, psGCodeExport }) ||
            (! equal_length && m_print->invalidate_step(psWipeTower)))
            status = PrintBase::APPLY_STATUS_INVALIDATED;
        m_instances = std::move(instances);
	    for (PrintInstance &i : m_instances)
	    	i.print_object = this;
    }
    return status;
}

std::vector<std::reference_wrapper<const PrintRegion>> PrintObject::all_regions() const
{
    std::vector<std::reference_wrapper<const PrintRegion>> out;
    out.reserve(m_shared_regions->all_regions.size());
    for (const std::unique_ptr<Slic3r::PrintRegion> &region : m_shared_regions->all_regions)
        out.emplace_back(*region.get());
    return out;
}

// 1) Merges typed region slices into stInternal type.
// 2) Increases an "extra perimeters" counter at region slices where needed.
// 3) Generates perimeters, gap fills and fill regions (fill regions of type stInternal).
void PrintObject::make_perimeters()
{
    // prerequisites
    this->slice();

    if (! this->set_started(posPerimeters))
        return;

    m_print->set_status(20, _u8L("Generating perimeters"));
    BOOST_LOG_TRIVIAL(info) << "Generating perimeters..." << log_memory_info();
    
    // Revert the typed slices into untyped slices.
    if (m_typed_slices) {
        for (Layer *layer : m_layers) {
            layer->clear_fills();
            layer->restore_untyped_slices();
            m_print->throw_if_canceled();
        }
        m_typed_slices = false;
    }
    
    // compare each layer to the one below, and mark those slices needing
    // one additional inner perimeter, like the top of domed objects-
    
    // this algorithm makes sure that at least one perimeter is overlapping
    // but we don't generate any extra perimeter if fill density is zero, as they would be floating
    // inside the object - infill_only_where_needed should be the method of choice for printing
    // hollow objects
    for (size_t region_id = 0; region_id < this->num_printing_regions(); ++ region_id) {
        const PrintRegion &region = this->printing_region(region_id);
        if (! region.config().extra_perimeters || region.config().perimeters == 0 || region.config().fill_density == 0 || this->layer_count() < 2)
            continue;

        BOOST_LOG_TRIVIAL(debug) << "Generating extra perimeters for region " << region_id << " in parallel - start";
        tbb::parallel_for(
            tbb::blocked_range<size_t>(0, m_layers.size() - 1),
            [this, &region, region_id](const tbb::blocked_range<size_t>& range) {
                PRINT_OBJECT_TIME_LIMIT_MILLIS(PRINT_OBJECT_TIME_LIMIT_DEFAULT);
                for (size_t layer_idx = range.begin(); layer_idx < range.end(); ++ layer_idx) {
                    m_print->throw_if_canceled();
                    LayerRegion &layerm                     = *m_layers[layer_idx]->get_region(region_id);
                    const LayerRegion &upper_layerm         = *m_layers[layer_idx+1]->get_region(region_id);
                    const Polygons upper_layerm_polygons    = to_polygons(upper_layerm.slices().surfaces);
                    // Filter upper layer polygons in intersection_ppl by their bounding boxes?
                    // my $upper_layerm_poly_bboxes= [ map $_->bounding_box, @{$upper_layerm_polygons} ];
                    const double total_loop_length      = total_length(upper_layerm_polygons);
                    const coord_t perimeter_spacing     = layerm.flow(frPerimeter).scaled_spacing();
                    const Flow ext_perimeter_flow       = layerm.flow(frExternalPerimeter);
                    const coord_t ext_perimeter_width   = ext_perimeter_flow.scaled_width();
                    const coord_t ext_perimeter_spacing = ext_perimeter_flow.scaled_spacing();

                    // slice is not const because slice.extra_perimeters is being incremented.
                    for (Surface &slice : layerm.m_slices.surfaces) {
                        for (;;) {
                            // compute the total thickness of perimeters
                            const coord_t perimeters_thickness = ext_perimeter_width/2 + ext_perimeter_spacing/2
                                + (region.config().perimeters-1 + slice.extra_perimeters) * perimeter_spacing;
                            // define a critical area where we don't want the upper slice to fall into
                            // (it should either lay over our perimeters or outside this area)
                            const coord_t critical_area_depth = coord_t(perimeter_spacing * 1.5);
                            const Polygons critical_area = diff(
                                offset(slice.expolygon, float(- perimeters_thickness)),
                                offset(slice.expolygon, float(- perimeters_thickness - critical_area_depth))
                            );
                            // check whether a portion of the upper slices falls inside the critical area
                            const Polylines intersection = intersection_pl(to_polylines(upper_layerm_polygons), critical_area);
                            // only add an additional loop if at least 30% of the slice loop would benefit from it
                            if (total_length(intersection) <=  total_loop_length*0.3)
                                break;
                            /*
                            if (0) {
                                require "Slic3r/SVG.pm";
                                Slic3r::SVG::output(
                                    "extra.svg",
                                    no_arrows   => 1,
                                    expolygons  => union_ex($critical_area),
                                    polylines   => [ map $_->split_at_first_point, map $_->p, @{$upper_layerm->slices} ],
                                );
                            }
                            */
                            ++ slice.extra_perimeters;
                        }
                        #ifdef DEBUG
                            if (slice.extra_perimeters > 0)
                                printf("  adding %d more perimeter(s) at layer %zu\n", slice.extra_perimeters, layer_idx);
                        #endif
                    }
                }
            });
        m_print->throw_if_canceled();
        BOOST_LOG_TRIVIAL(debug) << "Generating extra perimeters for region " << region_id << " in parallel - end";
    }

    BOOST_LOG_TRIVIAL(debug) << "Generating perimeters in parallel - start";
    tbb::parallel_for(
        tbb::blocked_range<size_t>(0, m_layers.size()),
        [this](const tbb::blocked_range<size_t>& range) {
            PRINT_OBJECT_TIME_LIMIT_MILLIS(PRINT_OBJECT_TIME_LIMIT_DEFAULT);
            for (size_t layer_idx = range.begin(); layer_idx < range.end(); ++ layer_idx) {
                m_print->throw_if_canceled();
                m_layers[layer_idx]->make_perimeters();
            }
        }
    );
    m_print->throw_if_canceled();
    BOOST_LOG_TRIVIAL(debug) << "Generating perimeters in parallel - end";

    this->set_done(posPerimeters);
}

void PrintObject::prepare_infill()
{
    if (! this->set_started(posPrepareInfill))
        return;

    m_print->set_status(30, _u8L("Preparing infill"));

    if (m_typed_slices) {
        // To improve robustness of detect_surfaces_type() when reslicing (working with typed slices), see GH issue #7442.
        // The preceding step (perimeter generator) only modifies extra_perimeters and the extra perimeters are only used by discover_vertical_shells()
        // with more than a single region. If this step does not use Surface::extra_perimeters or Surface::extra_perimeters is always zero, it is safe
        // to reset to the untyped slices before re-runnning detect_surfaces_type().
        for (Layer* layer : m_layers) {
            layer->restore_untyped_slices_no_extra_perimeters();
            m_print->throw_if_canceled();
        }
    }

    // This will assign a type (top/bottom/internal) to $layerm->slices.
    // Then the classifcation of $layerm->slices is transfered onto 
    // the $layerm->fill_surfaces by clipping $layerm->fill_surfaces
    // by the cummulative area of the previous $layerm->fill_surfaces.
    this->detect_surfaces_type();
    m_print->throw_if_canceled();
    
    // Decide what surfaces are to be filled.
    // Here the stTop / stBottomBridge / stBottom infill is turned to just stInternal if zero top / bottom infill layers are configured.
    // Also tiny stInternal surfaces are turned to stInternalSolid.
    BOOST_LOG_TRIVIAL(info) << "Preparing fill surfaces..." << log_memory_info();
    for (auto *layer : m_layers)
        for (auto *region : layer->m_regions) {
            region->prepare_fill_surfaces();
            m_print->throw_if_canceled();
        }


    // Add solid fills to ensure the shell vertical thickness.
    this->discover_vertical_shells();
    m_print->throw_if_canceled();

    // Debugging output.
#ifdef SLIC3R_DEBUG_SLICE_PROCESSING
    for (size_t region_id = 0; region_id < this->num_printing_regions(); ++ region_id) {
        for (const Layer *layer : m_layers) {
            LayerRegion *layerm = layer->m_regions[region_id];
            layerm->export_region_slices_to_svg_debug("3_discover_vertical_shells-final");
            layerm->export_region_fill_surfaces_to_svg_debug("3_discover_vertical_shells-final");
        } // for each layer
    } // for each region
#endif /* SLIC3R_DEBUG_SLICE_PROCESSING */

    // this will detect bridges and reverse bridges
    // and rearrange top/bottom/internal surfaces
    // It produces enlarged overlapping bridging areas.
    //
    // 1) stBottomBridge / stBottom infill is grown by 3mm and clipped by the total infill area. Bridges are detected. The areas may overlap.
    // 2) stTop is grown by 3mm and clipped by the grown bottom areas. The areas may overlap.
    // 3) Clip the internal surfaces by the grown top/bottom surfaces.
    // 4) Merge surfaces with the same style. This will mostly get rid of the overlaps.
    //FIXME This does not likely merge surfaces, which are supported by a material with different colors, but same properties.
    this->process_external_surfaces();
    m_print->throw_if_canceled();

    // Debugging output.
#ifdef SLIC3R_DEBUG_SLICE_PROCESSING
    for (size_t region_id = 0; region_id < this->num_printing_regions(); ++ region_id) {
        for (const Layer *layer : m_layers) {
            LayerRegion *layerm = layer->m_regions[region_id];
            layerm->export_region_slices_to_svg_debug("3_process_external_surfaces-final");
            layerm->export_region_fill_surfaces_to_svg_debug("3_process_external_surfaces-final");
        } // for each layer
    } // for each region
#endif /* SLIC3R_DEBUG_SLICE_PROCESSING */

    // Detect, which fill surfaces are near external layers.
    // They will be split in internal and internal-solid surfaces.
    // The purpose is to add a configurable number of solid layers to support the TOP surfaces
    // and to add a configurable number of solid layers above the BOTTOM / BOTTOMBRIDGE surfaces
    // to close these surfaces reliably.
    //FIXME Vojtech: Is this a good place to add supporting infills below sloping perimeters?
    this->discover_horizontal_shells();
    m_print->throw_if_canceled();

#ifdef SLIC3R_DEBUG_SLICE_PROCESSING
    for (size_t region_id = 0; region_id < this->num_printing_regions(); ++ region_id) {
        for (const Layer *layer : m_layers) {
            LayerRegion *layerm = layer->m_regions[region_id];
            layerm->export_region_slices_to_svg_debug("7_discover_horizontal_shells-final");
            layerm->export_region_fill_surfaces_to_svg_debug("7_discover_horizontal_shells-final");
        } // for each layer
    } // for each region
#endif /* SLIC3R_DEBUG_SLICE_PROCESSING */

    // Only active if config->infill_only_where_needed. This step trims the sparse infill,
    // so it acts as an internal support. It maintains all other infill types intact.
    // Here the internal surfaces and perimeters have to be supported by the sparse infill.
    //FIXME The surfaces are supported by a sparse infill, but the sparse infill is only as large as the area to support.
    // Likely the sparse infill will not be anchored correctly, so it will not work as intended.
    // Also one wishes the perimeters to be supported by a full infill.
    // this->clip_fill_surfaces();
    // m_print->throw_if_canceled();

#ifdef SLIC3R_DEBUG_SLICE_PROCESSING
    for (size_t region_id = 0; region_id < this->num_printing_regions(); ++ region_id) {
        for (const Layer *layer : m_layers) {
            LayerRegion *layerm = layer->m_regions[region_id];
            layerm->export_region_slices_to_svg_debug("8_clip_surfaces-final");
            layerm->export_region_fill_surfaces_to_svg_debug("8_clip_surfaces-final");
        } // for each layer
    } // for each region
#endif /* SLIC3R_DEBUG_SLICE_PROCESSING */
    
    // the following step needs to be done before combination because it may need
    // to remove only half of the combined infill
    this->bridge_over_infill();
    m_print->throw_if_canceled();

    // combine fill surfaces to honor the "infill every N layers" option
    this->combine_infill();
    m_print->throw_if_canceled();

#ifdef SLIC3R_DEBUG_SLICE_PROCESSING
    for (size_t region_id = 0; region_id < this->num_printing_regions(); ++ region_id) {
        for (const Layer *layer : m_layers) {
            LayerRegion *layerm = layer->m_regions[region_id];
            layerm->export_region_slices_to_svg_debug("9_prepare_infill-final");
            layerm->export_region_fill_surfaces_to_svg_debug("9_prepare_infill-final");
        } // for each layer
    } // for each region
    for (const Layer *layer : m_layers) {
        layer->export_region_slices_to_svg_debug("9_prepare_infill-final");
        layer->export_region_fill_surfaces_to_svg_debug("9_prepare_infill-final");
    } // for each layer
#endif /* SLIC3R_DEBUG_SLICE_PROCESSING */

    this->set_done(posPrepareInfill);
}

void PrintObject::clear_fills()
{
    for (Layer *layer : m_layers)
        layer->clear_fills();
}

void PrintObject::infill()
{
    // prerequisites
    this->prepare_infill();

    if (this->set_started(posInfill)) {
        // TRN Status for the Print calculation 
        m_print->set_status(45, _u8L("Making infill"));
        const auto& adaptive_fill_octree = this->m_adaptive_fill_octrees.first;
        const auto& support_fill_octree = this->m_adaptive_fill_octrees.second;

        BOOST_LOG_TRIVIAL(debug) << "Filling layers in parallel - start";
        tbb::parallel_for(
            tbb::blocked_range<size_t>(0, m_layers.size()),
            [this, &adaptive_fill_octree = adaptive_fill_octree, &support_fill_octree = support_fill_octree](const tbb::blocked_range<size_t>& range) {
                PRINT_OBJECT_TIME_LIMIT_MILLIS(PRINT_OBJECT_TIME_LIMIT_DEFAULT);
                for (size_t layer_idx = range.begin(); layer_idx < range.end(); ++ layer_idx) {
                    m_print->throw_if_canceled();
                    m_layers[layer_idx]->make_fills(adaptive_fill_octree.get(), support_fill_octree.get(), this->m_lightning_generator.get());
                }
            }
        );
        m_print->throw_if_canceled();
        BOOST_LOG_TRIVIAL(debug) << "Filling layers in parallel - end";
        /*  we could free memory now, but this would make this step not idempotent
        ### $_->fill_surfaces->clear for map @{$_->regions}, @{$object->layers};
        */
        this->set_done(posInfill);
    }
}

void PrintObject::ironing()
{
    if (this->set_started(posIroning)) {
        BOOST_LOG_TRIVIAL(debug) << "Ironing in parallel - start";
        tbb::parallel_for(
            // Ironing starting with layer 0 to support ironing all surfaces.
            tbb::blocked_range<size_t>(0, m_layers.size()),
            [this](const tbb::blocked_range<size_t>& range) {
                PRINT_OBJECT_TIME_LIMIT_MILLIS(PRINT_OBJECT_TIME_LIMIT_DEFAULT);
                for (size_t layer_idx = range.begin(); layer_idx < range.end(); ++ layer_idx) {
                    m_print->throw_if_canceled();
                    m_layers[layer_idx]->make_ironing();
                }
            }
        );
        m_print->throw_if_canceled();
        BOOST_LOG_TRIVIAL(debug) << "Ironing in parallel - end";
        this->set_done(posIroning);
    }
}

void PrintObject::generate_support_spots()
{
    if (this->set_started(posSupportSpotsSearch)) {
        BOOST_LOG_TRIVIAL(debug) << "Searching support spots - start";
        m_print->set_status(65, _u8L("Searching support spots"));
        if (!this->shared_regions()->generated_support_points.has_value()) {
            PrintTryCancel                cancel_func = m_print->make_try_cancel();
            SupportSpotsGenerator::Params params{this->print()->m_config.filament_type.values,
                                                 float(this->print()->m_config.perimeter_acceleration.getFloat()),
                                                 this->config().raft_layers.getInt(), this->config().brim_type.value,
                                                 float(this->config().brim_width.getFloat())};
            auto [supp_points, partial_objects] = SupportSpotsGenerator::full_search(this, cancel_func, params);
            Transform3d po_transform            = this->trafo_centered();
            if (this->layer_count() > 0) {
                po_transform = Geometry::translation_transform(Vec3d{0, 0, this->layers().front()->bottom_z()}) * po_transform;
            }
            this->m_shared_regions->generated_support_points = {po_transform, supp_points, partial_objects};
            m_print->throw_if_canceled();
        }
        BOOST_LOG_TRIVIAL(debug) << "Searching support spots - end";
        this->set_done(posSupportSpotsSearch);
    }
}

void PrintObject::generate_support_material()
{
    if (this->set_started(posSupportMaterial)) {
        this->clear_support_layers();
        if ((this->has_support() && m_layers.size() > 1) || (this->has_raft() && ! m_layers.empty())) {
            m_print->set_status(70, _u8L("Generating support material"));    
            this->_generate_support_material();
            m_print->throw_if_canceled();
        } else {
#if 0
            // Printing without supports. Empty layer means some objects or object parts are levitating,
            // therefore they cannot be printed without supports.
            for (const Layer *layer : m_layers)
                if (layer->empty())
                    throw Slic3r::SlicingError("Levitating objects cannot be printed without supports.");
#endif
        }
        this->set_done(posSupportMaterial);
    }
}

void PrintObject::estimate_curled_extrusions()
{
    if (this->set_started(posEstimateCurledExtrusions)) {
        if (this->print()->config().avoid_crossing_curled_overhangs ||
            std::any_of(this->print()->m_print_regions.begin(), this->print()->m_print_regions.end(),
                        [](const PrintRegion *region) { return region->config().enable_dynamic_overhang_speeds.getBool(); })) {
            BOOST_LOG_TRIVIAL(debug) << "Estimating areas with curled extrusions - start";
            m_print->set_status(88, _u8L("Estimating curled extrusions"));

            // Estimate curling of support material and add it to the malformaition lines of each layer
            float                         support_flow_width = support_material_flow(this, this->config().layer_height).width();
            SupportSpotsGenerator::Params params{this->print()->m_config.filament_type.values,
                                                 float(this->print()->m_config.perimeter_acceleration.getFloat()),
                                                 this->config().raft_layers.getInt(), this->config().brim_type.value,
                                                 float(this->config().brim_width.getFloat())};
            SupportSpotsGenerator::estimate_supports_malformations(this->support_layers(), support_flow_width, params);
            SupportSpotsGenerator::estimate_malformations(this->layers(), params);
            m_print->throw_if_canceled();
            BOOST_LOG_TRIVIAL(debug) << "Estimating areas with curled extrusions - end";
        }
        this->set_done(posEstimateCurledExtrusions);
    }
}

void PrintObject::calculate_overhanging_perimeters()
{
    if (this->set_started(posCalculateOverhangingPerimeters)) {
        BOOST_LOG_TRIVIAL(debug) << "Calculating overhanging perimeters - start";
        m_print->set_status(89, _u8L("Calculating overhanging perimeters"));
        std::vector<unsigned int>               extruders;
        std::unordered_set<const PrintRegion *> regions_with_dynamic_speeds;
        for (const PrintRegion *pr : this->print()->m_print_regions) {
            if (pr->config().enable_dynamic_overhang_speeds.getBool()) {
                regions_with_dynamic_speeds.insert(pr);
            }
            extruders.clear();
            pr->collect_object_printing_extruders(*this->print(), extruders);
            auto cfg = this->print()->config();
            if (std::any_of(extruders.begin(), extruders.end(),
                            [&cfg](unsigned int extruder_id) { return cfg.enable_dynamic_fan_speeds.get_at(extruder_id); })) {
                regions_with_dynamic_speeds.insert(pr);
            }
        }

        if (!regions_with_dynamic_speeds.empty()) {
            std::unordered_map<size_t, AABBTreeLines::LinesDistancer<CurledLine>> curled_lines;
            std::unordered_map<size_t, AABBTreeLines::LinesDistancer<Linef>>      unscaled_polygons_lines;
            for (const Layer *l : this->layers()) {
                curled_lines[l->id()]            = AABBTreeLines::LinesDistancer<CurledLine>{l->curled_lines};
                unscaled_polygons_lines[l->id()] = AABBTreeLines::LinesDistancer<Linef>{to_unscaled_linesf(l->lslices)};
            }
            curled_lines[size_t(-1)]            = {};
            unscaled_polygons_lines[size_t(-1)] = {};

            tbb::parallel_for(tbb::blocked_range<size_t>(0, m_layers.size()), [this, &curled_lines, &unscaled_polygons_lines,
                                                                               &regions_with_dynamic_speeds](
                                                                                  const tbb::blocked_range<size_t> &range) {
                PRINT_OBJECT_TIME_LIMIT_MILLIS(PRINT_OBJECT_TIME_LIMIT_DEFAULT);
                for (size_t layer_idx = range.begin(); layer_idx < range.end(); ++layer_idx) {
                    auto l = m_layers[layer_idx];
                    if (l->id() == 0) { // first layer, do not split
                        continue;
                    }
                    for (LayerRegion *layer_region : l->regions()) {
                        if (regions_with_dynamic_speeds.find(layer_region->m_region) == regions_with_dynamic_speeds.end()) {
                            continue;
                        }
                        size_t prev_layer_id = l->lower_layer ? l->lower_layer->id() : size_t(-1);
                        layer_region->m_perimeters =
                            ExtrusionProcessor::calculate_and_split_overhanging_extrusions(&layer_region->m_perimeters,
                                                                                           unscaled_polygons_lines[prev_layer_id],
                                                                                           curled_lines[l->id()]);
                    }
                }
            });

            m_print->throw_if_canceled();
            BOOST_LOG_TRIVIAL(debug) << "Calculating overhanging perimeters - end";
        }
        this->set_done(posCalculateOverhangingPerimeters);
    }
}
std::pair<FillAdaptive::OctreePtr, FillAdaptive::OctreePtr> PrintObject::prepare_adaptive_infill_data(
    const std::vector<std::pair<const Surface *, float>> &surfaces_w_bottom_z) const
{
    using namespace FillAdaptive;

    auto [adaptive_line_spacing, support_line_spacing] = adaptive_fill_line_spacing(*this);
    if ((adaptive_line_spacing == 0. && support_line_spacing == 0.) || this->layers().empty())
        return std::make_pair(OctreePtr(), OctreePtr());

    indexed_triangle_set mesh = this->model_object()->raw_indexed_triangle_set();
    // Rotate mesh and build octree on it with axis-aligned (standart base) cubes.
    auto to_octree = transform_to_octree().toRotationMatrix();
    its_transform(mesh, to_octree * this->trafo_centered(), true);

    // Triangulate internal bridging surfaces.
    std::vector<std::vector<Vec3d>> overhangs(std::max(surfaces_w_bottom_z.size(), size_t(1)));
    // ^ make sure vector is not empty, even with no briding surfaces we still want to build the adaptive trees later, some continue normally
    tbb::parallel_for(tbb::blocked_range<int>(0, surfaces_w_bottom_z.size()),
        [this, &to_octree, &overhangs, &surfaces_w_bottom_z](const tbb::blocked_range<int> &range) {
            PRINT_OBJECT_TIME_LIMIT_MILLIS(PRINT_OBJECT_TIME_LIMIT_DEFAULT);
            for (int surface_idx = range.begin(); surface_idx < range.end(); ++surface_idx) {
                std::vector<Vec3d> &out = overhangs[surface_idx];
                m_print->throw_if_canceled();
                append(out, triangulate_expolygon_3d(surfaces_w_bottom_z[surface_idx].first->expolygon,
                                                   surfaces_w_bottom_z[surface_idx].second));
                for (Vec3d &p : out)
                    p = (to_octree * p).eval();
            }
        });
    // and gather them.
    for (size_t i = 1; i < overhangs.size(); ++ i)
        append(overhangs.front(), std::move(overhangs[i]));

    return std::make_pair(
        adaptive_line_spacing ? build_octree(mesh, overhangs.front(), adaptive_line_spacing, false) : OctreePtr(),
        support_line_spacing  ? build_octree(mesh, overhangs.front(), support_line_spacing, true) : OctreePtr());
}

FillLightning::GeneratorPtr PrintObject::prepare_lightning_infill_data()
{
    bool     has_lightning_infill = false;
    coordf_t lightning_density    = 0.;
    size_t   lightning_cnt        = 0;
    for (size_t region_id = 0; region_id < this->num_printing_regions(); ++region_id)
        if (const PrintRegionConfig &config = this->printing_region(region_id).config(); config.fill_density > 0 && config.fill_pattern == ipLightning) {
            has_lightning_infill = true;
            lightning_density   += config.fill_density;
            ++lightning_cnt;
        }

    if (has_lightning_infill)
        lightning_density /= coordf_t(lightning_cnt);

    return has_lightning_infill ? FillLightning::build_generator(std::as_const(*this), lightning_density, [this]() -> void { this->throw_if_canceled(); }) : FillLightning::GeneratorPtr();
}

void PrintObject::clear_layers()
{
    for (Layer *l : m_layers)
        delete l;
    m_layers.clear();
}

Layer* PrintObject::add_layer(int id, coordf_t height, coordf_t print_z, coordf_t slice_z)
{
    m_layers.emplace_back(new Layer(id, this, height, print_z, slice_z));
    return m_layers.back();
}

void PrintObject::clear_support_layers()
{
    for (Layer *l : m_support_layers)
        delete l;
    m_support_layers.clear();
}

SupportLayer* PrintObject::add_support_layer(int id, int interface_id, coordf_t height, coordf_t print_z)
{
    m_support_layers.emplace_back(new SupportLayer(id, interface_id, this, height, print_z, -1));
    return m_support_layers.back();
}

SupportLayerPtrs::iterator PrintObject::insert_support_layer(SupportLayerPtrs::iterator pos, size_t id, size_t interface_id, coordf_t height, coordf_t print_z, coordf_t slice_z)
{
    return m_support_layers.insert(pos, new SupportLayer(id, interface_id, this, height, print_z, slice_z));
}

// Called by Print::apply().
// This method only accepts PrintObjectConfig and PrintRegionConfig option keys.
bool PrintObject::invalidate_state_by_config_options(
    const ConfigOptionResolver &old_config, const ConfigOptionResolver &new_config, const std::vector<t_config_option_key> &opt_keys)
{
    if (opt_keys.empty())
        return false;

    std::vector<PrintObjectStep> steps;
    bool invalidated = false;
    for (const t_config_option_key &opt_key : opt_keys) {
        if (   opt_key == "brim_width"
            || opt_key == "brim_separation"
            || opt_key == "brim_type") {
            steps.emplace_back(posSupportSpotsSearch);
            // Brim is printed below supports, support invalidates brim and skirt.
            steps.emplace_back(posSupportMaterial);
        } else if (
               opt_key == "perimeters"
            || opt_key == "extra_perimeters"
            || opt_key == "extra_perimeters_on_overhangs"
            || opt_key == "first_layer_extrusion_width"
            || opt_key == "perimeter_extrusion_width"
            || opt_key == "infill_overlap"
            || opt_key == "external_perimeters_first"
            || opt_key == "arc_fitting") {
            steps.emplace_back(posPerimeters);
        } else if (
               opt_key == "gap_fill_enabled"
            || opt_key == "gap_fill_speed") {
            // Return true if gap-fill speed has changed from zero value to non-zero or from non-zero value to zero.
            auto is_gap_fill_changed_state_due_to_speed = [&opt_key, &old_config, &new_config]() -> bool {
                if (opt_key == "gap_fill_speed") {
                    const auto *old_gap_fill_speed = old_config.option<ConfigOptionFloat>(opt_key);
                    const auto *new_gap_fill_speed = new_config.option<ConfigOptionFloat>(opt_key);
                    assert(old_gap_fill_speed && new_gap_fill_speed);
                    return (old_gap_fill_speed->value > 0.f && new_gap_fill_speed->value == 0.f) ||
                           (old_gap_fill_speed->value == 0.f && new_gap_fill_speed->value > 0.f);
                }
                return false;
            };

            // Filtering of unprintable regions in multi-material segmentation depends on if gap-fill is enabled or not.
            // So step posSlice is invalidated when gap-fill was enabled/disabled by option "gap_fill_enabled" or by
            // changing "gap_fill_speed" to force recomputation of the multi-material segmentation.
            if (this->is_mm_painted() && (opt_key == "gap_fill_enabled" || (opt_key == "gap_fill_speed" && is_gap_fill_changed_state_due_to_speed())))
                steps.emplace_back(posSlice);
            steps.emplace_back(posPerimeters);
        } else if (
               opt_key == "layer_height"
            || opt_key == "mmu_segmented_region_max_width"
            || opt_key == "mmu_segmented_region_interlocking_depth"
            || opt_key == "raft_layers"
            || opt_key == "raft_contact_distance"
            || opt_key == "slice_closing_radius"
            || opt_key == "slicing_mode") {
            steps.emplace_back(posSlice);
		} else if (
               opt_key == "elefant_foot_compensation"
            || opt_key == "support_material_contact_distance" 
            //w12
            || opt_key == "xy_size_compensation"
            || opt_key == "xy_hole_compensation"
            || opt_key == "xy_contour_compensation") {
            steps.emplace_back(posSlice);
        } else if (opt_key == "support_material") {
            steps.emplace_back(posSupportMaterial);
            if (m_config.support_material_contact_distance == 0.) {
            	// Enabling / disabling supports while soluble support interface is enabled.
            	// This changes the bridging logic (bridging enabled without supports, disabled with supports).
            	// Reset everything.
            	// See GH #1482 for details.
	            steps.emplace_back(posSlice);
	        }
        } else if (
        	   opt_key == "support_material_auto"
            || opt_key == "support_material_angle"
            || opt_key == "support_material_buildplate_only"
            || opt_key == "support_material_enforce_layers"
            || opt_key == "support_material_extruder"
            || opt_key == "support_material_extrusion_width"
            || opt_key == "support_material_bottom_contact_distance"
            || opt_key == "support_material_interface_layers"
            || opt_key == "support_material_bottom_interface_layers"
            || opt_key == "support_material_interface_pattern"
            || opt_key == "support_material_interface_contact_loops"
            || opt_key == "support_material_interface_extruder"
            || opt_key == "support_material_interface_spacing"
            || opt_key == "support_material_pattern"
            || opt_key == "support_material_style"
            || opt_key == "support_material_xy_spacing"
            || opt_key == "support_material_spacing"
            || opt_key == "support_material_closing_radius"
            || opt_key == "support_material_synchronize_layers"
            || opt_key == "support_material_threshold"
            || opt_key == "support_material_with_sheath"
            || opt_key == "support_tree_angle"
            || opt_key == "support_tree_angle_slow"
            || opt_key == "support_tree_branch_diameter"
            || opt_key == "support_tree_branch_diameter_angle"
            || opt_key == "support_tree_branch_diameter_double_wall"
            || opt_key == "support_tree_top_rate"
            || opt_key == "support_tree_branch_distance"
            || opt_key == "support_tree_tip_diameter"
            || opt_key == "raft_expansion"
            || opt_key == "raft_first_layer_density"
            || opt_key == "raft_first_layer_expansion"
            || opt_key == "dont_support_bridges"
            || opt_key == "first_layer_extrusion_width") {
            steps.emplace_back(posSupportMaterial);
        } else if (opt_key == "bottom_solid_layers") {
            steps.emplace_back(posPrepareInfill);
            if (m_print->config().spiral_vase) {
                // Changing the number of bottom layers when a spiral vase is enabled requires re-slicing the object again.
                // Otherwise, holes in the bottom layers could be filled, as is reported in GH #5528.
                steps.emplace_back(posSlice);
            }
        } else if (
               opt_key == "interface_shells"
            || opt_key == "infill_only_where_needed"
            || opt_key == "infill_every_layers"
            || opt_key == "solid_infill_every_layers"
            || opt_key == "bottom_solid_min_thickness"
            || opt_key == "top_solid_layers"
            || opt_key == "top_solid_min_thickness"
            || opt_key == "solid_infill_below_area"
            || opt_key == "infill_extruder"
            || opt_key == "solid_infill_extruder"
            || opt_key == "infill_extrusion_width"
            || opt_key == "bridge_angle") {
            steps.emplace_back(posPrepareInfill);
        } else if (
               opt_key == "top_fill_pattern"
            || opt_key == "bottom_fill_pattern"
            || opt_key == "external_fill_link_max_length"
            || opt_key == "fill_angle"
            || opt_key == "infill_anchor"
            || opt_key == "infill_anchor_max"
            || opt_key == "top_infill_extrusion_width"
            || opt_key == "first_layer_extrusion_width") {
            steps.emplace_back(posInfill);
        } else if (opt_key == "fill_pattern") {
            steps.emplace_back(posPrepareInfill);
        } else if (opt_key == "fill_density") {
            // One likely wants to reslice only when switching between zero infill to simulate boolean difference (subtracting volumes),
            // normal infill and 100% (solid) infill.
            const auto *old_density = old_config.option<ConfigOptionPercent>(opt_key);
            const auto *new_density = new_config.option<ConfigOptionPercent>(opt_key);
            assert(old_density && new_density);
            //FIXME Vojtech is not quite sure about the 100% here, maybe it is not needed.
            if (is_approx(old_density->value, 0.) || is_approx(old_density->value, 100.) ||
                is_approx(new_density->value, 0.) || is_approx(new_density->value, 100.))
                steps.emplace_back(posPerimeters);
            steps.emplace_back(posPrepareInfill);
        } else if (opt_key == "solid_infill_extrusion_width") {
            // This value is used for calculating perimeter - infill overlap, thus perimeters need to be recalculated.
            steps.emplace_back(posPerimeters);
            steps.emplace_back(posPrepareInfill);
        } else if (
               opt_key == "external_perimeter_extrusion_width"
            || opt_key == "perimeter_extruder"
            || opt_key == "fuzzy_skin"
            || opt_key == "fuzzy_skin_thickness"
            || opt_key == "fuzzy_skin_point_dist"
            || opt_key == "overhangs"
            || opt_key == "thin_walls"
            || opt_key == "thick_bridges") {
            steps.emplace_back(posPerimeters);
            steps.emplace_back(posSupportMaterial);
        } else if (opt_key == "bridge_flow_ratio") {
            if (m_config.support_material_contact_distance > 0.) {
            	// Only invalidate due to bridging if bridging is enabled.
            	// If later "support_material_contact_distance" is modified, the complete PrintObject is invalidated anyway.
            	steps.emplace_back(posPerimeters);
            	steps.emplace_back(posInfill);
	            steps.emplace_back(posSupportMaterial);
	        }
        } else if (
            opt_key == "perimeter_generator"
            || opt_key == "wall_transition_length"
            || opt_key == "wall_transition_filter_deviation"
            || opt_key == "wall_transition_angle"
            || opt_key == "wall_distribution_count"
            || opt_key == "min_feature_size"
            || opt_key == "min_bead_width") {
            steps.emplace_back(posSlice);
        } else if (
               opt_key == "seam_position"
            || opt_key == "seam_preferred_direction"
            || opt_key == "seam_preferred_direction_jitter"
            || opt_key == "support_material_speed"
            || opt_key == "support_material_interface_speed"
            || opt_key == "bridge_speed"
            || opt_key == "enable_dynamic_overhang_speeds"
            || opt_key == "overhang_speed_0"
            || opt_key == "overhang_speed_1"
            || opt_key == "overhang_speed_2"
            || opt_key == "overhang_speed_3"
            || opt_key == "external_perimeter_speed"
            || opt_key == "small_perimeter_speed"
            || opt_key == "solid_infill_speed"
            || opt_key == "top_solid_infill_speed") {
            invalidated |= m_print->invalidate_step(psGCodeExport);
        } else if (
               opt_key == "wipe_into_infill"
            || opt_key == "wipe_into_objects"
            || opt_key == "infill_speed"
            || opt_key == "perimeter_speed") {
            invalidated |= m_print->invalidate_step(psWipeTower);
            invalidated |= m_print->invalidate_step(psGCodeExport);
        } else {
            // for legacy, if we can't handle this option let's invalidate all steps
            this->invalidate_all_steps();
            invalidated = true;
        }
    }

    sort_remove_duplicates(steps);
    for (PrintObjectStep step : steps)
        invalidated |= this->invalidate_step(step);
    return invalidated;
}

bool PrintObject::invalidate_step(PrintObjectStep step)
{
	bool invalidated = Inherited::invalidate_step(step);
    
    // propagate to dependent steps
    if (step == posPerimeters) {
		invalidated |= this->invalidate_steps({ posPrepareInfill, posInfill, posIroning,  posSupportSpotsSearch, posEstimateCurledExtrusions, posCalculateOverhangingPerimeters });
        invalidated |= m_print->invalidate_steps({ psSkirtBrim });
    } else if (step == posPrepareInfill) {
        invalidated |= this->invalidate_steps({ posInfill, posIroning, posSupportSpotsSearch});
    } else if (step == posInfill) {
        invalidated |= this->invalidate_steps({ posIroning, posSupportSpotsSearch });
        invalidated |= m_print->invalidate_steps({ psSkirtBrim });
    } else if (step == posSlice) {
        invalidated |= this->invalidate_steps({posPerimeters, posPrepareInfill, posInfill, posIroning, posSupportSpotsSearch,
                                               posSupportMaterial, posEstimateCurledExtrusions, posCalculateOverhangingPerimeters});
        invalidated |= m_print->invalidate_steps({ psSkirtBrim });
        m_slicing_params.valid = false;
    } else if (step == posSupportMaterial) {
        invalidated |= m_print->invalidate_steps({ psSkirtBrim,  });
        invalidated |= this->invalidate_steps({ posEstimateCurledExtrusions });
        m_slicing_params.valid = false;
    }

    // invalidate alerts step always, since it depends on everything (except supports, but with supports enabled it is skipped anyway.)
    invalidated |= m_print->invalidate_step(psAlertWhenSupportsNeeded);
    // Wipe tower depends on the ordering of extruders, which in turn depends on everything.
    // It also decides about what the wipe_into_infill / wipe_into_object features will do,
    // and that too depends on many of the settings.
    invalidated |= m_print->invalidate_step(psWipeTower);
    // Invalidate G-code export in any case.
    invalidated |= m_print->invalidate_step(psGCodeExport);
    return invalidated;
}

bool PrintObject::invalidate_all_steps()
{
	// First call the "invalidate" functions, which may cancel background processing.
    bool result = Inherited::invalidate_all_steps() | m_print->invalidate_all_steps();
	// Then reset some of the depending values.
	m_slicing_params.valid = false;
	return result;
}

// Called on main thread with stopped or paused background processing to let PrintObject release data for its milestones that were invalidated or canceled.
void PrintObject::cleanup()
{
    if (this->query_reset_dirty_step_unguarded(posInfill))
        this->clear_fills();
    if (this->query_reset_dirty_step_unguarded(posSupportMaterial))
        this->clear_support_layers();
}

// This function analyzes slices of a region (SurfaceCollection slices).
// Each region slice (instance of Surface) is analyzed, whether it is supported or whether it is the top surface.
// Initially all slices are of type stInternal.
// Slices are compared against the top / bottom slices and regions and classified to the following groups:
// stTop          - Part of a region, which is not covered by any upper layer. This surface will be filled with a top solid infill.
// stBottomBridge - Part of a region, which is not fully supported, but it hangs in the air, or it hangs losely on a support or a raft.
// stBottom       - Part of a region, which is not supported by the same region, but it is supported either by another region, or by a soluble interface layer.
// stInternal     - Part of a region, which is supported by the same region type.
// If a part of a region is of stBottom and stTop, the stBottom wins.
void PrintObject::detect_surfaces_type()
{
    BOOST_LOG_TRIVIAL(info) << "Detecting solid surfaces..." << log_memory_info();

    // Interface shells: the intersecting parts are treated as self standing objects supporting each other.
    // Each of the objects will have a full number of top / bottom layers, even if these top / bottom layers
    // are completely hidden inside a collective body of intersecting parts.
    // This is useful if one of the parts is to be dissolved, or if it is transparent and the internal shells
    // should be visible.
    bool spiral_vase      = this->print()->config().spiral_vase.value;
    bool interface_shells = ! spiral_vase && m_config.interface_shells.value;
    size_t num_layers     = spiral_vase ? std::min(size_t(this->printing_region(0).config().bottom_solid_layers), m_layers.size()) : m_layers.size();

    for (size_t region_id = 0; region_id < this->num_printing_regions(); ++ region_id) {
        BOOST_LOG_TRIVIAL(debug) << "Detecting solid surfaces for region " << region_id << " in parallel - start";
#ifdef SLIC3R_DEBUG_SLICE_PROCESSING
        for (Layer *layer : m_layers)
            layer->m_regions[region_id]->export_region_fill_surfaces_to_svg_debug("1_detect_surfaces_type-initial");
#endif /* SLIC3R_DEBUG_SLICE_PROCESSING */

        // If interface shells are allowed, the region->surfaces cannot be overwritten as they may be used by other threads.
        // Cache the result of the following parallel_loop.
        std::vector<Surfaces> surfaces_new;
        if (interface_shells)
            surfaces_new.assign(num_layers, Surfaces());

        tbb::parallel_for(
            tbb::blocked_range<size_t>(0, 
            	spiral_vase ?
            		// In spiral vase mode, reserve the last layer for the top surface if more than 1 layer is planned for the vase bottom.
            		((num_layers > 1) ? num_layers - 1 : num_layers) :
            		// In non-spiral vase mode, go over all layers.
            		m_layers.size()),
            [this, region_id, interface_shells, &surfaces_new](const tbb::blocked_range<size_t>& range) {
                PRINT_OBJECT_TIME_LIMIT_MILLIS(PRINT_OBJECT_TIME_LIMIT_DEFAULT);
                // If we have soluble support material, don't bridge. The overhang will be squished against a soluble layer separating
                // the support from the print.
                SurfaceType surface_type_bottom_other =
                    (this->has_support() && m_config.support_material_contact_distance.value == 0) ?
                    stBottom : stBottomBridge;
                for (size_t idx_layer = range.begin(); idx_layer < range.end(); ++ idx_layer) {
                    m_print->throw_if_canceled();
                    // BOOST_LOG_TRIVIAL(trace) << "Detecting solid surfaces for region " << region_id << " and layer " << layer->print_z;
                    Layer       *layer  = m_layers[idx_layer];
                    LayerRegion *layerm = layer->m_regions[region_id];
                    // comparison happens against the *full* slices (considering all regions)
                    // unless internal shells are requested
                    Layer       *upper_layer = (idx_layer + 1 < this->layer_count()) ? m_layers[idx_layer + 1] : nullptr;
                    Layer       *lower_layer = (idx_layer > 0) ? m_layers[idx_layer - 1] : nullptr;
                    // collapse very narrow parts (using the safety offset in the diff is not enough)
                    float        offset = layerm->flow(frExternalPerimeter).scaled_width() / 10.f;

                    // find top surfaces (difference between current surfaces
                    // of current layer and upper one)
                    Surfaces top;
                    if (upper_layer) {
                        ExPolygons upper_slices = interface_shells ? 
                            diff_ex(layerm->slices().surfaces, upper_layer->m_regions[region_id]->slices().surfaces, ApplySafetyOffset::Yes) :
                            diff_ex(layerm->slices().surfaces, upper_layer->lslices, ApplySafetyOffset::Yes);
                        surfaces_append(top, opening_ex(upper_slices, offset), stTop);
                    } else {
                        // if no upper layer, all surfaces of this one are solid
                        // we clone surfaces because we're going to clear the slices collection
                        top = layerm->slices().surfaces;
                        for (Surface &surface : top)
                            surface.surface_type = stTop;
                    }
                    
                    // Find bottom surfaces (difference between current surfaces of current layer and lower one).
                    Surfaces bottom;
                    if (lower_layer) {
#if 0
                        //FIXME Why is this branch failing t\multi.t ?
                        Polygons lower_slices = interface_shells ? 
                            to_polygons(lower_layer->get_region(region_id)->slices.surfaces) : 
                            to_polygons(lower_layer->slices);
                        surfaces_append(bottom,
                            opening_ex(diff(layerm->slices.surfaces, lower_slices, true), offset),
                            surface_type_bottom_other);
#else
                        // Any surface lying on the void is a true bottom bridge (an overhang)
                        surfaces_append(
                            bottom,
                            opening_ex(
                                diff_ex(layerm->slices().surfaces, lower_layer->lslices, ApplySafetyOffset::Yes),
                                offset),
                            surface_type_bottom_other);
                        // if user requested internal shells, we need to identify surfaces
                        // lying on other slices not belonging to this region
                        if (interface_shells) {
                            // non-bridging bottom surfaces: any part of this layer lying 
                            // on something else, excluding those lying on our own region
                            surfaces_append(
                                bottom,
                                opening_ex(
                                    diff_ex(
                                        intersection(layerm->slices().surfaces, lower_layer->lslices), // supported
                                        lower_layer->m_regions[region_id]->slices().surfaces,
                                        ApplySafetyOffset::Yes),
                                    offset),
                                stBottom);
                        }
#endif
                    } else {
                        // if no lower layer, all surfaces of this one are solid
                        // we clone surfaces because we're going to clear the slices collection
                        bottom = layerm->slices().surfaces;
                        for (Surface &surface : bottom)
                            surface.surface_type = stBottom;
                    }
                    
                    // now, if the object contained a thin membrane, we could have overlapping bottom
                    // and top surfaces; let's do an intersection to discover them and consider them
                    // as bottom surfaces (to allow for bridge detection)
                    if (! top.empty() && ! bottom.empty()) {
        //                Polygons overlapping = intersection(to_polygons(top), to_polygons(bottom));
        //                Slic3r::debugf "  layer %d contains %d membrane(s)\n", $layerm->layer->id, scalar(@$overlapping)
        //                    if $Slic3r::debug;
                        Polygons top_polygons = to_polygons(std::move(top));
                        top.clear();
                        surfaces_append(top, diff_ex(top_polygons, bottom), stTop);
                    }

        #ifdef SLIC3R_DEBUG_SLICE_PROCESSING
                    {
                        static int iRun = 0;
                        std::vector<std::pair<Slic3r::ExPolygons, SVG::ExPolygonAttributes>> expolygons_with_attributes;
                        expolygons_with_attributes.emplace_back(std::make_pair(union_ex(top),                             SVG::ExPolygonAttributes("green")));
                        expolygons_with_attributes.emplace_back(std::make_pair(union_ex(bottom),                          SVG::ExPolygonAttributes("brown")));
                        expolygons_with_attributes.emplace_back(std::make_pair(to_expolygons(layerm->slices().surfaces),  SVG::ExPolygonAttributes("black")));
                        SVG::export_expolygons(debug_out_path("1_detect_surfaces_type_%d_region%d-layer_%f.svg", iRun ++, region_id, layer->print_z).c_str(), expolygons_with_attributes);
                    }
        #endif /* SLIC3R_DEBUG_SLICE_PROCESSING */
                    
                    // save surfaces to layer
                    Surfaces &surfaces_out = interface_shells ? surfaces_new[idx_layer] : layerm->m_slices.surfaces;
                    Surfaces  surfaces_backup;
                    if (! interface_shells) {
                        surfaces_backup = std::move(surfaces_out);
                        surfaces_out.clear();
                    }
                    const Surfaces &surfaces_prev = interface_shells ? layerm->slices().surfaces : surfaces_backup;

                    // find internal surfaces (difference between top/bottom surfaces and others)
                    {
                        Polygons topbottom = to_polygons(top);
                        polygons_append(topbottom, to_polygons(bottom));
                        surfaces_append(surfaces_out, diff_ex(surfaces_prev, topbottom), stInternal);
                    }

                    surfaces_append(surfaces_out, std::move(top));
                    surfaces_append(surfaces_out, std::move(bottom));
                    
        //            Slic3r::debugf "  layer %d has %d bottom, %d top and %d internal surfaces\n",
        //                $layerm->layer->id, scalar(@bottom), scalar(@top), scalar(@internal) if $Slic3r::debug;

        #ifdef SLIC3R_DEBUG_SLICE_PROCESSING
                    layerm->export_region_slices_to_svg_debug("detect_surfaces_type-final");
        #endif /* SLIC3R_DEBUG_SLICE_PROCESSING */
                }
            }
        ); // for each layer of a region
        m_print->throw_if_canceled();

        if (interface_shells) {
            // Move surfaces_new to layerm->slices.surfaces
            for (size_t idx_layer = 0; idx_layer < num_layers; ++ idx_layer)
                m_layers[idx_layer]->m_regions[region_id]->m_slices.set(std::move(surfaces_new[idx_layer]));
        }

        if (spiral_vase) {
        	if (num_layers > 1)
	        	// Turn the last bottom layer infill to a top infill, so it will be extruded with a proper pattern.
	        	m_layers[num_layers - 1]->m_regions[region_id]->m_slices.set_type(stTop);
	        for (size_t i = num_layers; i < m_layers.size(); ++ i)
	        	m_layers[i]->m_regions[region_id]->m_slices.set_type(stInternal);
        }

        BOOST_LOG_TRIVIAL(debug) << "Detecting solid surfaces for region " << region_id << " - clipping in parallel - start";
        // Fill in layerm->fill_surfaces by trimming the layerm->slices by the cummulative layerm->fill_surfaces.
        tbb::parallel_for(
            tbb::blocked_range<size_t>(0, m_layers.size()),
            [this, region_id](const tbb::blocked_range<size_t>& range) {
                PRINT_OBJECT_TIME_LIMIT_MILLIS(PRINT_OBJECT_TIME_LIMIT_DEFAULT);
                for (size_t idx_layer = range.begin(); idx_layer < range.end(); ++ idx_layer) {
                    m_print->throw_if_canceled();
                    LayerRegion *layerm = m_layers[idx_layer]->m_regions[region_id];
                    layerm->slices_to_fill_surfaces_clipped();
#ifdef SLIC3R_DEBUG_SLICE_PROCESSING
                    layerm->export_region_fill_surfaces_to_svg_debug("1_detect_surfaces_type-final");
#endif /* SLIC3R_DEBUG_SLICE_PROCESSING */
                } // for each layer of a region
            });
        m_print->throw_if_canceled();
        BOOST_LOG_TRIVIAL(debug) << "Detecting solid surfaces for region " << region_id << " - clipping in parallel - end";
    } // for each this->print->region_count

    // Mark the object to have the region slices classified (typed, which also means they are split based on whether they are supported, bridging, top layers etc.)
    m_typed_slices = true;
}

void PrintObject::process_external_surfaces()
{
    BOOST_LOG_TRIVIAL(info) << "Processing external surfaces..." << log_memory_info();

    // Cached surfaces covered by some extrusion, defining regions, over which the from the surfaces one layer higher are allowed to expand.
    std::vector<Polygons> surfaces_covered;
    // Is there any printing region, that has zero infill? If so, then we don't want the expansion to be performed over the complete voids, but only
    // over voids, which are supported by the layer below.
    bool 				  has_voids = false;
	for (size_t region_id = 0; region_id < this->num_printing_regions(); ++ region_id)
		if (this->printing_region(region_id).config().fill_density == 0) {
			has_voids = true;
			break;
		}
	if (has_voids && m_layers.size() > 1) {
	    // All but stInternal fill surfaces will get expanded and possibly trimmed.
	    std::vector<unsigned char> layer_expansions_and_voids(m_layers.size(), false);
	    for (size_t layer_idx = 1; layer_idx < m_layers.size(); ++ layer_idx) {
	    	const Layer *layer = m_layers[layer_idx];
	    	bool expansions = false;
	    	bool voids      = false;
	    	for (const LayerRegion *layerm : layer->regions()) {
	    		for (const Surface &surface : layerm->fill_surfaces()) {
	    			if (surface.surface_type == stInternal)
	    				voids = true;
	    			else
	    				expansions = true;
	    			if (voids && expansions) {
	    				layer_expansions_and_voids[layer_idx] = true;
	    				goto end;
	    			}
	    		}
	    	}
		end:;
		}
	    BOOST_LOG_TRIVIAL(debug) << "Collecting surfaces covered with extrusions in parallel - start";
	    surfaces_covered.resize(m_layers.size() - 1, Polygons());
    	auto unsupported_width = - float(scale_(0.3 * EXTERNAL_INFILL_MARGIN));
	    tbb::parallel_for(
	        tbb::blocked_range<size_t>(0, m_layers.size() - 1),
	        [this, &surfaces_covered, &layer_expansions_and_voids, unsupported_width](const tbb::blocked_range<size_t>& range) {
                PRINT_OBJECT_TIME_LIMIT_MILLIS(PRINT_OBJECT_TIME_LIMIT_DEFAULT);
	            for (size_t layer_idx = range.begin(); layer_idx < range.end(); ++ layer_idx)
	            	if (layer_expansions_and_voids[layer_idx + 1]) {
                        // Layer above is partially filled with solid infill (top, bottom, bridging...),
                        // while some sparse inill regions are empty (0% infill).
		                m_print->throw_if_canceled();
		                Polygons voids;
		                for (const LayerRegion *layerm : m_layers[layer_idx]->regions()) {
		                	if (layerm->region().config().fill_density.value == 0.)
		                		for (const Surface &surface : layerm->fill_surfaces())
		                			// Shrink the holes, let the layer above expand slightly inside the unsupported areas.
		                			polygons_append(voids, offset(surface.expolygon, unsupported_width));
		                }
		                surfaces_covered[layer_idx] = diff(m_layers[layer_idx]->lslices, voids);
	            	}
	        }
	    );
	    m_print->throw_if_canceled();
	    BOOST_LOG_TRIVIAL(debug) << "Collecting surfaces covered with extrusions in parallel - end";
	}

	for (size_t region_id = 0; region_id < this->num_printing_regions(); ++region_id) {
        BOOST_LOG_TRIVIAL(debug) << "Processing external surfaces for region " << region_id << " in parallel - start";
        tbb::parallel_for(
            tbb::blocked_range<size_t>(0, m_layers.size()),
            [this, &surfaces_covered, region_id](const tbb::blocked_range<size_t>& range) {
                PRINT_OBJECT_TIME_LIMIT_MILLIS(PRINT_OBJECT_TIME_LIMIT_DEFAULT);
                for (size_t layer_idx = range.begin(); layer_idx < range.end(); ++ layer_idx) {
                    m_print->throw_if_canceled();
                    // BOOST_LOG_TRIVIAL(trace) << "Processing external surface, layer" << m_layers[layer_idx]->print_z;
                    m_layers[layer_idx]->get_region(int(region_id))->process_external_surfaces(
                        // lower layer
                    	(layer_idx == 0) ? nullptr : m_layers[layer_idx - 1],
                        // lower layer polygons with density > 0%
                    	(layer_idx == 0 || surfaces_covered.empty() || surfaces_covered[layer_idx - 1].empty()) ? nullptr : &surfaces_covered[layer_idx - 1]);
                }
            }
        );
        m_print->throw_if_canceled();
        BOOST_LOG_TRIVIAL(debug) << "Processing external surfaces for region " << region_id << " in parallel - end";
    }

    if (this->has_raft() && ! m_layers.empty()) {
        // Adjust bridge direction of 1st object layer over raft to be perpendicular to the raft contact layer direction.
        Layer &layer = *m_layers.front();
        assert(layer.id() > 0);
        for (LayerRegion *layerm : layer.regions())
            for (Surface &fill : layerm->m_fill_surfaces)
                fill.bridge_angle = -1;
    }
} // void PrintObject::process_external_surfaces()

void PrintObject::discover_vertical_shells()
{
    BOOST_LOG_TRIVIAL(info) << "Discovering vertical shells..." << log_memory_info();

    struct DiscoverVerticalShellsCacheEntry
    {
        // Collected polygons, offsetted
        Polygons    top_surfaces;
        Polygons    bottom_surfaces;
        Polygons    holes;
    };
    bool     spiral_vase      = this->print()->config().spiral_vase.value;
    size_t   num_layers       = spiral_vase ? std::min(size_t(this->printing_region(0).config().bottom_solid_layers), m_layers.size()) : m_layers.size();
    std::vector<DiscoverVerticalShellsCacheEntry> cache_top_botom_regions(num_layers, DiscoverVerticalShellsCacheEntry());
    bool top_bottom_surfaces_all_regions = this->num_printing_regions() > 1 && ! m_config.interface_shells.value;
//    static constexpr const float top_bottom_expansion_coeff = 1.05f;
    // Just a tiny fraction of an infill extrusion width to merge neighbor regions reliably.
    static constexpr const float top_bottom_expansion_coeff = 0.05f;
    if (top_bottom_surfaces_all_regions) {
        // This is a multi-material print and interface_shells are disabled, meaning that the vertical shell thickness
        // is calculated over all materials.
        BOOST_LOG_TRIVIAL(debug) << "Discovering vertical shells in parallel - start : cache top / bottom";
        //FIXME Improve the heuristics for a grain size.
        size_t grain_size = std::max(num_layers / 16, size_t(1));
        tbb::parallel_for(
            tbb::blocked_range<size_t>(0, num_layers, grain_size),
            [this, &cache_top_botom_regions](const tbb::blocked_range<size_t>& range) {
                PRINT_OBJECT_TIME_LIMIT_MILLIS(PRINT_OBJECT_TIME_LIMIT_DEFAULT);
                const std::initializer_list<SurfaceType> surfaces_bottom { stBottom, stBottomBridge };
                const size_t num_regions = this->num_printing_regions();
                for (size_t idx_layer = range.begin(); idx_layer < range.end(); ++ idx_layer) {
                    m_print->throw_if_canceled();
                    const Layer                      &layer = *m_layers[idx_layer];
                    DiscoverVerticalShellsCacheEntry &cache = cache_top_botom_regions[idx_layer];
                    // Simulate single set of perimeters over all merged regions.
                    float                             perimeter_offset = 0.f;
                    float                             perimeter_min_spacing = FLT_MAX;
#ifdef SLIC3R_DEBUG_SLICE_PROCESSING
                    static size_t debug_idx = 0;
                    ++ debug_idx;
#endif /* SLIC3R_DEBUG_SLICE_PROCESSING */
                    for (size_t region_id = 0; region_id < num_regions; ++ region_id) {
                        LayerRegion &layerm               = *layer.m_regions[region_id];
                        float        top_bottom_expansion = float(layerm.flow(frSolidInfill).scaled_spacing()) * top_bottom_expansion_coeff;
                        // Top surfaces.
                        append(cache.top_surfaces, offset(layerm.slices().filter_by_type(stTop), top_bottom_expansion));
//                        append(cache.top_surfaces, offset(layerm.fill_surfaces().filter_by_type(stTop), top_bottom_expansion));
                        // Bottom surfaces.
                        append(cache.bottom_surfaces, offset(layerm.slices().filter_by_types(surfaces_bottom), top_bottom_expansion));
//                        append(cache.bottom_surfaces, offset(layerm.fill_surfaces().filter_by_types(surfaces_bottom), top_bottom_expansion));
                        // Calculate the maximum perimeter offset as if the slice was extruded with a single extruder only.
                        // First find the maxium number of perimeters per region slice.
                        unsigned int perimeters = 0;
                        for (const Surface &s : layerm.slices())
                            perimeters = std::max<unsigned int>(perimeters, s.extra_perimeters);
                        perimeters += layerm.region().config().perimeters.value;
                        // Then calculate the infill offset.
                        if (perimeters > 0) {
                            Flow extflow = layerm.flow(frExternalPerimeter);
                            Flow flow    = layerm.flow(frPerimeter);
                            perimeter_offset = std::max(perimeter_offset,
                                0.5f * float(extflow.scaled_width() + extflow.scaled_spacing()) + (float(perimeters) - 1.f) * flow.scaled_spacing());
                            perimeter_min_spacing = std::min(perimeter_min_spacing, float(std::min(extflow.scaled_spacing(), flow.scaled_spacing())));
                        }
                        polygons_append(cache.holes, to_polygons(layerm.fill_expolygons()));
                    }
                    // Save some computing time by reducing the number of polygons.
                    cache.top_surfaces    = union_(cache.top_surfaces);
                    cache.bottom_surfaces = union_(cache.bottom_surfaces);
                    // For a multi-material print, simulate perimeter / infill split as if only a single extruder has been used for the whole print.
                    if (perimeter_offset > 0.) {
                        // The layer.lslices are forced to merge by expanding them first.
                        polygons_append(cache.holes, offset2(layer.lslices, 0.3f * perimeter_min_spacing, - perimeter_offset - 0.3f * perimeter_min_spacing));
#ifdef SLIC3R_DEBUG_SLICE_PROCESSING
                        {
                            Slic3r::SVG svg(debug_out_path("discover_vertical_shells-extra-holes-%d.svg", debug_idx), get_extents(layer.lslices));
                            svg.draw(layer.lslices, "blue");
                            svg.draw(union_ex(cache.holes), "red");
                            svg.draw_outline(union_ex(cache.holes), "black", "blue", scale_(0.05));
                            svg.Close(); 
                        }
#endif /* SLIC3R_DEBUG_SLICE_PROCESSING */
                    }
                    cache.holes = union_(cache.holes);
                }
            });
        m_print->throw_if_canceled();
        BOOST_LOG_TRIVIAL(debug) << "Discovering vertical shells in parallel - end : cache top / bottom";
    }

    for (size_t region_id = 0; region_id < this->num_printing_regions(); ++ region_id) {
        //FIXME Improve the heuristics for a grain size.
        size_t grain_size = std::max(num_layers / 16, size_t(1));

        if (! top_bottom_surfaces_all_regions) {
            // This is either a single material print, or a multi-material print and interface_shells are enabled, meaning that the vertical shell thickness
            // is calculated over a single material.
            BOOST_LOG_TRIVIAL(debug) << "Discovering vertical shells for region " << region_id << " in parallel - start : cache top / bottom";
            tbb::parallel_for(
                tbb::blocked_range<size_t>(0, num_layers, grain_size),
                [this, region_id, &cache_top_botom_regions](const tbb::blocked_range<size_t>& range) {
                    PRINT_OBJECT_TIME_LIMIT_MILLIS(PRINT_OBJECT_TIME_LIMIT_DEFAULT);
                    const std::initializer_list<SurfaceType> surfaces_bottom { stBottom, stBottomBridge };
                    for (size_t idx_layer = range.begin(); idx_layer < range.end(); ++ idx_layer) {
                        m_print->throw_if_canceled();
                        Layer       &layer                = *m_layers[idx_layer];
                        LayerRegion &layerm               = *layer.m_regions[region_id];
                        float        top_bottom_expansion = float(layerm.flow(frSolidInfill).scaled_spacing()) * top_bottom_expansion_coeff;
                        // Top surfaces.
                        auto &cache = cache_top_botom_regions[idx_layer];
                        cache.top_surfaces = offset(layerm.slices().filter_by_type(stTop), top_bottom_expansion);
//                        append(cache.top_surfaces, offset(layerm.fill_surfaces().filter_by_type(stTop), top_bottom_expansion));
                        // Bottom surfaces.
                        cache.bottom_surfaces = offset(layerm.slices().filter_by_types(surfaces_bottom), top_bottom_expansion);
//                        append(cache.bottom_surfaces, offset(layerm.fill_surfaces().filter_by_types(surfaces_bottom), top_bottom_expansion));
                        // Holes over all regions. Only collect them once, they are valid for all region_id iterations.
                        if (cache.holes.empty()) {
                            for (size_t region_id = 0; region_id < layer.regions().size(); ++ region_id)
                                polygons_append(cache.holes, to_polygons(layer.regions()[region_id]->fill_expolygons()));
                        }
                    }
                });
            m_print->throw_if_canceled();
            BOOST_LOG_TRIVIAL(debug) << "Discovering vertical shells for region " << region_id << " in parallel - end : cache top / bottom";
        }

        BOOST_LOG_TRIVIAL(debug) << "Discovering vertical shells for region " << region_id << " in parallel - start : ensure vertical wall thickness";
        grain_size = 1;
        tbb::parallel_for(
            tbb::blocked_range<size_t>(0, num_layers, grain_size),
            [this, region_id, &cache_top_botom_regions]
            (const tbb::blocked_range<size_t>& range) {
                PRINT_OBJECT_TIME_LIMIT_MILLIS(PRINT_OBJECT_TIME_LIMIT_DEFAULT);
                // printf("discover_vertical_shells from %d to %d\n", range.begin(), range.end());
                for (size_t idx_layer = range.begin(); idx_layer < range.end(); ++ idx_layer) {
                    m_print->throw_if_canceled();
#ifdef SLIC3R_DEBUG_SLICE_PROCESSING
        			static size_t debug_idx = 0;
        			++ debug_idx;
#endif /* SLIC3R_DEBUG_SLICE_PROCESSING */

                    Layer       	        *layer          = m_layers[idx_layer];
                    LayerRegion 	        *layerm         = layer->m_regions[region_id];
                    const PrintRegionConfig &region_config  = layerm->region().config();

#ifdef SLIC3R_DEBUG_SLICE_PROCESSING
                    layerm->export_region_slices_to_svg_debug("3_discover_vertical_shells-initial");
                    layerm->export_region_fill_surfaces_to_svg_debug("3_discover_vertical_shells-initial");
#endif /* SLIC3R_DEBUG_SLICE_PROCESSING */

                    Flow         solid_infill_flow   = layerm->flow(frSolidInfill);
                    coord_t      infill_line_spacing = solid_infill_flow.scaled_spacing(); 
                    // Find a union of perimeters below / above this surface to guarantee a minimum shell thickness.
                    Polygons shell;
                    Polygons holes;
#ifdef SLIC3R_DEBUG_SLICE_PROCESSING
                    ExPolygons shell_ex;
#endif /* SLIC3R_DEBUG_SLICE_PROCESSING */
                    float min_perimeter_infill_spacing = float(infill_line_spacing) * 1.05f;
#if 0
// #ifdef SLIC3R_DEBUG_SLICE_PROCESSING
                    {
        				Slic3r::SVG svg_cummulative(debug_out_path("discover_vertical_shells-perimeters-before-union-run%d.svg", debug_idx), this->bounding_box());
                        for (int n = (int)idx_layer - n_extra_bottom_layers; n <= (int)idx_layer + n_extra_top_layers; ++ n) {
                            if (n < 0 || n >= (int)m_layers.size())
                                continue;
                            ExPolygons &expolys = m_layers[n]->perimeter_expolygons;
                            for (size_t i = 0; i < expolys.size(); ++ i) {
        						Slic3r::SVG svg(debug_out_path("discover_vertical_shells-perimeters-before-union-run%d-layer%d-expoly%d.svg", debug_idx, n, i), get_extents(expolys[i]));
                                svg.draw(expolys[i]);
                                svg.draw_outline(expolys[i].contour, "black", scale_(0.05));
                                svg.draw_outline(expolys[i].holes, "blue", scale_(0.05));
                                svg.Close();

                                svg_cummulative.draw(expolys[i]);
                                svg_cummulative.draw_outline(expolys[i].contour, "black", scale_(0.05));
                                svg_cummulative.draw_outline(expolys[i].holes, "blue", scale_(0.05));
                            }
                        }
                    }
#endif /* SLIC3R_DEBUG_SLICE_PROCESSING */
			        polygons_append(holes, cache_top_botom_regions[idx_layer].holes);
                    auto combine_holes = [&holes](const Polygons &holes2) {
                        if (holes.empty() || holes2.empty())
                            holes.clear();
                        else
                            holes = intersection(holes, holes2);
                    };
                    auto combine_shells = [&shell](const Polygons &shells2) {
                        if (shell.empty())
                            shell = std::move(shells2);
                        else if (! shells2.empty()) {
                            polygons_append(shell, shells2);
                            // Running the union_ using the Clipper library piece by piece is cheaper 
                            // than running the union_ all at once.
                            shell = union_(shell);
                        }
                    };
                    static constexpr const bool one_more_layer_below_top_bottom_surfaces = false;
			        if (int n_top_layers = region_config.top_solid_layers.value; n_top_layers > 0) {
                        // Gather top regions projected to this layer.
                        coordf_t print_z = layer->print_z;
                        int i = int(idx_layer) + 1;
                        int itop = int(idx_layer) + n_top_layers;
                        bool at_least_one_top_projected = false;
	                    for (; i < int(cache_top_botom_regions.size()) &&
	                         (i < itop || m_layers[i]->print_z - print_z < region_config.top_solid_min_thickness - EPSILON);
	                        ++ i) {
                            at_least_one_top_projected = true;
	                        const DiscoverVerticalShellsCacheEntry &cache = cache_top_botom_regions[i];
                            combine_holes(cache.holes);
                            combine_shells(cache.top_surfaces);
	                    }
                        if (!at_least_one_top_projected && i < int(cache_top_botom_regions.size())) {
                            // Lets consider this a special case - with only 1 top solid and minimal shell thickness settings, the
                            // boundaries of solid layers are not anchored over/under perimeters, so lets fix it by adding at least one
                            // perimeter width of area
                            Polygons anchor_area = intersection(expand(cache_top_botom_regions[idx_layer].top_surfaces,
                                                                       layerm->flow(frExternalPerimeter).scaled_spacing()),
                                                                to_polygons(m_layers[i]->lslices));
                            combine_shells(anchor_area);
                        }

                        if (one_more_layer_below_top_bottom_surfaces)
                            if (i < int(cache_top_botom_regions.size()) &&
                                (i <= itop || m_layers[i]->bottom_z() - print_z < region_config.top_solid_min_thickness - EPSILON))
                                combine_holes(cache_top_botom_regions[i].holes);
	                }
	                if (int n_bottom_layers = region_config.bottom_solid_layers.value; n_bottom_layers > 0) {
                        // Gather bottom regions projected to this layer.
                        coordf_t bottom_z = layer->bottom_z();
                        int i = int(idx_layer) - 1;
                        int ibottom = int(idx_layer) - n_bottom_layers;
                        bool at_least_one_bottom_projected = false;
	                    for (; i >= 0 &&
	                         (i > ibottom || bottom_z - m_layers[i]->bottom_z() < region_config.bottom_solid_min_thickness - EPSILON);
	                        -- i) {
                                at_least_one_bottom_projected = true;
	                        const DiscoverVerticalShellsCacheEntry &cache = cache_top_botom_regions[i];
							combine_holes(cache.holes);
                            combine_shells(cache.bottom_surfaces);
	                    }

                        if (!at_least_one_bottom_projected && i >= 0) {
                            Polygons anchor_area = intersection(expand(cache_top_botom_regions[idx_layer].bottom_surfaces,
                                                                       layerm->flow(frExternalPerimeter).scaled_spacing()),
                                                                to_polygons(m_layers[i]->lslices));
                            combine_shells(anchor_area);
                        }

                        if (one_more_layer_below_top_bottom_surfaces)
                            if (i >= 0 &&
                                (i > ibottom || bottom_z - m_layers[i]->print_z < region_config.bottom_solid_min_thickness - EPSILON))
                                combine_holes(cache_top_botom_regions[i].holes);
	                }
#ifdef SLIC3R_DEBUG_SLICE_PROCESSING
                    {
        				Slic3r::SVG svg(debug_out_path("discover_vertical_shells-perimeters-before-union-%d.svg", debug_idx), get_extents(shell));
                        svg.draw(shell);
                        svg.draw_outline(shell, "black", scale_(0.05));
                        svg.Close(); 
                    }
#endif /* SLIC3R_DEBUG_SLICE_PROCESSING */
#if 0
//                    shell = union_(shell, true);
                    shell = union_(shell, false); 
#endif
#ifdef SLIC3R_DEBUG_SLICE_PROCESSING
                    shell_ex = union_safety_offset_ex(shell);
#endif /* SLIC3R_DEBUG_SLICE_PROCESSING */

                    //if (shell.empty())
                    //    continue;

#ifdef SLIC3R_DEBUG_SLICE_PROCESSING
                    {
                        Slic3r::SVG svg(debug_out_path("discover_vertical_shells-perimeters-after-union-%d.svg", debug_idx), get_extents(shell));
                        svg.draw(shell_ex);
                        svg.draw_outline(shell_ex, "black", "blue", scale_(0.05));
                        svg.Close();  
                    }
#endif /* SLIC3R_DEBUG_SLICE_PROCESSING */

#ifdef SLIC3R_DEBUG_SLICE_PROCESSING
                    {
                        Slic3r::SVG svg(debug_out_path("discover_vertical_shells-internal-wshell-%d.svg", debug_idx), get_extents(shell));
                        svg.draw(layerm->fill_surfaces().filter_by_type(stInternal), "yellow", 0.5);
                        svg.draw_outline(layerm->fill_surfaces().filter_by_type(stInternal), "black", "blue", scale_(0.05));
                        svg.draw(shell_ex, "blue", 0.5);
                        svg.draw_outline(shell_ex, "black", "blue", scale_(0.05));
                        svg.Close();
                    } 
                    {
                        Slic3r::SVG svg(debug_out_path("discover_vertical_shells-internalvoid-wshell-%d.svg", debug_idx), get_extents(shell));
                        svg.draw(layerm->fill_surfaces().filter_by_type(stInternalVoid), "yellow", 0.5);
                        svg.draw_outline(layerm->fill_surfaces().filter_by_type(stInternalVoid), "black", "blue", scale_(0.05));
                        svg.draw(shell_ex, "blue", 0.5);
                        svg.draw_outline(shell_ex, "black", "blue", scale_(0.05));
                        svg.Close();
                    } 
                    {
                        Slic3r::SVG svg(debug_out_path("discover_vertical_shells-internalsolid-wshell-%d.svg", debug_idx), get_extents(shell));
                        svg.draw(layerm->fill_surfaces().filter_by_type(stInternalSolid), "yellow", 0.5);
                        svg.draw_outline(layerm->fill_surfaces().filter_by_type(stInternalSolid), "black", "blue", scale_(0.05));
                        svg.draw(shell_ex, "blue", 0.5);
                        svg.draw_outline(shell_ex, "black", "blue", scale_(0.05));
                        svg.Close();
                    } 
#endif /* SLIC3R_DEBUG_SLICE_PROCESSING */

                    // Trim the shells region by the internal & internal void surfaces.
                    const Polygons    polygonsInternal = to_polygons(layerm->fill_surfaces().filter_by_types({ stInternal, stInternalVoid, stInternalSolid }));
                    shell = intersection(shell, polygonsInternal, ApplySafetyOffset::Yes);
                    polygons_append(shell, diff(polygonsInternal, holes));
                    if (shell.empty())
                        continue;

                    // Append the internal solids, so they will be merged with the new ones.
                    polygons_append(shell, to_polygons(layerm->fill_surfaces().filter_by_type(stInternalSolid)));

                    // These regions will be filled by a rectilinear full infill. Currently this type of infill
                    // only fills regions, which fit at least a single line. To avoid gaps in the sparse infill,
                    // make sure that this region does not contain parts narrower than the infill spacing width.
#ifdef SLIC3R_DEBUG_SLICE_PROCESSING
                    Polygons shell_before = shell;
#endif /* SLIC3R_DEBUG_SLICE_PROCESSING */
                    ExPolygons regularized_shell;
                    {
                        // Open to remove (filter out) regions narrower than a bit less than an infill extrusion line width.
                        // Such narrow regions are difficult to fill in with a gap fill algorithm (or Arachne), however they are most likely
                        // not needed for print stability / quality.
                        //W11
                        const float narrow_ensure_vertical_wall_thickness_region_radius = 0.65f * 0.7f * min_perimeter_infill_spacing;//0.7f*0.75f may error in complex model/ original parameter 0.5f * 0.65f 
                        // Then close gaps narrower than 1.2 * line width, such gaps are difficult to fill in with sparse infill,
                        // thus they will be merged into the solid infill.
                        //W11
                        const float narrow_sparse_infill_region_radius                  = 0.7f * 1.25f * min_perimeter_infill_spacing;//0.7f*1.25f may error in complex model /original parameter 0.5f * 1.2f 
                        // Finally expand the infill a bit to remove tiny gaps between solid infill and the other regions.
                        //W11
                        const float tiny_overlap_radius                                 = 0.19f        * min_perimeter_infill_spacing;// original parameter 0.2f       
                        regularized_shell = shrink_ex(offset2_ex(union_ex(shell),
                            // Open to remove (filter out) regions narrower than an infill extrusion line width.
                            -narrow_ensure_vertical_wall_thickness_region_radius,
                            // Then close gaps narrower than 1.2 * line width, such gaps are difficult to fill in with sparse infill.
                            narrow_ensure_vertical_wall_thickness_region_radius + narrow_sparse_infill_region_radius, ClipperLib::jtSquare),
                            // Finally expand the infill a bit to remove tiny gaps between solid infill and the other regions.
                            narrow_sparse_infill_region_radius - tiny_overlap_radius, ClipperLib::jtSquare);

                        Polygons object_volume;
                        Polygons internal_volume;
                        {
                            Polygons shrinked_bottom_slice = idx_layer > 0 ? to_polygons(m_layers[idx_layer - 1]->lslices) : Polygons{};
                            Polygons shrinked_upper_slice  = (idx_layer + 1) < m_layers.size() ?
                                                                 to_polygons(m_layers[idx_layer + 1]->lslices) :
                                                                 Polygons{};
                            object_volume = intersection(shrinked_bottom_slice, shrinked_upper_slice);
                            internal_volume = closing(polygonsInternal, float(SCALED_EPSILON));
                        }

                        // The regularization operation may cause scattered tiny drops on the smooth parts of the model, filter them out
                        // If the region checks both following conditions, it is removed:
                        //   1. the area is very small,
                        //      OR the area is quite small and it is fully wrapped in model (not visible)
                        //      the in-model condition is there due to small sloping surfaces, e.g. top of the hull of the benchy
                        //   2. the area does not fully cover an internal polygon
                        //         This is there mainly for a very thin parts, where the solid layers would be missing if the part area is quite small
                        regularized_shell.erase(std::remove_if(regularized_shell.begin(), regularized_shell.end(),
                                                               [&internal_volume, &min_perimeter_infill_spacing,
                                                                &object_volume](const ExPolygon &p) {
                                                                    //W11
                                                                   return (p.area() < min_perimeter_infill_spacing * scaled(2.0) ||//original parameter scaled(1.5)
                                                                       //W11
                                                                          (p.area() < min_perimeter_infill_spacing * scaled(10.0) &&//original parameter scaled(8.0) 
                                                                            diff(to_polygons(p), object_volume).empty())) &&
                                                                          diff(internal_volume,
                                                                               expand(to_polygons(p), min_perimeter_infill_spacing))
                                                                                  .size() >= internal_volume.size();
                                                               }),
                                                regularized_shell.end());
                    }
                    if (regularized_shell.empty())
                        continue;

                    ExPolygons new_internal_solid = intersection_ex(polygonsInternal, regularized_shell);
#ifdef SLIC3R_DEBUG_SLICE_PROCESSING
                    {
                        Slic3r::SVG svg(debug_out_path("discover_vertical_shells-regularized-%d.svg", debug_idx), get_extents(shell_before));
                        // Source shell.
                        svg.draw(union_safety_offset_ex(shell_before));
                        // Shell trimmed to the internal surfaces.
                        svg.draw_outline(union_safety_offset_ex(shell), "black", "blue", scale_(0.05));
                        // Regularized infill region.
                        svg.draw_outline(new_internal_solid, "red", "magenta", scale_(0.05));
                        svg.Close();  
                    }
#endif /* SLIC3R_DEBUG_SLICE_PROCESSING */

                    // Trim the internal & internalvoid by the shell.
                    Slic3r::ExPolygons new_internal = diff_ex(layerm->fill_surfaces().filter_by_type(stInternal), regularized_shell);
                    Slic3r::ExPolygons new_internal_void = diff_ex(layerm->fill_surfaces().filter_by_type(stInternalVoid), regularized_shell);

#ifdef SLIC3R_DEBUG_SLICE_PROCESSING
                    {
                        SVG::export_expolygons(debug_out_path("discover_vertical_shells-new_internal-%d.svg", debug_idx), get_extents(shell), new_internal, "black", "blue", scale_(0.05));
        				SVG::export_expolygons(debug_out_path("discover_vertical_shells-new_internal_void-%d.svg", debug_idx), get_extents(shell), new_internal_void, "black", "blue", scale_(0.05));
        				SVG::export_expolygons(debug_out_path("discover_vertical_shells-new_internal_solid-%d.svg", debug_idx), get_extents(shell), new_internal_solid, "black", "blue", scale_(0.05));
                    }
#endif /* SLIC3R_DEBUG_SLICE_PROCESSING */

                    // Assign resulting internal surfaces to layer.
                    layerm->m_fill_surfaces.keep_types({ stTop, stBottom, stBottomBridge });
                    layerm->m_fill_surfaces.append(new_internal,       stInternal);
                    layerm->m_fill_surfaces.append(new_internal_void,  stInternalVoid);
                    layerm->m_fill_surfaces.append(new_internal_solid, stInternalSolid);
                } // for each layer
            });
        m_print->throw_if_canceled();
        BOOST_LOG_TRIVIAL(debug) << "Discovering vertical shells for region " << region_id << " in parallel - end";

#ifdef SLIC3R_DEBUG_SLICE_PROCESSING
		for (size_t idx_layer = 0; idx_layer < m_layers.size(); ++idx_layer) {
			LayerRegion *layerm = m_layers[idx_layer]->get_region(region_id);
			layerm->export_region_slices_to_svg_debug("3_discover_vertical_shells-final");
			layerm->export_region_fill_surfaces_to_svg_debug("3_discover_vertical_shells-final");
		}
#endif /* SLIC3R_DEBUG_SLICE_PROCESSING */
    } // for each region
} // void PrintObject::discover_vertical_shells()

// #define DEBUG_BRIDGE_OVER_INFILL
#ifdef DEBUG_BRIDGE_OVER_INFILL
template<typename T> void debug_draw(std::string name, const T& a, const T& b, const T& c, const T& d)
{
    std::vector<std::string> colors = {"red", "green", "blue", "orange"};
    BoundingBox              bbox   = get_extents(a);
    bbox.merge(get_extents(b));
    bbox.merge(get_extents(c));
    bbox.merge(get_extents(d));
    bbox.offset(scale_(1.));
    ::Slic3r::SVG svg(debug_out_path(name.c_str()).c_str(), bbox);   
    svg.draw(a, colors[0], scale_(0.3));
    svg.draw(b, colors[1], scale_(0.23));
    svg.draw(c, colors[2], scale_(0.16));
    svg.draw(d, colors[3], scale_(0.10));
    svg.Close();
}
#endif

// This method applies bridge flow to the first internal solid layer above sparse infill.
void PrintObject::bridge_over_infill()
{
    BOOST_LOG_TRIVIAL(info) << "Bridge over infill - Start" << log_memory_info();

    struct CandidateSurface
    {
        CandidateSurface(const Surface     *original_surface,
                         int                layer_index,
                         Polygons           new_polys,
                         const LayerRegion *region,
                         double             bridge_angle)
            : original_surface(original_surface)
            , layer_index(layer_index)
            , new_polys(new_polys)
            , region(region)
            , bridge_angle(bridge_angle)
        {}
        const Surface     *original_surface;
        int                layer_index;
        Polygons           new_polys;
        const LayerRegion *region;
        double             bridge_angle;
    };

    std::map<size_t, std::vector<CandidateSurface>> surfaces_by_layer;

    // SECTION to gather and filter surfaces for expanding, and then cluster them by layer
    {
        tbb::concurrent_vector<CandidateSurface> candidate_surfaces;
        tbb::parallel_for(tbb::blocked_range<size_t>(0, this->layers().size()), [po = static_cast<const PrintObject *>(this),
                                                                                 &candidate_surfaces](tbb::blocked_range<size_t> r) {
            PRINT_OBJECT_TIME_LIMIT_MILLIS(PRINT_OBJECT_TIME_LIMIT_DEFAULT);
            for (size_t lidx = r.begin(); lidx < r.end(); lidx++) {
                const Layer *layer = po->get_layer(lidx);
                if (layer->lower_layer == nullptr) {
                    continue;
                }
                double spacing = layer->regions().front()->flow(frSolidInfill).scaled_spacing();
                // unsupported area will serve as a filter for polygons worth bridging.
                Polygons   unsupported_area;
                Polygons   lower_layer_solids;
                for (const LayerRegion *region : layer->lower_layer->regions()) {
                    Polygons fill_polys = to_polygons(region->fill_expolygons());
                    // initially consider the whole layer unsupported, but also gather solid layers to later cut off supported parts
                    unsupported_area.insert(unsupported_area.end(), fill_polys.begin(), fill_polys.end());
                    for (const Surface &surface : region->fill_surfaces()) {
                        if (surface.surface_type != stInternal || region->region().config().fill_density.value == 100) {
                            Polygons p = to_polygons(surface.expolygon);
                            lower_layer_solids.insert(lower_layer_solids.end(), p.begin(), p.end());
                        }
                    }
                }
                unsupported_area = closing(unsupported_area, float(SCALED_EPSILON));
                // By expanding the lower layer solids, we avoid making bridges from the tiny internal overhangs that are (very likely) supported by previous layer solids
                // NOTE that we cannot filter out polygons worth bridging by their area, because sometimes there is a very small internal island that will grow into large hole
                lower_layer_solids = shrink(lower_layer_solids, 1 * spacing); // first remove thin regions that will not support anything
                lower_layer_solids = expand(lower_layer_solids, (1 + 3) * spacing); // then expand back (opening), and further for parts supported by internal solids
                // By shrinking the unsupported area, we avoid making bridges from narrow ensuring region along perimeters.
                unsupported_area   = shrink(unsupported_area, 3 * spacing);
                unsupported_area   = diff(unsupported_area, lower_layer_solids);

                for (const LayerRegion *region : layer->regions()) {
                    SurfacesPtr region_internal_solids = region->fill_surfaces().filter_by_type(stInternalSolid);
                    for (const Surface *s : region_internal_solids) {
                        Polygons unsupported         = intersection(to_polygons(s->expolygon), unsupported_area);
                        // The following flag marks those surfaces, which overlap with unuspported area, but at least part of them is supported. 
                        // These regions can be filtered by area, because they for sure are touching solids on lower layers, and it does not make sense to bridge their tiny overhangs 
                        bool     partially_supported = area(unsupported) < area(to_polygons(s->expolygon)) - EPSILON;
                        if (!unsupported.empty() && (!partially_supported || area(unsupported) > 3 * 3 * spacing * spacing)) {
                            Polygons worth_bridging = intersection(to_polygons(s->expolygon), expand(unsupported, 4 * spacing));
                            // after we extracted the part worth briding, we go over the leftovers and merge the tiny ones back, to not brake the surface too much
                            for (const Polygon& p : diff(to_polygons(s->expolygon), expand(worth_bridging, spacing))) {
                                double area = p.area();
                                if (area < spacing * scale_(12.0) && area > spacing * spacing) {
                                    worth_bridging.push_back(p);
                                }
                            }
                            worth_bridging = intersection(closing(worth_bridging, float(SCALED_EPSILON)), s->expolygon);
                            candidate_surfaces.push_back(CandidateSurface(s, lidx, worth_bridging, region, 0));

#ifdef DEBUG_BRIDGE_OVER_INFILL
                            debug_draw(std::to_string(lidx) + "_candidate_surface_" + std::to_string(area(s->expolygon)),
                                       to_lines(region->layer()->lslices), to_lines(s->expolygon), to_lines(worth_bridging),
                                       to_lines(unsupported_area));
#endif
#ifdef DEBUG_BRIDGE_OVER_INFILL
                            debug_draw(std::to_string(lidx) + "_candidate_processing_" + std::to_string(area(unsupported)),
                                       to_lines(unsupported), to_lines(intersection(to_polygons(s->expolygon), expand(unsupported, 5 * spacing))), 
                                       to_lines(diff(to_polygons(s->expolygon), expand(worth_bridging, spacing))),
                                       to_lines(unsupported_area));
#endif
                        }
                    }
                }
            }
        });

        for (const CandidateSurface &c : candidate_surfaces) {
            surfaces_by_layer[c.layer_index].push_back(c);
        }
    }

    // LIGHTNING INFILL SECTION - If lightning infill is used somewhere, we check the areas that are going to be bridges, and those that rely on the 
    // lightning infill under them get expanded. This somewhat helps to ensure that most of the extrusions are anchored to the lightning infill at the ends.
    // It requires modifying this instance of print object in a specific way, so that we do not invalidate the pointers in our surfaces_by_layer structure.
    bool has_lightning_infill = false;
    for (size_t i = 0; i < this->num_printing_regions(); i++) {
        if (this->printing_region(i).config().fill_pattern == ipLightning) {
            has_lightning_infill = true;
            break;
        }
    }
    if (has_lightning_infill) {
        // Prepare backup data for the Layer Region infills. Before modfiyng the layer region, we backup its fill surfaces by moving! them into this map.
        // then a copy is created, modifiyed and passed to lightning infill generator. After generator is created, we restore the original state of the fills
        // again by moving the data from this map back to the layer regions. This ensures that pointers to surfaces stay valid.
        std::map<size_t, std::map<const LayerRegion *, SurfaceCollection>> backup_surfaces;
        for (size_t lidx = 0; lidx < this->layer_count(); lidx++) {
            backup_surfaces[lidx] = {};
        }

        tbb::parallel_for(tbb::blocked_range<size_t>(0, this->layers().size()), [po = this, &backup_surfaces,
                                                                                 &surfaces_by_layer](tbb::blocked_range<size_t> r) {
            PRINT_OBJECT_TIME_LIMIT_MILLIS(PRINT_OBJECT_TIME_LIMIT_DEFAULT);
            for (size_t lidx = r.begin(); lidx < r.end(); lidx++) {
                if (surfaces_by_layer.find(lidx) == surfaces_by_layer.end())
                    continue;

                Layer       *layer       = po->get_layer(lidx);
                const Layer *lower_layer = layer->lower_layer;
                if (lower_layer == nullptr)
                    continue;

                Polygons lightning_fill;
                for (const LayerRegion *region : lower_layer->regions()) {
                    if (region->region().config().fill_pattern == ipLightning) {
                        Polygons lf = to_polygons(region->fill_surfaces().filter_by_type(stInternal));
                        lightning_fill.insert(lightning_fill.end(), lf.begin(), lf.end());
                    }
                }

                if (lightning_fill.empty())
                    continue;

                for (LayerRegion *region : layer->regions()) {
                    backup_surfaces[lidx][region] = std::move(
                        region->m_fill_surfaces); // Make backup copy by move!! so that pointers in candidate surfaces stay valid
                    // Copy the surfaces back, this will make copy, but we will later discard it anyway
                    region->m_fill_surfaces = backup_surfaces[lidx][region];
                }

                for (LayerRegion *region : layer->regions()) {
                    ExPolygons sparse_infill = to_expolygons(region->fill_surfaces().filter_by_type(stInternal));
                    ExPolygons solid_infill  = to_expolygons(region->fill_surfaces().filter_by_type(stInternalSolid));

                    if (sparse_infill.empty()) {
                        break;
                    }
                    for (const auto &surface : surfaces_by_layer[lidx]) {
                        if (surface.region != region)
                            continue;
                        ExPolygons expansion = intersection_ex(sparse_infill, expand(surface.new_polys, scaled<float>(3.0)));
                        solid_infill.insert(solid_infill.end(), expansion.begin(), expansion.end());
                    }

                    solid_infill  = union_safety_offset_ex(solid_infill);
                    sparse_infill = diff_ex(sparse_infill, solid_infill);

                    region->m_fill_surfaces.remove_types({stInternalSolid, stInternal});
                    for (const ExPolygon &ep : solid_infill) {
                        region->m_fill_surfaces.surfaces.emplace_back(stInternalSolid, ep);
                    }
                    for (const ExPolygon &ep : sparse_infill) {
                        region->m_fill_surfaces.surfaces.emplace_back(stInternal, ep);
                    }
                }
            }
        });

        // Use the modified surfaces to generate expanded lightning anchors
        this->m_lightning_generator = this->prepare_lightning_infill_data();

        // And now restore carefully the original surfaces, again using move to avoid reallocation and preserving the validity of the
        // pointers in surface candidates
        for (size_t lidx = 0; lidx < this->layer_count(); lidx++) {
            Layer *layer = this->get_layer(lidx);
            for (LayerRegion *region : layer->regions()) {
                if (backup_surfaces[lidx].find(region) != backup_surfaces[lidx].end()) {
                    region->m_fill_surfaces = std::move(backup_surfaces[lidx][region]);
                }
            }
        }
    }

    std::map<size_t, Polylines> infill_lines;
    // SECTION to generate infill polylines
    {
        std::vector<std::pair<const Surface *, float>> surfaces_w_bottom_z;
        for (const auto &pair : surfaces_by_layer) {
            for (const CandidateSurface &c : pair.second) {
                surfaces_w_bottom_z.emplace_back(c.original_surface, c.region->m_layer->bottom_z());
            }
        }

        this->m_adaptive_fill_octrees = this->prepare_adaptive_infill_data(surfaces_w_bottom_z);

        std::vector<size_t> layers_to_generate_infill;
        for (const auto &pair : surfaces_by_layer) {
            assert(pair.first > 0);
            infill_lines[pair.first - 1] = {};
            layers_to_generate_infill.push_back(pair.first - 1);
        }

        tbb::parallel_for(tbb::blocked_range<size_t>(0, layers_to_generate_infill.size()), [po = static_cast<const PrintObject *>(this),
                                                                                            &layers_to_generate_infill,
                                                                                            &infill_lines](tbb::blocked_range<size_t> r) {
            PRINT_OBJECT_TIME_LIMIT_MILLIS(PRINT_OBJECT_TIME_LIMIT_DEFAULT);
            for (size_t job_idx = r.begin(); job_idx < r.end(); job_idx++) {
                size_t lidx = layers_to_generate_infill[job_idx];
                infill_lines.at(
                    lidx) = po->get_layer(lidx)->generate_sparse_infill_polylines_for_anchoring(po->m_adaptive_fill_octrees.first.get(),
                                                                                                po->m_adaptive_fill_octrees.second.get(),
                                                                                                po->m_lightning_generator.get());
            }
        });
#ifdef DEBUG_BRIDGE_OVER_INFILL
        for (const auto &il : infill_lines) {
            debug_draw(std::to_string(il.first) + "_infill_lines", to_lines(get_layer(il.first)->lslices), to_lines(il.second), {}, {});
        }
#endif
    }

    // cluster layers by depth needed for thick bridges. Each cluster is to be processed by single thread sequentially, so that bridges cannot appear one on another
    std::vector<std::vector<size_t>> clustered_layers_for_threads;
    float target_flow_height_factor = 0.9f;
    {
        std::vector<size_t> layers_with_candidates;
        std::map<size_t, Polygons> layer_area_covered_by_candidates;
        for (const auto& pair : surfaces_by_layer) {
            layers_with_candidates.push_back(pair.first);
            layer_area_covered_by_candidates[pair.first] = {};
        }

        // prepare inflated filter for each candidate on each layer. layers will be put into single thread cluster if they are close to each other (z-axis-wise)
        // and if the inflated AABB polygons overlap somewhere
        tbb::parallel_for(tbb::blocked_range<size_t>(0, layers_with_candidates.size()), [&layers_with_candidates, &surfaces_by_layer,
                                                                                         &layer_area_covered_by_candidates](
                                                                                            tbb::blocked_range<size_t> r) {
            PRINT_OBJECT_TIME_LIMIT_MILLIS(PRINT_OBJECT_TIME_LIMIT_DEFAULT);
            for (size_t job_idx = r.begin(); job_idx < r.end(); job_idx++) {
                size_t lidx = layers_with_candidates[job_idx];
                for (const auto &candidate : surfaces_by_layer.at(lidx)) {
                    Polygon candiate_inflated_aabb = get_extents(candidate.new_polys).inflated(scale_(7)).polygon();
                    layer_area_covered_by_candidates.at(lidx) = union_(layer_area_covered_by_candidates.at(lidx),
                                                                       Polygons{candiate_inflated_aabb});
                }
            }
        });

        // note: surfaces_by_layer is ordered map
        for (auto pair : surfaces_by_layer) {
            if (clustered_layers_for_threads.empty() ||
                this->get_layer(clustered_layers_for_threads.back().back())->print_z <
                    this->get_layer(pair.first)->print_z -
                        this->get_layer(pair.first)->regions()[0]->bridging_flow(frSolidInfill, true).height() * target_flow_height_factor -
                        EPSILON ||
                intersection(layer_area_covered_by_candidates[clustered_layers_for_threads.back().back()],
                             layer_area_covered_by_candidates[pair.first])
                    .empty()) {
                clustered_layers_for_threads.push_back({pair.first});
            } else {
                clustered_layers_for_threads.back().push_back(pair.first);
            }
        }

#ifdef DEBUG_BRIDGE_OVER_INFILL
        std::cout << "BRIDGE OVER INFILL CLUSTERED LAYERS FOR SINGLE THREAD" << std::endl;
        for (auto cluster : clustered_layers_for_threads) {
            std::cout << "CLUSTER: ";
            for (auto l : cluster) {
                std::cout << l << "  ";
            }
            std::cout << std::endl;
        }
#endif
    }

    // LAMBDA to gather areas with sparse infill deep enough that we can fit thick bridges there.
    auto gather_areas_w_depth = [target_flow_height_factor](const PrintObject *po, int lidx, float target_flow_height) {
        // Gather layers sparse infill areas, to depth defined by used bridge flow
        ExPolygons layers_sparse_infill{};
        ExPolygons not_sparse_infill{};
        double   bottom_z = po->get_layer(lidx)->print_z - target_flow_height * target_flow_height_factor - EPSILON;
        for (int i = int(lidx) - 1; i >= 0; --i) {
            // Stop iterating if layer is lower than bottom_z and at least one iteration was made
            const Layer *layer = po->get_layer(i);
            if (layer->print_z < bottom_z && i < int(lidx) - 1)
                break;

            for (const LayerRegion *region : layer->regions()) {
                bool has_low_density = region->region().config().fill_density.value < 100;
                for (const Surface &surface : region->fill_surfaces()) {
                    if ((surface.surface_type == stInternal && has_low_density) || surface.surface_type == stInternalVoid ) {
                        layers_sparse_infill.push_back(surface.expolygon);
                    } else {
                        not_sparse_infill.push_back(surface.expolygon);
                    }
                }
            }
        }
        layers_sparse_infill = union_ex(layers_sparse_infill);
        layers_sparse_infill = closing_ex(layers_sparse_infill, float(SCALED_EPSILON));
        not_sparse_infill    = union_ex(not_sparse_infill);
        not_sparse_infill    = closing_ex(not_sparse_infill, float(SCALED_EPSILON));
        return diff(layers_sparse_infill, not_sparse_infill);
    };

    // LAMBDA do determine optimal bridging angle
    auto determine_bridging_angle = [](const Polygons &bridged_area, const Lines &anchors, InfillPattern dominant_pattern) {
        AABBTreeLines::LinesDistancer<Line> lines_tree(anchors);

        std::map<double, int> counted_directions;
        for (const Polygon &p : bridged_area) {
            double acc_distance = 0;
            for (int point_idx = 0; point_idx < int(p.points.size()) - 1; ++point_idx) {
                Vec2d  start        = p.points[point_idx].cast<double>();
                Vec2d  next         = p.points[point_idx + 1].cast<double>();
                Vec2d  v            = next - start; // vector from next to current
                double dist_to_next = v.norm();
                acc_distance += dist_to_next;
                if (acc_distance > scaled(2.0)) {
                    acc_distance = 0.0;
                    v.normalize();
                    int   lines_count = int(std::ceil(dist_to_next / scaled(2.0)));
                    float step_size   = dist_to_next / lines_count;
                    for (int i = 0; i < lines_count; ++i) {
                        Point a                   = (start + v * (i * step_size)).cast<coord_t>();
                        auto [distance, index, p] = lines_tree.distance_from_lines_extra<false>(a);
                        double angle = lines_tree.get_line(index).orientation();
                        if (angle > PI) {
                            angle -= PI;
                        }
                        angle += PI * 0.5;
                        counted_directions[angle]++;
                    }
                }
            }
        }

        std::pair<double, int> best_dir{0, 0};
        // sliding window accumulation
        for (const auto &dir : counted_directions) {
            int    score_acc          = 0;
            double dir_acc            = 0;
            double window_start_angle = dir.first - PI * 0.1;
            double window_end_angle   = dir.first + PI * 0.1;
            for (auto dirs_window = counted_directions.lower_bound(window_start_angle);
                 dirs_window != counted_directions.upper_bound(window_end_angle); dirs_window++) {
                dir_acc += dirs_window->first * dirs_window->second;
                score_acc += dirs_window->second;
            }
            // current span of directions is 0.5 PI to 1.5 PI (due to the aproach.). Edge values should also account for the
            //  opposite direction.
            if (window_start_angle < 0.5 * PI) {
                for (auto dirs_window = counted_directions.lower_bound(1.5 * PI - (0.5 * PI - window_start_angle));
                     dirs_window != counted_directions.end(); dirs_window++) {
                    dir_acc += dirs_window->first * dirs_window->second;
                    score_acc += dirs_window->second;
                }
            }
            if (window_start_angle > 1.5 * PI) {
                for (auto dirs_window = counted_directions.begin();
                     dirs_window != counted_directions.upper_bound(window_start_angle - 1.5 * PI); dirs_window++) {
                    dir_acc += dirs_window->first * dirs_window->second;
                    score_acc += dirs_window->second;
                }
            }

            if (score_acc > best_dir.second) {
                best_dir = {dir_acc / score_acc, score_acc};
            }
        }
        double bridging_angle = best_dir.first;
        if (bridging_angle == 0) {
            bridging_angle = 0.001;
        }
        switch (dominant_pattern) {
        case ipHilbertCurve: bridging_angle += 0.25 * PI; break;
        case ipOctagramSpiral: bridging_angle += (1.0 / 16.0) * PI; break;
        default: break;
        }

        return bridging_angle;
    };

    // LAMBDA that will fill given polygons with lines, exapand the lines to the nearest anchor, and reconstruct polygons from the newly
    // generated lines
    auto construct_anchored_polygon = [](Polygons bridged_area, Lines anchors, const Flow &bridging_flow, double bridging_angle) {
        auto lines_rotate = [](Lines &lines, double cos_angle, double sin_angle) {
            for (Line &l : lines) {
                double ax = double(l.a.x());
                double ay = double(l.a.y());
                l.a.x()   = coord_t(round(cos_angle * ax - sin_angle * ay));
                l.a.y()   = coord_t(round(cos_angle * ay + sin_angle * ax));
                double bx = double(l.b.x());
                double by = double(l.b.y());
                l.b.x()   = coord_t(round(cos_angle * bx - sin_angle * by));
                l.b.y()   = coord_t(round(cos_angle * by + sin_angle * bx));
            }
        };

        auto segments_overlap = [](coord_t alow, coord_t ahigh, coord_t blow, coord_t bhigh) {
            return (alow >= blow && alow <= bhigh) || (ahigh >= blow && ahigh <= bhigh) || (blow >= alow && blow <= ahigh) ||
                   (bhigh >= alow && bhigh <= ahigh);
        };

        Polygons expanded_bridged_area{};
        double   aligning_angle = -bridging_angle + PI * 0.5;
        {
            polygons_rotate(bridged_area, aligning_angle);
            lines_rotate(anchors, cos(aligning_angle), sin(aligning_angle));
            BoundingBox bb_x = get_extents(bridged_area);
            BoundingBox bb_y = get_extents(anchors);

            const size_t n_vlines = (bb_x.max.x() - bb_x.min.x() + bridging_flow.scaled_spacing() - 1) / bridging_flow.scaled_spacing();
            std::vector<Line> vertical_lines(n_vlines);
            for (size_t i = 0; i < n_vlines; i++) {
                coord_t x           = bb_x.min.x() + i * bridging_flow.scaled_spacing();
                coord_t y_min       = bb_y.min.y() - bridging_flow.scaled_spacing();
                coord_t y_max       = bb_y.max.y() + bridging_flow.scaled_spacing();
                vertical_lines[i].a = Point{x, y_min};
                vertical_lines[i].b = Point{x, y_max};
            }

            auto anchors_and_walls_tree = AABBTreeLines::LinesDistancer<Line>{std::move(anchors)};
            auto bridged_area_tree      = AABBTreeLines::LinesDistancer<Line>{to_lines(bridged_area)};

            std::vector<std::vector<Line>> polygon_sections(n_vlines);
            for (size_t i = 0; i < n_vlines; i++) {
                auto area_intersections = bridged_area_tree.intersections_with_line<true>(vertical_lines[i]);
                for (int intersection_idx = 0; intersection_idx < int(area_intersections.size()) - 1; intersection_idx++) {
                    if (bridged_area_tree.outside(
                            (area_intersections[intersection_idx].first + area_intersections[intersection_idx + 1].first) / 2) < 0) {
                        polygon_sections[i].emplace_back(area_intersections[intersection_idx].first,
                                                         area_intersections[intersection_idx + 1].first);
                    }
                }
                auto anchors_intersections = anchors_and_walls_tree.intersections_with_line<true>(vertical_lines[i]);

                for (Line &section : polygon_sections[i]) {
                    auto maybe_below_anchor = std::upper_bound(anchors_intersections.rbegin(), anchors_intersections.rend(), section.a,
                                                               [](const Point &a, const std::pair<Point, size_t> &b) {
                                                                   return a.y() > b.first.y();
                                                               });
                    if (maybe_below_anchor != anchors_intersections.rend()) {
                        section.a = maybe_below_anchor->first;
                        section.a.y() -= bridging_flow.scaled_width() * (0.5 + 0.5);
                    }

                    auto maybe_upper_anchor = std::upper_bound(anchors_intersections.begin(), anchors_intersections.end(), section.b,
                                                               [](const Point &a, const std::pair<Point, size_t> &b) {
                                                                   return a.y() < b.first.y();
                                                               });
                    if (maybe_upper_anchor != anchors_intersections.end()) {
                        section.b = maybe_upper_anchor->first;
                        section.b.y() += bridging_flow.scaled_width() * (0.5 + 0.5);
                    }
                }

                for (int section_idx = 0; section_idx < int(polygon_sections[i].size()) - 1; section_idx++) {
                    Line &section_a = polygon_sections[i][section_idx];
                    Line &section_b = polygon_sections[i][section_idx + 1];
                    if (segments_overlap(section_a.a.y(), section_a.b.y(), section_b.a.y(), section_b.b.y())) {
                        section_b.a = section_a.a.y() < section_b.a.y() ? section_a.a : section_b.a;
                        section_b.b = section_a.b.y() < section_b.b.y() ? section_b.b : section_a.b;
                        section_a.a = section_a.b;
                    }
                }

                polygon_sections[i].erase(std::remove_if(polygon_sections[i].begin(), polygon_sections[i].end(),
                                                         [](const Line &s) { return s.a == s.b; }),
                                          polygon_sections[i].end());
                std::sort(polygon_sections[i].begin(), polygon_sections[i].end(),
                          [](const Line &a, const Line &b) { return a.a.y() < b.b.y(); });
            }

            // reconstruct polygon from polygon sections
            struct TracedPoly
            {
                Points lows;
                Points highs;
            };

            std::vector<TracedPoly> current_traced_polys;
            for (const auto &polygon_slice : polygon_sections) {
                std::unordered_set<const Line *> used_segments;
                for (TracedPoly &traced_poly : current_traced_polys) {
                    auto candidates_begin = std::upper_bound(polygon_slice.begin(), polygon_slice.end(), traced_poly.lows.back(),
                                                             [](const Point &low, const Line &seg) { return seg.b.y() > low.y(); });
                    auto candidates_end   = std::upper_bound(polygon_slice.begin(), polygon_slice.end(), traced_poly.highs.back(),
                                                             [](const Point &high, const Line &seg) { return seg.a.y() > high.y(); });

                    bool segment_added = false;
                    for (auto candidate = candidates_begin; candidate != candidates_end && !segment_added; candidate++) {
                        if (used_segments.find(&(*candidate)) != used_segments.end()) {
                            continue;
                        }

                        if ((traced_poly.lows.back() - candidate->a).cast<double>().squaredNorm() <
                            36.0 * double(bridging_flow.scaled_spacing()) * bridging_flow.scaled_spacing()) {
                            traced_poly.lows.push_back(candidate->a);
                        } else {
                            traced_poly.lows.push_back(traced_poly.lows.back() + Point{bridging_flow.scaled_spacing() / 2, 0});
                            traced_poly.lows.push_back(candidate->a - Point{bridging_flow.scaled_spacing() / 2, 0});
                            traced_poly.lows.push_back(candidate->a);
                        }

                        if ((traced_poly.highs.back() - candidate->b).cast<double>().squaredNorm() <
                            36.0 * double(bridging_flow.scaled_spacing()) * bridging_flow.scaled_spacing()) {
                            traced_poly.highs.push_back(candidate->b);
                        } else {
                            traced_poly.highs.push_back(traced_poly.highs.back() + Point{bridging_flow.scaled_spacing() / 2, 0});
                            traced_poly.highs.push_back(candidate->b - Point{bridging_flow.scaled_spacing() / 2, 0});
                            traced_poly.highs.push_back(candidate->b);
                        }
                        segment_added = true;
                        used_segments.insert(&(*candidate));
                    }

                    if (!segment_added) {
                        // Zero overlapping segments, we just close this polygon
                        traced_poly.lows.push_back(traced_poly.lows.back() + Point{bridging_flow.scaled_spacing() / 2, 0});
                        traced_poly.highs.push_back(traced_poly.highs.back() + Point{bridging_flow.scaled_spacing() / 2, 0});
                        Polygon &new_poly = expanded_bridged_area.emplace_back(std::move(traced_poly.lows));
                        new_poly.points.insert(new_poly.points.end(), traced_poly.highs.rbegin(), traced_poly.highs.rend());
                        traced_poly.lows.clear();
                        traced_poly.highs.clear();
                    }
                }

                current_traced_polys.erase(std::remove_if(current_traced_polys.begin(), current_traced_polys.end(),
                                                          [](const TracedPoly &tp) { return tp.lows.empty(); }),
                                           current_traced_polys.end());

                for (const auto &segment : polygon_slice) {
                    if (used_segments.find(&segment) == used_segments.end()) {
                        TracedPoly &new_tp = current_traced_polys.emplace_back();
                        new_tp.lows.push_back(segment.a - Point{bridging_flow.scaled_spacing() / 2, 0});
                        new_tp.lows.push_back(segment.a);
                        new_tp.highs.push_back(segment.b - Point{bridging_flow.scaled_spacing() / 2, 0});
                        new_tp.highs.push_back(segment.b);
                    }
                }
            }

            // add not closed polys
            for (TracedPoly &traced_poly : current_traced_polys) {
                Polygon &new_poly = expanded_bridged_area.emplace_back(std::move(traced_poly.lows));
                new_poly.points.insert(new_poly.points.end(), traced_poly.highs.rbegin(), traced_poly.highs.rend());
            }
            expanded_bridged_area = union_safety_offset(expanded_bridged_area);
        }

        polygons_rotate(expanded_bridged_area, -aligning_angle);
        return expanded_bridged_area;
    };

    tbb::parallel_for(tbb::blocked_range<size_t>(0, clustered_layers_for_threads.size()), [po = static_cast<const PrintObject *>(this),
                                                                                           target_flow_height_factor, &surfaces_by_layer,
                                                                                           &clustered_layers_for_threads,
                                                                                           gather_areas_w_depth, &infill_lines,
                                                                                           determine_bridging_angle,
                                                                                           construct_anchored_polygon](
                                                                                              tbb::blocked_range<size_t> r) {
        PRINT_OBJECT_TIME_LIMIT_MILLIS(PRINT_OBJECT_TIME_LIMIT_DEFAULT);
        for (size_t cluster_idx = r.begin(); cluster_idx < r.end(); cluster_idx++) {
            for (size_t job_idx = 0; job_idx < clustered_layers_for_threads[cluster_idx].size(); job_idx++) {
                size_t       lidx  = clustered_layers_for_threads[cluster_idx][job_idx];
                const Layer *layer = po->get_layer(lidx);
                // this thread has exclusive access to all surfaces in layers enumerated in
                // clustered_layers_for_threads[cluster_idx]

                // Presort the candidate polygons. This will help choose the same angle for neighbournig surfaces, that
                // would otherwise compete over anchoring sparse infill lines, leaving one area unachored
                std::sort(surfaces_by_layer[lidx].begin(), surfaces_by_layer[lidx].end(),
                          [](const CandidateSurface &left, const CandidateSurface &right) {
                              auto a = get_extents(left.new_polys);
                              auto b = get_extents(right.new_polys);

                              if (a.min.x() == b.min.x()) {
                                  return a.min.y() < b.min.y();
                              };
                              return a.min.x() < b.min.x();
                          });
                if (surfaces_by_layer[lidx].size() > 2) {
                    Vec2d origin = get_extents(surfaces_by_layer[lidx].front().new_polys).max.cast<double>();
                    std::stable_sort(surfaces_by_layer[lidx].begin() + 1, surfaces_by_layer[lidx].end(),
                                     [origin](const CandidateSurface &left, const CandidateSurface &right) {
                                         auto a = get_extents(left.new_polys);
                                         auto b = get_extents(right.new_polys);

                                         return (origin - a.min.cast<double>()).squaredNorm() <
                                                (origin - b.min.cast<double>()).squaredNorm();
                                     });
                }

                // Gather deep infill areas, where thick bridges fit
                coordf_t spacing            = surfaces_by_layer[lidx].front().region->bridging_flow(frSolidInfill, true).scaled_spacing();
                coordf_t target_flow_height = surfaces_by_layer[lidx].front().region->bridging_flow(frSolidInfill, true).height() *
                                              target_flow_height_factor;
                Polygons deep_infill_area = gather_areas_w_depth(po, lidx, target_flow_height);

                {
                    // Now also remove area that has been already filled on lower layers by bridging expansion - For this
                    // reason we did the clustering of layers per thread.
                    Polygons filled_polyons_on_lower_layers;
                    double   bottom_z = layer->print_z - target_flow_height - EPSILON;
                    if (job_idx > 0) {
                        for (int lower_job_idx = job_idx - 1; lower_job_idx >= 0; lower_job_idx--) {
                            size_t       lower_layer_idx = clustered_layers_for_threads[cluster_idx][lower_job_idx];
                            const Layer *lower_layer     = po->get_layer(lower_layer_idx);
                            if (lower_layer->print_z >= bottom_z) {
                                for (const auto &c : surfaces_by_layer[lower_layer_idx]) {
                                    filled_polyons_on_lower_layers.insert(filled_polyons_on_lower_layers.end(), c.new_polys.begin(),
                                                                          c.new_polys.end());
                                }
                            } else {
                                break;
                            }
                        }
                    }
                    deep_infill_area = diff(deep_infill_area, filled_polyons_on_lower_layers);
                }

                deep_infill_area = expand(deep_infill_area, spacing * 1.5);

                // Now gather expansion polygons - internal infill on current layer, from which we can cut off anchors
                Polygons lightning_area;
                Polygons expansion_area;
                Polygons total_fill_area;
                for (const LayerRegion *region : layer->regions()) {
                    Polygons internal_polys = to_polygons(region->fill_surfaces().filter_by_types({stInternal, stInternalSolid}));
                    expansion_area.insert(expansion_area.end(), internal_polys.begin(), internal_polys.end());
                    Polygons fill_polys = to_polygons(region->fill_expolygons());
                    total_fill_area.insert(total_fill_area.end(), fill_polys.begin(), fill_polys.end());
                    if (region->region().config().fill_pattern == ipLightning) {
                        Polygons l = to_polygons(region->fill_surfaces().filter_by_type(stInternal));
                        lightning_area.insert(lightning_area.end(), l.begin(), l.end());
                    }
                }
                total_fill_area   = closing(total_fill_area, float(SCALED_EPSILON));
                expansion_area    = closing(expansion_area, float(SCALED_EPSILON));
                expansion_area    = intersection(expansion_area, deep_infill_area);
                Polylines anchors = intersection_pl(infill_lines[lidx - 1], shrink(expansion_area, spacing));
                Polygons internal_unsupported_area = shrink(deep_infill_area, spacing * 4.5);

#ifdef DEBUG_BRIDGE_OVER_INFILL
                debug_draw(std::to_string(lidx) + "_" + std::to_string(cluster_idx) + "_" + std::to_string(job_idx) + "_" + "_total_area",
                           to_lines(total_fill_area), to_lines(expansion_area), to_lines(deep_infill_area), to_lines(anchors));
#endif

                std::vector<CandidateSurface> expanded_surfaces;
                expanded_surfaces.reserve(surfaces_by_layer[lidx].size());
                for (const CandidateSurface &candidate : surfaces_by_layer[lidx]) {
                    const Flow &flow              = candidate.region->bridging_flow(frSolidInfill, true);
                    Polygons    area_to_be_bridge = expand(candidate.new_polys, flow.scaled_spacing());
                    area_to_be_bridge             = intersection(area_to_be_bridge, deep_infill_area);

                    area_to_be_bridge.erase(std::remove_if(area_to_be_bridge.begin(), area_to_be_bridge.end(),
                                                           [internal_unsupported_area](const Polygon &p) {
                                                               return intersection({p}, internal_unsupported_area).empty();
                                                           }),
                                            area_to_be_bridge.end());

                    Polygons limiting_area = union_(area_to_be_bridge, expansion_area);

                    if (area_to_be_bridge.empty())
                        continue;

                    Polylines boundary_plines = to_polylines(expand(total_fill_area, 1.3 * flow.scaled_spacing()));
                    {
                        Polylines limiting_plines = to_polylines(expand(limiting_area, 0.3*flow.spacing()));
                        boundary_plines.insert(boundary_plines.end(), limiting_plines.begin(), limiting_plines.end());
                    }

#ifdef DEBUG_BRIDGE_OVER_INFILL
                    int r = rand();
                    debug_draw(std::to_string(lidx) + "_" + std::to_string(cluster_idx) + "_" + std::to_string(job_idx) + "_" +
                                   "_anchors_" + std::to_string(r),
                               to_lines(area_to_be_bridge), to_lines(boundary_plines), to_lines(anchors), to_lines(expansion_area));
#endif

                    double bridging_angle = 0;
                    if (!anchors.empty()) {
                        bridging_angle = determine_bridging_angle(area_to_be_bridge, to_lines(anchors),
                                                                  candidate.region->region().config().fill_pattern.value);
                    } else {
                        // use expansion boundaries as anchors.
                        // Also, use Infill pattern that is neutral for angle determination, since there are no infill lines.
                        bridging_angle = determine_bridging_angle(area_to_be_bridge, to_lines(boundary_plines), InfillPattern::ipLine);
                    }

                    boundary_plines.insert(boundary_plines.end(), anchors.begin(), anchors.end());
                    if (!lightning_area.empty() && !intersection(area_to_be_bridge, lightning_area).empty()) {
                        boundary_plines = intersection_pl(boundary_plines, expand(area_to_be_bridge, scale_(10)));
                    }
                    Polygons bridging_area = construct_anchored_polygon(area_to_be_bridge, to_lines(boundary_plines), flow, bridging_angle);

                    // Check collision with other expanded surfaces
                    {
                        bool     reconstruct       = false;
                        Polygons tmp_expanded_area = expand(bridging_area, 3.0 * flow.scaled_spacing());
                        for (const CandidateSurface &s : expanded_surfaces) {
                            if (!intersection(s.new_polys, tmp_expanded_area).empty()) {
                                bridging_angle = s.bridge_angle;
                                reconstruct    = true;
                                break;
                            }
                        }
                        if (reconstruct) {
                            bridging_area = construct_anchored_polygon(area_to_be_bridge, to_lines(boundary_plines), flow, bridging_angle);
                        }
                    }

                    bridging_area          = opening(bridging_area, flow.scaled_spacing());
                    bridging_area          = closing(bridging_area, flow.scaled_spacing());
                    bridging_area          = intersection(bridging_area, limiting_area);
                    bridging_area          = intersection(bridging_area, total_fill_area);
                    expansion_area         = diff(expansion_area, bridging_area);

#ifdef DEBUG_BRIDGE_OVER_INFILL
                    debug_draw(std::to_string(lidx) + "_" + std::to_string(cluster_idx) + "_" + std::to_string(job_idx) + "_" + "_expanded_bridging" +  std::to_string(r),
                               to_lines(layer->lslices), to_lines(boundary_plines), to_lines(candidate.new_polys), to_lines(bridging_area));
#endif

                    expanded_surfaces.push_back(CandidateSurface(candidate.original_surface, candidate.layer_index, bridging_area,
                                                                 candidate.region, bridging_angle));
                }
                surfaces_by_layer[lidx].swap(expanded_surfaces);
                expanded_surfaces.clear();
            }
        }
    });

    BOOST_LOG_TRIVIAL(info) << "Bridge over infill - Directions and expanded surfaces computed" << log_memory_info();

    tbb::parallel_for(tbb::blocked_range<size_t>(0, this->layers().size()), [po = this, &surfaces_by_layer](tbb::blocked_range<size_t> r) {
        PRINT_OBJECT_TIME_LIMIT_MILLIS(PRINT_OBJECT_TIME_LIMIT_DEFAULT);
        for (size_t lidx = r.begin(); lidx < r.end(); lidx++) {
            if (surfaces_by_layer.find(lidx) == surfaces_by_layer.end() && surfaces_by_layer.find(lidx + 1) == surfaces_by_layer.end())
                continue;
            Layer *layer = po->get_layer(lidx);

            Polygons cut_from_infill{};
            if (surfaces_by_layer.find(lidx) != surfaces_by_layer.end()) {
                for (const auto &surface : surfaces_by_layer.at(lidx)) {
                    cut_from_infill.insert(cut_from_infill.end(), surface.new_polys.begin(), surface.new_polys.end());
                }
            }

            Polygons additional_ensuring_areas{};
            if (surfaces_by_layer.find(lidx + 1) != surfaces_by_layer.end()) {
                for (const auto &surface : surfaces_by_layer.at(lidx + 1)) {
                    auto additional_area = diff(surface.new_polys,
                                                shrink(surface.new_polys, surface.region->flow(frSolidInfill).scaled_spacing()));
                    additional_ensuring_areas.insert(additional_ensuring_areas.end(), additional_area.begin(), additional_area.end());
                }
            }

            for (LayerRegion *region : layer->regions()) {
                Surfaces new_surfaces;

                Polygons near_perimeters = to_polygons(union_safety_offset_ex(to_polygons(region->fill_surfaces().surfaces)));
                near_perimeters          = diff(near_perimeters, shrink(near_perimeters, region->flow(frSolidInfill).scaled_spacing()));
                ExPolygons additional_ensuring = intersection_ex(additional_ensuring_areas, near_perimeters);

                SurfacesPtr internal_infills = region->m_fill_surfaces.filter_by_type(stInternal);
                ExPolygons new_internal_infills = diff_ex(internal_infills, cut_from_infill);
                new_internal_infills            = diff_ex(new_internal_infills, additional_ensuring);
                for (const ExPolygon &ep : new_internal_infills) {
                    new_surfaces.emplace_back(stInternal, ep);
                }

                SurfacesPtr internal_solids = region->m_fill_surfaces.filter_by_type(stInternalSolid);
                if (surfaces_by_layer.find(lidx) != surfaces_by_layer.end()) {
                    for (const CandidateSurface &cs : surfaces_by_layer.at(lidx)) {
                        for (const Surface *surface : internal_solids) {
                            if (cs.original_surface == surface) {
                                Surface tmp{*surface, {}};
                                tmp.surface_type = stInternalBridge;
                                tmp.bridge_angle = cs.bridge_angle;
                                for (const ExPolygon &ep : union_ex(cs.new_polys)) {
                                    new_surfaces.emplace_back(tmp, ep);
                                }
                                break;
                            }
                        }
                    }
                }
                ExPolygons new_internal_solids = to_expolygons(internal_solids);
                new_internal_solids.insert(new_internal_solids.end(), additional_ensuring.begin(), additional_ensuring.end());
                new_internal_solids = diff_ex(new_internal_solids, cut_from_infill);
                new_internal_solids = union_safety_offset_ex(new_internal_solids);
                for (const ExPolygon &ep : new_internal_solids) {
                    new_surfaces.emplace_back(stInternalSolid, ep);
                }

#ifdef DEBUG_BRIDGE_OVER_INFILL
                debug_draw("Aensuring_" + std::to_string(reinterpret_cast<uint64_t>(&region)), to_polylines(additional_ensuring),
                           to_polylines(near_perimeters), to_polylines(to_polygons(internal_infills)),
                           to_polylines(to_polygons(internal_solids)));
                debug_draw("Aensuring_" + std::to_string(reinterpret_cast<uint64_t>(&region)) + "_new", to_polylines(additional_ensuring),
                           to_polylines(near_perimeters), to_polylines(to_polygons(new_internal_infills)),
                           to_polylines(to_polygons(new_internal_solids)));
#endif

                region->m_fill_surfaces.remove_types({stInternalSolid, stInternal});
                region->m_fill_surfaces.append(new_surfaces);
            }
        }
    });

    BOOST_LOG_TRIVIAL(info) << "Bridge over infill - End" << log_memory_info();

} // void PrintObject::bridge_over_infill()

static void clamp_exturder_to_default(ConfigOptionInt &opt, size_t num_extruders)
{
    if (opt.value > (int)num_extruders)
        // assign the default extruder
        opt.value = 1;
}

PrintObjectConfig PrintObject::object_config_from_model_object(const PrintObjectConfig &default_object_config, const ModelObject &object, size_t num_extruders)
{
    PrintObjectConfig config = default_object_config;
    {
        DynamicPrintConfig src_normalized(object.config.get());
        src_normalized.normalize_fdm();
        config.apply(src_normalized, true);
    }
    // Clamp invalid extruders to the default extruder (with index 1).
    clamp_exturder_to_default(config.support_material_extruder,           num_extruders);
    clamp_exturder_to_default(config.support_material_interface_extruder, num_extruders);
    return config;
}

const std::string                                                    key_extruder { "extruder" };
static constexpr const std::initializer_list<const std::string_view> keys_extruders { "infill_extruder"sv, "solid_infill_extruder"sv, "perimeter_extruder"sv };

static void apply_to_print_region_config(PrintRegionConfig &out, const DynamicPrintConfig &in)
{
    // 1) Copy the "extruder key to infill_extruder and perimeter_extruder.
    auto *opt_extruder = in.opt<ConfigOptionInt>(key_extruder);
    if (opt_extruder)
        if (int extruder = opt_extruder->value; extruder != 0) {
            // Not a default extruder.
            out.infill_extruder      .value = extruder;
            out.solid_infill_extruder.value = extruder;
            out.perimeter_extruder   .value = extruder;
        }
    // 2) Copy the rest of the values.
    for (auto it = in.cbegin(); it != in.cend(); ++ it)
        if (it->first != key_extruder)
            if (ConfigOption* my_opt = out.option(it->first, false); my_opt != nullptr) {
                if (one_of(it->first, keys_extruders)) {
                    // Ignore "default" extruders.
                    int extruder = static_cast<const ConfigOptionInt*>(it->second.get())->value;
                    if (extruder > 0)
                        my_opt->setInt(extruder);
                } else
                    my_opt->set(it->second.get());
            }
}

PrintRegionConfig region_config_from_model_volume(const PrintRegionConfig &default_or_parent_region_config, const DynamicPrintConfig *layer_range_config, const ModelVolume &volume, size_t num_extruders)
{
    PrintRegionConfig config = default_or_parent_region_config;
    if (volume.is_model_part()) {
        // default_or_parent_region_config contains the Print's PrintRegionConfig.
        // Override with ModelObject's PrintRegionConfig values.
        apply_to_print_region_config(config, volume.get_object()->config.get());
    } else {
        // default_or_parent_region_config contains parent PrintRegion config, which already contains ModelVolume's config.
    }
    if (layer_range_config != nullptr) {
        // Not applicable to modifiers.
        assert(volume.is_model_part());
    	apply_to_print_region_config(config, *layer_range_config);
    }
    apply_to_print_region_config(config, volume.config.get());
    if (! volume.material_id().empty())
        apply_to_print_region_config(config, volume.material()->config.get());
    // Clamp invalid extruders to the default extruder (with index 1).
    clamp_exturder_to_default(config.infill_extruder,       num_extruders);
    clamp_exturder_to_default(config.perimeter_extruder,    num_extruders);
    clamp_exturder_to_default(config.solid_infill_extruder, num_extruders);
    if (config.fill_density.value < 0.00011f)
        // Switch of infill for very low infill rates, also avoid division by zero in infill generator for these very low rates.
        // See GH issue #5910.
        config.fill_density.value = 0;
    else 
        config.fill_density.value = std::min(config.fill_density.value, 100.);
    if (config.fuzzy_skin.value != FuzzySkinType::None && (config.fuzzy_skin_point_dist.value < 0.01 || config.fuzzy_skin_thickness.value < 0.001))
        config.fuzzy_skin.value = FuzzySkinType::None;
    return config;
}

void PrintObject::update_slicing_parameters()
{
    if (!m_slicing_params.valid)
        m_slicing_params = SlicingParameters::create_from_config(
            this->print()->config(), m_config, this->model_object()->max_z(), this->object_extruders());
}

SlicingParameters PrintObject::slicing_parameters(const DynamicPrintConfig& full_config, const ModelObject& model_object, float object_max_z)
{
	PrintConfig         print_config;
	PrintObjectConfig   object_config;
	PrintRegionConfig   default_region_config;
	print_config.apply(full_config, true);
	object_config.apply(full_config, true);
	default_region_config.apply(full_config, true);
	size_t              num_extruders = print_config.nozzle_diameter.size();
	object_config = object_config_from_model_object(object_config, model_object, num_extruders);

	std::vector<unsigned int> object_extruders;
	for (const ModelVolume* model_volume : model_object.volumes)
		if (model_volume->is_model_part()) {
			PrintRegion::collect_object_printing_extruders(
				print_config,
				region_config_from_model_volume(default_region_config, nullptr, *model_volume, num_extruders),
                object_config.brim_type != btNoBrim && object_config.brim_width > 0.,
				object_extruders);
			for (const std::pair<const t_layer_height_range, ModelConfig> &range_and_config : model_object.layer_config_ranges)
				if (range_and_config.second.has("perimeter_extruder") ||
					range_and_config.second.has("infill_extruder") ||
					range_and_config.second.has("solid_infill_extruder"))
					PrintRegion::collect_object_printing_extruders(
						print_config,
						region_config_from_model_volume(default_region_config, &range_and_config.second.get(), *model_volume, num_extruders),
                        object_config.brim_type != btNoBrim && object_config.brim_width > 0.,
						object_extruders);
		}
    sort_remove_duplicates(object_extruders);
    //FIXME add painting extruders

    if (object_max_z <= 0.f)
        object_max_z = (float)model_object.raw_bounding_box().size().z();
    return SlicingParameters::create_from_config(print_config, object_config, object_max_z, object_extruders);
}

// returns 0-based indices of extruders used to print the object (without brim, support and other helper extrusions)
std::vector<unsigned int> PrintObject::object_extruders() const
{
    std::vector<unsigned int> extruders;
    extruders.reserve(this->all_regions().size() * 3);
    for (const PrintRegion &region : this->all_regions())
        region.collect_object_printing_extruders(*this->print(), extruders);
    sort_remove_duplicates(extruders);
    return extruders;
}

bool PrintObject::update_layer_height_profile(const ModelObject &model_object, const SlicingParameters &slicing_parameters, std::vector<coordf_t> &layer_height_profile)
{
    bool updated = false;

    if (layer_height_profile.empty()) {
        // use the constructor because the assignement is crashing on ASAN OsX
        layer_height_profile = model_object.layer_height_profile.get();
//        layer_height_profile = model_object.layer_height_profile;
        // The layer height returned is sampled with high density for the UI layer height painting
        // and smoothing tool to work.
        updated = true;
    }

    // Verify the layer_height_profile.
    if (!layer_height_profile.empty() &&
        // Must not be of even length.
        ((layer_height_profile.size() & 1) != 0 ||
            // Last entry must be at the top of the object.
            std::abs(layer_height_profile[layer_height_profile.size() - 2] - slicing_parameters.object_print_z_max + slicing_parameters.object_print_z_min) > 1e-3))
        layer_height_profile.clear();

    if (layer_height_profile.empty()) {
        //layer_height_profile = layer_height_profile_adaptive(slicing_parameters, model_object.layer_config_ranges, model_object.volumes);
        layer_height_profile = layer_height_profile_from_ranges(slicing_parameters, model_object.layer_config_ranges);
        // The layer height profile is already compressed.
        updated = true;
    }
    return updated;
}

// Only active if config->infill_only_where_needed. This step trims the sparse infill,
// so it acts as an internal support. It maintains all other infill types intact.
// Here the internal surfaces and perimeters have to be supported by the sparse infill.
//FIXME The surfaces are supported by a sparse infill, but the sparse infill is only as large as the area to support.
// Likely the sparse infill will not be anchored correctly, so it will not work as intended.
// Also one wishes the perimeters to be supported by a full infill.
// Idempotence of this method is guaranteed by the fact that we don't remove things from
// fill_surfaces but we only turn them into VOID surfaces, thus preserving the boundaries.
// void PrintObject::clip_fill_surfaces()
// {
//     bool has_lightning_infill = false;
//     for (size_t region_id = 0; region_id < this->num_printing_regions(); ++region_id)
//         if (const PrintRegionConfig &config = this->printing_region(region_id).config(); config.fill_density > 0 && config.fill_pattern == ipLightning)
//             has_lightning_infill = true;

//     // For Lightning infill, infill_only_where_needed is ignored because both
//     // do a similar thing, and their combination doesn't make much sense.
//     if (! m_config.infill_only_where_needed.value || has_lightning_infill)
//         return;
//     bool has_infill = false;
//     for (size_t i = 0; i < this->num_printing_regions(); ++ i)
//         if (this->printing_region(i).config().fill_density > 0) {
//             has_infill = true;
//             break;
//         }
//     if (! has_infill)
//         return;

//     // We only want infill under ceilings; this is almost like an
//     // internal support material.
//     // Proceed top-down, skipping the bottom layer.
//     Polygons upper_internal;
//     for (int layer_id = int(m_layers.size()) - 1; layer_id > 0; -- layer_id) {
//         Layer *layer       = m_layers[layer_id];
//         Layer *lower_layer = m_layers[layer_id - 1];
//         // Detect things that we need to support.
//         // Cummulative fill surfaces.
//         Polygons fill_surfaces;
//         // Solid surfaces to be supported.
//         Polygons overhangs;
//         for (const LayerRegion *layerm : layer->m_regions)
//             for (const Surface &surface : layerm->fill_surfaces()) {
//                 Polygons polygons = to_polygons(surface.expolygon);
//                 if (surface.is_solid())
//                     polygons_append(overhangs, polygons);
//                 polygons_append(fill_surfaces, std::move(polygons));
//             }
//         Polygons lower_layer_fill_surfaces;
//         Polygons lower_layer_internal_surfaces;
//         for (const LayerRegion *layerm : lower_layer->m_regions)
//             for (const Surface &surface : layerm->fill_surfaces()) {
//                 Polygons polygons = to_polygons(surface.expolygon);
//                 if (surface.surface_type == stInternal || surface.surface_type == stInternalVoid)
//                     polygons_append(lower_layer_internal_surfaces, polygons);
//                 polygons_append(lower_layer_fill_surfaces, std::move(polygons));
//             }
//         // We also need to support perimeters when there's at least one full unsupported loop
//         {
//             // Get perimeters area as the difference between slices and fill_surfaces
//             // Only consider the area that is not supported by lower perimeters
//             Polygons perimeters = intersection(diff(layer->lslices, fill_surfaces), lower_layer_fill_surfaces);
//             // Only consider perimeter areas that are at least one extrusion width thick.
//             //FIXME Offset2 eats out from both sides, while the perimeters are create outside in.
//             //Should the pw not be half of the current value?
//             float pw = FLT_MAX;
//             for (const LayerRegion *layerm : layer->m_regions)
//                 pw = std::min(pw, (float)layerm->flow(frPerimeter).scaled_width());
//             // Append such thick perimeters to the areas that need support
//             polygons_append(overhangs, opening(perimeters, pw));
//         }
//         // Merge the new overhangs, find new internal infill.
//         polygons_append(upper_internal, std::move(overhangs));
//         static constexpr const auto closing_radius = scaled<float>(2.f);
//         upper_internal = intersection(
//             // Regularize the overhang regions, so that the infill areas will not become excessively jagged.
//             smooth_outward(
//                 closing(upper_internal, closing_radius, ClipperLib::jtSquare, 0.),
//                 scaled<coord_t>(0.1)), 
//             lower_layer_internal_surfaces);
//         // Apply new internal infill to regions.
//         for (LayerRegion *layerm : lower_layer->m_regions) {
//             if (layerm->region().config().fill_density.value == 0)
//                 continue;
//             Polygons internal;
//             for (Surface &surface : layerm->m_fill_surfaces.surfaces)
//                 if (surface.surface_type == stInternal || surface.surface_type == stInternalVoid)
//                     polygons_append(internal, std::move(surface.expolygon));
//             layerm->m_fill_surfaces.remove_types({ stInternal, stInternalVoid });
//             layerm->m_fill_surfaces.append(intersection_ex(internal, upper_internal, ApplySafetyOffset::Yes), stInternal);
//             layerm->m_fill_surfaces.append(diff_ex        (internal, upper_internal, ApplySafetyOffset::Yes), stInternalVoid);
//             // If there are voids it means that our internal infill is not adjacent to
//             // perimeters. In this case it would be nice to add a loop around infill to
//             // make it more robust and nicer. TODO.
// #ifdef SLIC3R_DEBUG_SLICE_PROCESSING
//             layerm->export_region_fill_surfaces_to_svg_debug("6_clip_fill_surfaces");
// #endif
//         }
//         m_print->throw_if_canceled();
//     }
// } // void PrintObject::clip_fill_surfaces()

void PrintObject::discover_horizontal_shells()
{
    BOOST_LOG_TRIVIAL(trace) << "discover_horizontal_shells()";

    for (size_t region_id = 0; region_id < this->num_printing_regions(); ++region_id) {
        for (size_t i = 0; i < m_layers.size(); ++i) {
            m_print->throw_if_canceled();
            Layer                   *layer         = m_layers[i];
            LayerRegion             *layerm        = layer->regions()[region_id];
            const PrintRegionConfig &region_config = layerm->region().config();
            if (region_config.solid_infill_every_layers.value > 0 && region_config.fill_density.value > 0 &&
                (i % region_config.solid_infill_every_layers) == 0) {
                // Insert a solid internal layer. Mark stInternal surfaces as stInternalSolid or stInternalBridge.
                SurfaceType type = (region_config.fill_density == 100 || region_config.solid_infill_every_layers == 1) ? stInternalSolid :
                                                                                                                         stInternalBridge;
                for (Surface &surface : layerm->m_fill_surfaces.surfaces)
                    if (surface.surface_type == stInternal)
                        surface.surface_type = type;
            }
            // The rest has already been performed by discover_vertical_shells().
        } // for each layer
    }     // for each region

#ifdef SLIC3R_DEBUG_SLICE_PROCESSING
    for (size_t region_id = 0; region_id < this->num_printing_regions(); ++region_id) {
        for (const Layer *layer : m_layers) {
            const LayerRegion *layerm = layer->m_regions[region_id];
            layerm->export_region_slices_to_svg_debug("5_discover_horizontal_shells");
            layerm->export_region_fill_surfaces_to_svg_debug("5_discover_horizontal_shells");
        } // for each layer
    }     // for each region
#endif    /* SLIC3R_DEBUG_SLICE_PROCESSING */
} // void PrintObject::discover_horizontal_shells()

// combine fill surfaces across layers to honor the "infill every N layers" option
// Idempotence of this method is guaranteed by the fact that we don't remove things from
// fill_surfaces but we only turn them into VOID surfaces, thus preserving the boundaries.
void PrintObject::combine_infill()
{
    // Work on each region separately.
    for (size_t region_id = 0; region_id < this->num_printing_regions(); ++ region_id) {
        const PrintRegion &region = this->printing_region(region_id);
        const size_t every = region.config().infill_every_layers.value;
        if (every < 2 || region.config().fill_density == 0.)
            continue;
        // Limit the number of combined layers to the maximum height allowed by this regions' nozzle.
        //FIXME limit the layer height to max_layer_height
        double nozzle_diameter = std::min(
            this->print()->config().nozzle_diameter.get_at(region.config().infill_extruder.value - 1),
            this->print()->config().nozzle_diameter.get_at(region.config().solid_infill_extruder.value - 1));
        // define the combinations
        std::vector<size_t> combine(m_layers.size(), 0);
        {
            double current_height = 0.;
            size_t num_layers = 0;
            for (size_t layer_idx = 0; layer_idx < m_layers.size(); ++ layer_idx) {
                m_print->throw_if_canceled();
                const Layer *layer = m_layers[layer_idx];
                if (layer->id() == 0)
                    // Skip first print layer (which may not be first layer in array because of raft).
                    continue;
                // Check whether the combination of this layer with the lower layers' buffer
                // would exceed max layer height or max combined layer count.
                if (current_height + layer->height >= nozzle_diameter + EPSILON || num_layers >= every) {
                    // Append combination to lower layer.
                    combine[layer_idx - 1] = num_layers;
                    current_height = 0.;
                    num_layers = 0;
                }
                current_height += layer->height;
                ++ num_layers;
            }
            
            // Append lower layers (if any) to uppermost layer.
            combine[m_layers.size() - 1] = num_layers;
        }
        
        // loop through layers to which we have assigned layers to combine
        for (size_t layer_idx = 0; layer_idx < m_layers.size(); ++ layer_idx) {
            m_print->throw_if_canceled();
            size_t num_layers = combine[layer_idx];
			if (num_layers <= 1)
                continue;
            // Get all the LayerRegion objects to be combined.
            std::vector<LayerRegion*> layerms;
            layerms.reserve(num_layers);
			for (size_t i = layer_idx + 1 - num_layers; i <= layer_idx; ++ i)
                layerms.emplace_back(m_layers[i]->regions()[region_id]);
            // We need to perform a multi-layer intersection, so let's split it in pairs.
            // Initialize the intersection with the candidates of the lowest layer.
            ExPolygons intersection = to_expolygons(layerms.front()->fill_surfaces().filter_by_type(stInternal));
            // Start looping from the second layer and intersect the current intersection with it.
            for (size_t i = 1; i < layerms.size(); ++ i)
                intersection = intersection_ex(layerms[i]->fill_surfaces().filter_by_type(stInternal), intersection);
            double area_threshold = layerms.front()->infill_area_threshold();
            if (! intersection.empty() && area_threshold > 0.)
                intersection.erase(std::remove_if(intersection.begin(), intersection.end(), 
                    [area_threshold](const ExPolygon &expoly) { return expoly.area() <= area_threshold; }), 
                    intersection.end());
            if (intersection.empty())
                continue;
//            Slic3r::debugf "  combining %d %s regions from layers %d-%d\n",
//                scalar(@$intersection),
//                ($type == stInternal ? 'internal' : 'internal-solid'),
//                $layer_idx-($every-1), $layer_idx;
            // intersection now contains the regions that can be combined across the full amount of layers,
            // so let's remove those areas from all layers.
            Polygons intersection_with_clearance;
            intersection_with_clearance.reserve(intersection.size());
            float clearance_offset = 
                0.5f * layerms.back()->flow(frPerimeter).scaled_width() +
             // Because fill areas for rectilinear and honeycomb are grown 
             // later to overlap perimeters, we need to counteract that too.
                ((region.config().fill_pattern == ipRectilinear   ||
                  region.config().fill_pattern == ipMonotonic     ||
                  region.config().fill_pattern == ipGrid          ||
                  region.config().fill_pattern == ipLine          ||
                  region.config().fill_pattern == ipHoneycomb) ? 1.5f : 0.5f) * 
                    layerms.back()->flow(frSolidInfill).scaled_width();
            for (ExPolygon &expoly : intersection)
                polygons_append(intersection_with_clearance, offset(expoly, clearance_offset));
            for (LayerRegion *layerm : layerms) {
                Polygons internal = to_polygons(std::move(layerm->fill_surfaces().filter_by_type(stInternal)));
                layerm->m_fill_surfaces.remove_type(stInternal);
                layerm->m_fill_surfaces.append(diff_ex(internal, intersection_with_clearance), stInternal);
                if (layerm == layerms.back()) {
                    // Apply surfaces back with adjusted depth to the uppermost layer.
                    Surface templ(stInternal, ExPolygon());
                    templ.thickness = 0.;
                    for (LayerRegion *layerm2 : layerms)
                        templ.thickness += layerm2->layer()->height;
                    templ.thickness_layers = (unsigned short)layerms.size();
                    layerm->m_fill_surfaces.append(intersection, templ);
                } else {
                    // Save void surfaces.
                    layerm->m_fill_surfaces.append(
                        intersection_ex(internal, intersection_with_clearance),
                        stInternalVoid);
                }
            }
        }
    }
} // void PrintObject::combine_infill()

void PrintObject::_generate_support_material()
{
    if (this->has_support() && (m_config.support_material_style == smsTree || m_config.support_material_style == smsOrganic)) {
        fff_tree_support_generate(*this, std::function<void()>([this](){ this->throw_if_canceled(); }));
    } else {
        // If support style is set to Organic however only raft will be built but no support,
        // build snug raft instead.
        PrintObjectSupportMaterial support_material(this, m_slicing_params);
        support_material.generate(*this);
    }
}

static void project_triangles_to_slabs(SpanOfConstPtrs<Layer> layers, const indexed_triangle_set &custom_facets, const Transform3f &tr, bool seam, std::vector<Polygons> &out)
{
    if (custom_facets.indices.empty())
        return;

    const float tr_det_sign = (tr.matrix().determinant() > 0. ? 1.f : -1.f);

    // The projection will be at most a pentagon. Let's minimize heap
    // reallocations by saving in in the following struct.
    // Points are used so that scaling can be done in parallel
    // and they can be moved from to create an ExPolygon later.
    struct LightPolygon {
        LightPolygon() { pts.reserve(5); }
        LightPolygon(const std::array<Vec2f, 3>& tri) {
            pts.reserve(3);
            pts.emplace_back(scaled<coord_t>(tri.front()));
            pts.emplace_back(scaled<coord_t>(tri[1]));
            pts.emplace_back(scaled<coord_t>(tri.back()));
        }

        Points pts;

        void add(const Vec2f& pt) {
            pts.emplace_back(scaled<coord_t>(pt));
            assert(pts.size() <= 5);
        }
    };

    // Structure to collect projected polygons. One element for each triangle.
    // Saves vector of polygons and layer_id of the first one.
    struct TriangleProjections {
        size_t first_layer_id;
        std::vector<LightPolygon> polygons;
    };

    // Vector to collect resulting projections from each triangle.
    std::vector<TriangleProjections> projections_of_triangles(custom_facets.indices.size());

    // Iterate over all triangles.
    tbb::parallel_for(
        tbb::blocked_range<size_t>(0, custom_facets.indices.size()),
        [&custom_facets, &tr, tr_det_sign, seam, layers, &projections_of_triangles](const tbb::blocked_range<size_t>& range) {
        for (size_t idx = range.begin(); idx < range.end(); ++ idx) {

        PRINT_OBJECT_TIME_LIMIT_MILLIS(PRINT_OBJECT_TIME_LIMIT_DEFAULT);
        std::array<Vec3f, 3> facet;

        // Transform the triangle into worlds coords.
        for (int i=0; i<3; ++i)
            facet[i] = tr * custom_facets.vertices[custom_facets.indices[idx](i)];

        // Ignore triangles with upward-pointing normal. Don't forget about mirroring.
        float z_comp = (facet[1]-facet[0]).cross(facet[2]-facet[0]).z();
        if (! seam && tr_det_sign * z_comp > 0.)
            continue;

        // The algorithm does not process vertical triangles, but it should for seam.
        // In that case, tilt the triangle a bit so the projection does not degenerate.
        if (seam && z_comp == 0.f)
            facet[0].x() += float(EPSILON);

        // Sort the three vertices according to z-coordinate.
        std::sort(facet.begin(), facet.end(),
                  [](const Vec3f& pt1, const Vec3f&pt2) {
                      return pt1.z() < pt2.z();
                  });

        std::array<Vec2f, 3> trianglef;
        for (int i=0; i<3; ++i)
            trianglef[i] = to_2d(facet[i]);

        // Find lowest slice not below the triangle.
        auto it = std::lower_bound(layers.begin(), layers.end(), facet[0].z()+EPSILON,
                      [](const Layer* l1, float z) {
                           return l1->slice_z < z;
                      });

        // Count how many projections will be generated for this triangle
        // and allocate respective amount in projections_of_triangles.
        size_t first_layer_id = projections_of_triangles[idx].first_layer_id = it - layers.begin();
        size_t last_layer_id  = first_layer_id;
        // The cast in the condition below is important. The comparison must
        // be an exact opposite of the one lower in the code where
        // the polygons are appended. And that one is on floats.
        while (last_layer_id + 1 < layers.size()
            && float(layers[last_layer_id]->slice_z) <= facet[2].z())
            ++last_layer_id;

        if (first_layer_id == last_layer_id) {
            // The triangle fits just a single slab, just project it. This also avoids division by zero for horizontal triangles.
            float dz = facet[2].z() - facet[0].z();
            assert(dz >= 0);
            // The face is nearly horizontal and it crosses the slicing plane at first_layer_id - 1.
            // Rather add this face to both the planes.
            bool add_below = dz < float(2. * EPSILON) && first_layer_id > 0 && layers[first_layer_id - 1]->slice_z > facet[0].z() - EPSILON;
            projections_of_triangles[idx].polygons.reserve(add_below ? 2 : 1);
            projections_of_triangles[idx].polygons.emplace_back(trianglef);
            if (add_below) {
                -- projections_of_triangles[idx].first_layer_id;
                projections_of_triangles[idx].polygons.emplace_back(trianglef);
            }
            continue;
        }

        projections_of_triangles[idx].polygons.resize(last_layer_id - first_layer_id + 1);

        // Calculate how to move points on triangle sides per unit z increment.
        Vec2f ta(trianglef[1] - trianglef[0]);
        Vec2f tb(trianglef[2] - trianglef[0]);
        ta *= 1.f/(facet[1].z() - facet[0].z());
        tb *= 1.f/(facet[2].z() - facet[0].z());

        // Projection on current slice will be built directly in place.
        LightPolygon* proj = &projections_of_triangles[idx].polygons[0];
        proj->add(trianglef[0]);

        bool passed_first = false;
        bool stop = false;

        // Project a sub-polygon on all slices intersecting the triangle.
        while (it != layers.end()) {
            const float z = float((*it)->slice_z);

            // Projections of triangle sides intersections with slices.
            // a moves along one side, b tracks the other.
            Vec2f a;
            Vec2f b;

            // If the middle vertex was already passed, append the vertex
            // and use ta for tracking the remaining side.
            if (z > facet[1].z() && ! passed_first) {
                proj->add(trianglef[1]);
                ta = trianglef[2]-trianglef[1];
                ta *= 1.f/(facet[2].z() - facet[1].z());
                passed_first = true;
            }

            // This slice is above the triangle already.
            if (z > facet[2].z() || it+1 == layers.end()) {
                proj->add(trianglef[2]);
                stop = true;
            }
            else {
                // Move a, b along the side it currently tracks to get
                // projected intersection with current slice.
                a = passed_first ? (trianglef[1]+ta*(z-facet[1].z()))
                                 : (trianglef[0]+ta*(z-facet[0].z()));
                b = trianglef[0]+tb*(z-facet[0].z());
                proj->add(a);
                proj->add(b);
            }

           if (stop)
                break;

            // Advance to the next layer.
            ++it;
            ++proj;
            assert(proj <= &projections_of_triangles[idx].polygons.back() );

            // a, b are first two points of the polygon for the next layer.
            proj->add(b);
            proj->add(a);
        }
    }
    }); // end of parallel_for

    // Make sure that the output vector can be used.
    out.resize(layers.size());

    // Now append the collected polygons to respective layers.
    for (auto& trg : projections_of_triangles) {
        int layer_id = int(trg.first_layer_id);
        for (LightPolygon &poly : trg.polygons) {
            if (layer_id >= int(out.size()))
                break; // part of triangle could be projected above top layer
            assert(! poly.pts.empty());
            // The resulting triangles are fed to the Clipper library, which seem to handle flipped triangles well.
//                if (cross2(Vec2d((poly.pts[1] - poly.pts[0]).cast<double>()), Vec2d((poly.pts[2] - poly.pts[1]).cast<double>())) < 0)
//                    std::swap(poly.pts.front(), poly.pts.back());
                
            out[layer_id].emplace_back(std::move(poly.pts));
            ++layer_id;
        }
    }
}

void PrintObject::project_and_append_custom_facets(
        bool seam, EnforcerBlockerType type, std::vector<Polygons>& out) const
{
    for (const ModelVolume* mv : this->model_object()->volumes)
        if (mv->is_model_part()) {
            const indexed_triangle_set custom_facets = seam
                    ? mv->seam_facets.get_facets_strict(*mv, type)
                    : mv->supported_facets.get_facets_strict(*mv, type);
            if (! custom_facets.indices.empty()) {
                if (seam)
                    project_triangles_to_slabs(this->layers(), custom_facets,
                        (this->trafo_centered() * mv->get_matrix()).cast<float>(),
                        seam, out);
                else {
                    std::vector<Polygons> projected;
                    // Support blockers or enforcers. Project downward facing painted areas upwards to their respective slicing plane.
                    slice_mesh_slabs(custom_facets, zs_from_layers(this->layers()), this->trafo_centered() * mv->get_matrix(), nullptr, &projected, [](){});
                    // Merge these projections with the output, layer by layer.
                    assert(! projected.empty());
                    assert(out.empty() || out.size() == projected.size());
                    if (out.empty())
                        out = std::move(projected);
                    else
                        for (size_t i = 0; i < out.size(); ++ i)
                            append(out[i], std::move(projected[i]));
                }
            }
        }
}

const Layer* PrintObject::get_layer_at_printz(coordf_t print_z) const {
    auto it = Slic3r::lower_bound_by_predicate(m_layers.begin(), m_layers.end(), [print_z](const Layer *layer) { return layer->print_z < print_z; });
    return (it == m_layers.end() || (*it)->print_z != print_z) ? nullptr : *it;
}



Layer* PrintObject::get_layer_at_printz(coordf_t print_z) { return const_cast<Layer*>(std::as_const(*this).get_layer_at_printz(print_z)); }



// Get a layer approximately at print_z.
const Layer* PrintObject::get_layer_at_printz(coordf_t print_z, coordf_t epsilon) const {
    coordf_t limit = print_z - epsilon;
    auto it = Slic3r::lower_bound_by_predicate(m_layers.begin(), m_layers.end(), [limit](const Layer *layer) { return layer->print_z < limit; });
    return (it == m_layers.end() || (*it)->print_z > print_z + epsilon) ? nullptr : *it;
}



Layer* PrintObject::get_layer_at_printz(coordf_t print_z, coordf_t epsilon) { return const_cast<Layer*>(std::as_const(*this).get_layer_at_printz(print_z, epsilon)); }

const Layer *PrintObject::get_first_layer_bellow_printz(coordf_t print_z, coordf_t epsilon) const
{
    coordf_t limit = print_z + epsilon;
    auto it = Slic3r::lower_bound_by_predicate(m_layers.begin(), m_layers.end(), [limit](const Layer *layer) { return layer->print_z < limit; });
    return (it == m_layers.begin()) ? nullptr : *(--it);
}

} // namespace Slic3r