#include "SLAPrint.hpp" #include "SLA/SLASupportTree.hpp" #include "SLA/SLABasePool.hpp" #include "SLA/SLAAutoSupports.hpp" #include "ClipperUtils.hpp" #include "MTUtils.hpp" #include #include #include #include #include //#include //#include "tbb/mutex.h" #include "I18N.hpp" //! macro used to mark string used at localization, //! return same string #define L(s) Slic3r::I18N::translate(s) namespace Slic3r { using SupportTreePtr = std::unique_ptr; class SLAPrintObject::SupportData { public: sla::EigenMesh3D emesh; // index-triangle representation std::vector support_points; // all the support points (manual/auto) SupportTreePtr support_tree_ptr; // the supports SlicedSupports support_slices; // sliced supports std::vector level_ids; inline SupportData(const TriangleMesh& trmesh): emesh(trmesh) {} }; namespace { // should add up to 100 (%) const std::array OBJ_STEP_LEVELS = { 10, // slaposObjectSlice, 30, // slaposSupportPoints, 25, // slaposSupportTree, 25, // slaposBasePool, 5, // slaposSliceSupports, 5 // slaposIndexSlices }; const std::array OBJ_STEP_LABELS = { L("Slicing model"), // slaposObjectSlice, L("Generating support points"), // slaposSupportPoints, L("Generating support tree"), // slaposSupportTree, L("Generating pad"), // slaposBasePool, L("Slicing supports"), // slaposSliceSupports, L("Slicing supports") // slaposIndexSlices, }; // Should also add up to 100 (%) const std::array PRINT_STEP_LEVELS = { 80, // slapsRasterize 20, // slapsValidate }; const std::array PRINT_STEP_LABELS = { L("Rasterizing layers"), // slapsRasterize L("Validating"), // slapsValidate }; } void SLAPrint::clear() { tbb::mutex::scoped_lock lock(this->state_mutex()); // The following call should stop background processing if it is running. this->invalidate_all_steps(); for (SLAPrintObject *object : m_objects) delete object; m_objects.clear(); m_model.clear_objects(); } // Transformation without rotation around Z and without a shift by X and Y. static Transform3d sla_trafo(const ModelObject &model_object) { ModelInstance &model_instance = *model_object.instances.front(); Vec3d offset = model_instance.get_offset(); Vec3d rotation = model_instance.get_rotation(); offset(0) = 0.; offset(1) = 0.; rotation(2) = 0.; return Geometry::assemble_transform(offset, rotation, model_instance.get_scaling_factor(), model_instance.get_mirror()); } // List of instances, where the ModelInstance transformation is a composite of sla_trafo and the transformation defined by SLAPrintObject::Instance. static std::vector sla_instances(const ModelObject &model_object) { std::vector instances; for (ModelInstance *model_instance : model_object.instances) if (model_instance->is_printable()) { instances.emplace_back(SLAPrintObject::Instance( model_instance->id(), Point::new_scale(model_instance->get_offset(X), model_instance->get_offset(Y)), float(model_instance->get_rotation(Z)))); } return instances; } SLAPrint::ApplyStatus SLAPrint::apply(const Model &model, const DynamicPrintConfig &config_in) { #ifdef _DEBUG check_model_ids_validity(model); #endif /* _DEBUG */ // Make a copy of the config, normalize it. DynamicPrintConfig config(config_in); config.normalize(); // Collect changes to print config. t_config_option_keys print_diff = m_print_config.diff(config); t_config_option_keys printer_diff = m_printer_config.diff(config); t_config_option_keys material_diff = m_material_config.diff(config); t_config_option_keys object_diff = m_default_object_config.diff(config); t_config_option_keys placeholder_parser_diff = this->placeholder_parser().config_diff(config); // Do not use the ApplyStatus as we will use the max function when updating apply_status. unsigned int apply_status = APPLY_STATUS_UNCHANGED; auto update_apply_status = [&apply_status](bool invalidated) { apply_status = std::max(apply_status, invalidated ? APPLY_STATUS_INVALIDATED : APPLY_STATUS_CHANGED); }; if (! (print_diff.empty() && printer_diff.empty() && material_diff.empty() && object_diff.empty())) update_apply_status(false); // Grab the lock for the Print / PrintObject milestones. tbb::mutex::scoped_lock lock(this->state_mutex()); // The following call may stop the background processing. if (! print_diff.empty()) update_apply_status(this->invalidate_state_by_config_options(print_diff)); if (! printer_diff.empty()) update_apply_status(this->invalidate_state_by_config_options(printer_diff)); if (! material_diff.empty()) update_apply_status(this->invalidate_state_by_config_options(material_diff)); // Apply variables to placeholder parser. The placeholder parser is currently used // only to generate the output file name. if (! placeholder_parser_diff.empty()) { // update_apply_status(this->invalidate_step(slapsRasterize)); PlaceholderParser &pp = this->placeholder_parser(); pp.apply_config(config); // Set the profile aliases for the PrintBase::output_filename() pp.set("print_preset", config_in.option("sla_print_settings_id")->clone()); pp.set("material_preset", config_in.option("sla_material_settings_id")->clone()); pp.set("printer_preset", config_in.option("printer_settings_id")->clone()); } // It is also safe to change m_config now after this->invalidate_state_by_config_options() call. m_print_config.apply_only(config, print_diff, true); m_printer_config.apply_only(config, printer_diff, true); // Handle changes to material config. m_material_config.apply_only(config, material_diff, true); // Handle changes to object config defaults m_default_object_config.apply_only(config, object_diff, true); struct ModelObjectStatus { enum Status { Unknown, Old, New, Moved, Deleted, }; ModelObjectStatus(ModelID id, Status status = Unknown) : id(id), status(status) {} ModelID id; Status status; // Search by id. bool operator<(const ModelObjectStatus &rhs) const { return id < rhs.id; } }; std::set model_object_status; // 1) Synchronize model objects. if (model.id() != m_model.id()) { // Kill everything, initialize from scratch. // Stop background processing. this->call_cancel_callback(); update_apply_status(this->invalidate_all_steps()); for (SLAPrintObject *object : m_objects) { model_object_status.emplace(object->model_object()->id(), ModelObjectStatus::Deleted); update_apply_status(object->invalidate_all_steps()); delete object; } m_objects.clear(); m_model.assign_copy(model); for (const ModelObject *model_object : m_model.objects) model_object_status.emplace(model_object->id(), ModelObjectStatus::New); } else { if (model_object_list_equal(m_model, model)) { // The object list did not change. for (const ModelObject *model_object : m_model.objects) model_object_status.emplace(model_object->id(), ModelObjectStatus::Old); } else if (model_object_list_extended(m_model, model)) { // Add new objects. Their volumes and configs will be synchronized later. update_apply_status(this->invalidate_step(slapsRasterize)); for (const ModelObject *model_object : m_model.objects) model_object_status.emplace(model_object->id(), ModelObjectStatus::Old); for (size_t i = m_model.objects.size(); i < model.objects.size(); ++ i) { model_object_status.emplace(model.objects[i]->id(), ModelObjectStatus::New); m_model.objects.emplace_back(ModelObject::new_copy(*model.objects[i])); m_model.objects.back()->set_model(&m_model); } } else { // Reorder the objects, add new objects. // First stop background processing before shuffling or deleting the PrintObjects in the object list. this->call_cancel_callback(); update_apply_status(this->invalidate_step(slapsRasterize)); // Second create a new list of objects. std::vector model_objects_old(std::move(m_model.objects)); m_model.objects.clear(); m_model.objects.reserve(model.objects.size()); auto by_id_lower = [](const ModelObject *lhs, const ModelObject *rhs){ return lhs->id() < rhs->id(); }; std::sort(model_objects_old.begin(), model_objects_old.end(), by_id_lower); for (const ModelObject *mobj : model.objects) { auto it = std::lower_bound(model_objects_old.begin(), model_objects_old.end(), mobj, by_id_lower); if (it == model_objects_old.end() || (*it)->id() != mobj->id()) { // New ModelObject added. m_model.objects.emplace_back(ModelObject::new_copy(*mobj)); m_model.objects.back()->set_model(&m_model); model_object_status.emplace(mobj->id(), ModelObjectStatus::New); } else { // Existing ModelObject re-added (possibly moved in the list). m_model.objects.emplace_back(*it); model_object_status.emplace(mobj->id(), ModelObjectStatus::Moved); } } bool deleted_any = false; for (ModelObject *&model_object : model_objects_old) { if (model_object_status.find(ModelObjectStatus(model_object->id())) == model_object_status.end()) { model_object_status.emplace(model_object->id(), ModelObjectStatus::Deleted); deleted_any = true; } else // Do not delete this ModelObject instance. model_object = nullptr; } if (deleted_any) { // Delete PrintObjects of the deleted ModelObjects. std::vector print_objects_old = std::move(m_objects); m_objects.clear(); m_objects.reserve(print_objects_old.size()); for (SLAPrintObject *print_object : print_objects_old) { auto it_status = model_object_status.find(ModelObjectStatus(print_object->model_object()->id())); assert(it_status != model_object_status.end()); if (it_status->status == ModelObjectStatus::Deleted) { update_apply_status(print_object->invalidate_all_steps()); delete print_object; } else m_objects.emplace_back(print_object); } for (ModelObject *model_object : model_objects_old) delete model_object; } } } // 2) Map print objects including their transformation matrices. struct PrintObjectStatus { enum Status { Unknown, Deleted, Reused, New }; PrintObjectStatus(SLAPrintObject *print_object, Status status = Unknown) : id(print_object->model_object()->id()), print_object(print_object), trafo(print_object->trafo()), status(status) {} PrintObjectStatus(ModelID id) : id(id), print_object(nullptr), trafo(Transform3d::Identity()), status(Unknown) {} // ID of the ModelObject & PrintObject ModelID id; // Pointer to the old PrintObject SLAPrintObject *print_object; // Trafo generated with model_object->world_matrix(true) Transform3d trafo; Status status; // Search by id. bool operator<(const PrintObjectStatus &rhs) const { return id < rhs.id; } }; std::multiset print_object_status; for (SLAPrintObject *print_object : m_objects) print_object_status.emplace(PrintObjectStatus(print_object)); // 3) Synchronize ModelObjects & PrintObjects. std::vector print_objects_new; print_objects_new.reserve(std::max(m_objects.size(), m_model.objects.size())); bool new_objects = false; for (size_t idx_model_object = 0; idx_model_object < model.objects.size(); ++ idx_model_object) { ModelObject &model_object = *m_model.objects[idx_model_object]; auto it_status = model_object_status.find(ModelObjectStatus(model_object.id())); assert(it_status != model_object_status.end()); assert(it_status->status != ModelObjectStatus::Deleted); if (it_status->status == ModelObjectStatus::New) // PrintObject instances will be added in the next loop. continue; // Update the ModelObject instance, possibly invalidate the linked PrintObjects. assert(it_status->status == ModelObjectStatus::Old || it_status->status == ModelObjectStatus::Moved); const ModelObject &model_object_new = *model.objects[idx_model_object]; auto it_print_object_status = print_object_status.lower_bound(PrintObjectStatus(model_object.id())); if (it_print_object_status != print_object_status.end() && it_print_object_status->id != model_object.id()) it_print_object_status = print_object_status.end(); // Check whether a model part volume was added or removed, their transformations or order changed. bool model_parts_differ = model_volume_list_changed(model_object, model_object_new, ModelVolumeType::MODEL_PART); bool sla_trafo_differs = model_object.instances.empty() != model_object_new.instances.empty() || (! model_object.instances.empty() && ! sla_trafo(model_object).isApprox(sla_trafo(model_object_new))); if (model_parts_differ || sla_trafo_differs) { // The very first step (the slicing step) is invalidated. One may freely remove all associated PrintObjects. if (it_print_object_status != print_object_status.end()) { update_apply_status(it_print_object_status->print_object->invalidate_all_steps()); const_cast(*it_print_object_status).status = PrintObjectStatus::Deleted; } // Copy content of the ModelObject including its ID, do not change the parent. model_object.assign_copy(model_object_new); } else { // Synchronize Object's config. bool object_config_changed = model_object.config != model_object_new.config; if (object_config_changed) model_object.config = model_object_new.config; if (! object_diff.empty() || object_config_changed) { SLAPrintObjectConfig new_config = m_default_object_config; normalize_and_apply_config(new_config, model_object.config); if (it_print_object_status != print_object_status.end()) { t_config_option_keys diff = it_print_object_status->print_object->config().diff(new_config); if (! diff.empty()) { update_apply_status(it_print_object_status->print_object->invalidate_state_by_config_options(diff)); it_print_object_status->print_object->config_apply_only(new_config, diff, true); } } } if (model_object.sla_support_points != model_object_new.sla_support_points) { model_object.sla_support_points = model_object_new.sla_support_points; if (it_print_object_status != print_object_status.end()) update_apply_status(it_print_object_status->print_object->invalidate_step(slaposSupportPoints)); } // Copy the ModelObject name, input_file and instances. The instances will compared against PrintObject instances in the next step. model_object.name = model_object_new.name; model_object.input_file = model_object_new.input_file; model_object.clear_instances(); model_object.instances.reserve(model_object_new.instances.size()); for (const ModelInstance *model_instance : model_object_new.instances) { model_object.instances.emplace_back(new ModelInstance(*model_instance)); model_object.instances.back()->set_model_object(&model_object); } } std::vector new_instances = sla_instances(model_object); if (it_print_object_status != print_object_status.end() && it_print_object_status->status != PrintObjectStatus::Deleted) { // The SLAPrintObject is already there. if (new_instances.empty()) { const_cast(*it_print_object_status).status = PrintObjectStatus::Deleted; } else { if (new_instances != it_print_object_status->print_object->instances()) { // Instances changed. it_print_object_status->print_object->set_instances(new_instances); update_apply_status(this->invalidate_step(slapsRasterize)); } print_objects_new.emplace_back(it_print_object_status->print_object); const_cast(*it_print_object_status).status = PrintObjectStatus::Reused; } } else if (! new_instances.empty()) { auto print_object = new SLAPrintObject(this, &model_object); // FIXME: this invalidates the transformed mesh in SLAPrintObject // which is expensive to calculate (especially the raw_mesh() call) print_object->set_trafo(sla_trafo(model_object)); print_object->set_instances(new_instances); print_object->config_apply(config, true); print_objects_new.emplace_back(print_object); new_objects = true; } } if (m_objects != print_objects_new) { this->call_cancel_callback(); update_apply_status(this->invalidate_all_steps()); m_objects = print_objects_new; // Delete the PrintObjects marked as Unknown or Deleted. bool deleted_objects = false; for (auto &pos : print_object_status) if (pos.status == PrintObjectStatus::Unknown || pos.status == PrintObjectStatus::Deleted) { update_apply_status(pos.print_object->invalidate_all_steps()); delete pos.print_object; deleted_objects = true; } if (new_objects) update_apply_status(false); } #ifdef _DEBUG check_model_ids_equal(m_model, model); #endif /* _DEBUG */ return static_cast(apply_status); } // After calling the apply() function, set_task() may be called to limit the task to be processed by process(). void SLAPrint::set_task(const TaskParams ¶ms) { // Grab the lock for the Print / PrintObject milestones. tbb::mutex::scoped_lock lock(this->state_mutex()); int n_object_steps = int(params.to_object_step) + 1; if (n_object_steps == 0) n_object_steps = (int)slaposCount; if (params.single_model_object.valid()) { SLAPrintObject *print_object = nullptr; size_t idx_print_object = 0; for (; idx_print_object < m_objects.size(); ++idx_print_object) if (m_objects[idx_print_object]->model_object()->id() == params.single_model_object) { print_object = m_objects[idx_print_object]; break; } assert(print_object != nullptr); bool shall_cancel = false; for (int istep = 0; istep < n_object_steps; ++istep) if (! print_object->m_stepmask[istep]) { shall_cancel = true; break; } bool running = false; if (!shall_cancel) { for (int istep = 0; istep < n_object_steps; ++ istep) if (print_object->is_step_started_unguarded(SLAPrintObjectStep(istep))) { running = true; break; } } if (!running) this->call_cancel_callback(); // Now the background process is either stopped, or it is inside one of the print object steps to be calculated anyway. if (params.single_model_instance_only) { // Suppress all the steps of other instances. for (SLAPrintObject *po : m_objects) for (int istep = 0; istep < (int)slaposCount; ++istep) po->m_stepmask[istep] = false; } else if (!running) { // Swap the print objects, so that the selected print_object is first in the row. // At this point the background processing must be stopped, so it is safe to shuffle print objects. if (idx_print_object != 0) std::swap(m_objects.front(), m_objects[idx_print_object]); } for (int istep = 0; istep < n_object_steps; ++ istep) print_object->m_stepmask[istep] = true; for (int istep = n_object_steps; istep < (int)slaposCount; ++istep) print_object->m_stepmask[istep] = false; } if (params.to_object_step != -1 || params.to_print_step != -1) { // Limit the print steps. size_t istep = (params.to_object_step != -1) ? 0 : size_t(params.to_print_step) + 1; for (; istep < m_stepmask.size(); ++ istep) m_stepmask[istep] = false; } } // Clean up after process() finished, either with success, error or if canceled. // The adjustments on the SLAPrint / SLAPrintObject data due to set_task() are to be reverted here. void SLAPrint::finalize() { for (SLAPrintObject *po : m_objects) for (int istep = 0; istep < (int)slaposCount; ++ istep) po->m_stepmask[istep] = true; for (int istep = 0; istep < (int)slapsCount; ++ istep) m_stepmask[istep] = true; } // Generate a recommended output file name based on the format template, default extension, and template parameters // (timestamps, object placeholders derived from the model, current placeholder prameters and print statistics. // Use the final print statistics if available, or just keep the print statistics placeholders if not available yet (before the output is finalized). std::string SLAPrint::output_filename() const { DynamicConfig config = this->finished() ? this->print_statistics().config() : this->print_statistics().placeholders(); return this->PrintBase::output_filename(m_print_config.output_filename_format.value, "zip", &config); } namespace { // Compile the argument for support creation from the static print config. sla::SupportConfig make_support_cfg(const SLAPrintObjectConfig& c) { sla::SupportConfig scfg; scfg.head_front_radius_mm = 0.5*c.support_head_front_diameter.getFloat(); scfg.head_back_radius_mm = 0.5*c.support_pillar_diameter.getFloat(); scfg.head_penetration_mm = c.support_head_penetration.getFloat(); scfg.head_width_mm = c.support_head_width.getFloat(); scfg.object_elevation_mm = c.support_object_elevation.getFloat(); scfg.tilt = c.support_critical_angle.getFloat() * PI / 180.0 ; scfg.max_bridge_length_mm = c.support_max_bridge_length.getFloat(); scfg.headless_pillar_radius_mm = 0.375*c.support_pillar_diameter.getFloat(); switch(c.support_pillar_connection_mode.getInt()) { case slapcmZigZag: scfg.pillar_connection_mode = sla::PillarConnectionMode::zigzag; break; case slapcmCross: scfg.pillar_connection_mode = sla::PillarConnectionMode::cross; break; case slapcmDynamic: scfg.pillar_connection_mode = sla::PillarConnectionMode::dynamic; break; } scfg.ground_facing_only = c.support_buildplate_only.getBool(); scfg.pillar_widening_factor = c.support_pillar_widening_factor.getFloat(); scfg.base_radius_mm = 0.5*c.support_base_diameter.getFloat(); scfg.base_height_mm = c.support_base_height.getFloat(); return scfg; } void swapXY(ExPolygon& expoly) { for(auto& p : expoly.contour.points) std::swap(p(X), p(Y)); for(auto& h : expoly.holes) for(auto& p : h.points) std::swap(p(X), p(Y)); } } std::vector SLAPrint::calculate_heights(const BoundingBoxf3& bb3d, float elevation, float initial_layer_height, float layer_height) const { std::vector heights; float minZ = float(bb3d.min(Z)) - float(elevation); float maxZ = float(bb3d.max(Z)); auto flh = float(layer_height); auto gnd = float(bb3d.min(Z)); for(float h = minZ + initial_layer_height; h < maxZ; h += flh) if(h >= gnd) heights.emplace_back(h); return heights; } template void report_status(SLAPrint& p, int st, const std::string& msg, Args&&...args) { BOOST_LOG_TRIVIAL(info) << st << "% " << msg; p.set_status(st, msg, std::forward(args)...); } void SLAPrint::process() { using namespace sla; using ExPolygon = Slic3r::ExPolygon; // Assumption: at this point the print objects should be populated only with // the model objects we have to process and the instances are also filtered // shortcut to initial layer height double ilhd = m_material_config.initial_layer_height.getFloat(); auto ilh = float(ilhd); const size_t objcount = m_objects.size(); const unsigned min_objstatus = 0; // where the per object operations start const unsigned max_objstatus = PRINT_STEP_LEVELS[slapsRasterize]; // where the per object operations end // the coefficient that multiplies the per object status values which // are set up for <0, 100>. They need to be scaled into the whole process const double ostepd = (max_objstatus - min_objstatus) / (objcount * 100.0); // The slicing will be performed on an imaginary 1D grid which starts from // the bottom of the bounding box created around the supported model. So // the first layer which is usually thicker will be part of the supports // not the model geometry. Exception is when the model is not in the air // (elevation is zero) and no pad creation was requested. In this case the // model geometry starts on the ground level and the initial layer is part // of it. In any case, the model and the supports have to be sliced in the // same imaginary grid (the height vector argument to TriangleMeshSlicer). // Slicing the model object. This method is oversimplified and needs to // be compared with the fff slicing algorithm for verification auto slice_model = [this, ilh](SLAPrintObject& po) { double lh = po.m_config.layer_height.getFloat(); TriangleMesh mesh = po.transformed_mesh(); TriangleMeshSlicer slicer(&mesh); // The 1D grid heights std::vector heights = calculate_heights(mesh.bounding_box(), float(po.get_elevation()), ilh, float(lh)); auto& layers = po.m_model_slices; layers.clear(); slicer.slice(heights, &layers, [this](){ throw_if_canceled(); }); }; // In this step we check the slices, identify island and cover them with // support points. Then we sprinkle the rest of the mesh. auto support_points = [this, ilh](SLAPrintObject& po) { const ModelObject& mo = *po.m_model_object; po.m_supportdata.reset( new SLAPrintObject::SupportData(po.transformed_mesh()) ); // If supports are disabled, we can skip the model scan. if(!po.m_config.supports_enable.getBool()) return; BOOST_LOG_TRIVIAL(debug) << "Support point count " << mo.sla_support_points.size(); // If there are no points on the front-end, we will do the // autoplacement. Otherwise we will just blindly copy the frontend data // into the backend cache. if(mo.sla_support_points.empty()) { // calculate heights of slices (slices are calculated already) double lh = po.m_config.layer_height.getFloat(); std::vector heights = calculate_heights(po.transformed_mesh().bounding_box(), float(po.get_elevation()), ilh, float(lh)); this->throw_if_canceled(); SLAAutoSupports::Config config; const SLAPrintObjectConfig& cfg = po.config(); config.density_relative = float(cfg.support_points_density_relative / 100.f); // the config value is in percents config.minimal_distance = float(cfg.support_points_minimal_distance); // Construction of this object does the calculation. this->throw_if_canceled(); SLAAutoSupports auto_supports(po.transformed_mesh(), po.m_supportdata->emesh, po.get_model_slices(), heights, config, [this]() { throw_if_canceled(); }); // Now let's extract the result. const std::vector& points = auto_supports.output(); this->throw_if_canceled(); po.m_supportdata->support_points = points; BOOST_LOG_TRIVIAL(debug) << "Automatic support points: " << po.m_supportdata->support_points.size(); // Using RELOAD_SLA_SUPPORT_POINTS to tell the Plater to pass the update status to GLGizmoSlaSupports report_status(*this, -1, L("Generating support points"), SlicingStatus::RELOAD_SLA_SUPPORT_POINTS); } else { // There are some points on the front-end, no calculation will be done. po.m_supportdata->support_points = po.transformed_support_points(); } }; // In this step we create the supports auto support_tree = [this, objcount, ostepd](SLAPrintObject& po) { if(!po.m_supportdata) return; if(!po.m_config.supports_enable.getBool()) { // Generate empty support tree. It can still host a pad po.m_supportdata->support_tree_ptr.reset(new SLASupportTree()); return; } try { sla::SupportConfig scfg = make_support_cfg(po.m_config); sla::Controller ctl; // some magic to scale the status values coming from the support // tree creation into the whole print process auto stfirst = OBJ_STEP_LEVELS.begin(); auto stthis = stfirst + slaposSupportTree; // we need to add up the status portions until this operation int init = std::accumulate(stfirst, stthis, 0); init = int(init * ostepd); // scale the init portion // scaling for the sub operations double d = *stthis / (objcount * 100.0); ctl.statuscb = [this, init, d](unsigned st, const std::string& msg) { //FIXME this status line scaling does not seem to be correct. // How does it account for an increasing object index? report_status(*this, int(init + st*d), msg); }; ctl.stopcondition = [this](){ return canceled(); }; ctl.cancelfn = [this]() { throw_if_canceled(); }; po.m_supportdata->support_tree_ptr.reset( new SLASupportTree(sla::to_point_set(po.m_supportdata->support_points), po.m_supportdata->emesh, scfg, ctl)); // Create the unified mesh auto rc = SlicingStatus::RELOAD_SCENE; // This is to prevent "Done." being displayed during merged_mesh() report_status(*this, -1, L("Visualizing supports")); po.m_supportdata->support_tree_ptr->merged_mesh(); BOOST_LOG_TRIVIAL(debug) << "Processed support point count " << po.m_supportdata->support_points.size(); // Check the mesh for later troubleshooting. if(po.support_mesh().empty()) BOOST_LOG_TRIVIAL(warning) << "Support mesh is empty"; report_status(*this, -1, L("Visualizing supports"), rc); } catch(sla::SLASupportsStoppedException&) { // no need to rethrow // throw_if_canceled(); } }; // This step generates the sla base pad auto base_pool = [this](SLAPrintObject& po) { // this step can only go after the support tree has been created // and before the supports had been sliced. (or the slicing has to be // repeated) if(!po.m_supportdata || !po.m_supportdata->support_tree_ptr) { BOOST_LOG_TRIVIAL(error) << "Uninitialized support data at " << "pad creation."; return; } if(po.m_config.pad_enable.getBool()) { double wt = po.m_config.pad_wall_thickness.getFloat(); double h = po.m_config.pad_wall_height.getFloat(); double md = po.m_config.pad_max_merge_distance.getFloat(); double er = po.m_config.pad_edge_radius.getFloat(); double lh = po.m_config.layer_height.getFloat(); double elevation = po.m_config.support_object_elevation.getFloat(); if(!po.m_config.supports_enable.getBool()) elevation = 0; sla::PoolConfig pcfg(wt, h, md, er); ExPolygons bp; double pad_h = sla::get_pad_fullheight(pcfg); auto&& trmesh = po.transformed_mesh(); // This call can get pretty time consuming auto thrfn = [this](){ throw_if_canceled(); }; if(elevation < pad_h) { // we have to count with the model geometry for the base plate sla::base_plate(trmesh, bp, float(pad_h), float(lh), thrfn); } pcfg.throw_on_cancel = thrfn; po.m_supportdata->support_tree_ptr->add_pad(bp, pcfg); } else { po.m_supportdata->support_tree_ptr->remove_pad(); } po.throw_if_canceled(); auto rc = SlicingStatus::RELOAD_SCENE; report_status(*this, -1, L("Visualizing supports"), rc); }; // Slicing the support geometries similarly to the model slicing procedure. // If the pad had been added previously (see step "base_pool" than it will // be part of the slices) auto slice_supports = [ilh](SLAPrintObject& po) { auto& sd = po.m_supportdata; if(sd && sd->support_tree_ptr) { auto lh = float(po.m_config.layer_height.getFloat()); sd->support_slices = sd->support_tree_ptr->slice(lh, ilh); } }; // We have the layer polygon collection but we need to unite them into // an index where the key is the height level in discrete levels (clipper) auto index_slices = [ilhd](SLAPrintObject& po) { po.m_slice_index.clear(); auto sih = LevelID(scale_(ilhd)); // Establish the slice grid boundaries auto bb = po.transformed_mesh().bounding_box(); double modelgnd = bb.min(Z); double elevation = po.get_elevation(); double lh = po.m_config.layer_height.getFloat(); double minZ = modelgnd - elevation; // scaled values: auto sminZ = LevelID(scale_(minZ)); auto smaxZ = LevelID(scale_(bb.max(Z))); auto smodelgnd = LevelID(scale_(modelgnd)); auto slh = LevelID(scale_(lh)); // It is important that the next levels match the levels in // model_slice method. Only difference is that here it works with // scaled coordinates po.m_level_ids.clear(); for(LevelID h = sminZ + sih; h < smaxZ; h += slh) if(h >= smodelgnd) po.m_level_ids.emplace_back(h); std::vector& oslices = po.m_model_slices; // If everything went well this code should not run at all, but // let's be robust... // assert(levelids.size() == oslices.size()); if(po.m_level_ids.size() < oslices.size()) { // extend the levels until... BOOST_LOG_TRIVIAL(warning) << "Height level mismatch at rasterization!\n"; LevelID lastlvl = po.m_level_ids.back(); while(po.m_level_ids.size() < oslices.size()) { lastlvl += slh; po.m_level_ids.emplace_back(lastlvl); } } for(size_t i = 0; i < oslices.size(); ++i) { LevelID h = po.m_level_ids[i]; float fh = float(double(h) * SCALING_FACTOR); // now for the public slice index: SLAPrintObject::SliceRecord& sr = po.m_slice_index[fh]; // There should be only one slice layer for each print object assert(sr.model_slices_idx == SLAPrintObject::SliceRecord::NONE); sr.model_slices_idx = i; } if(po.m_supportdata) { // deal with the support slices if present std::vector& sslices = po.m_supportdata->support_slices; po.m_supportdata->level_ids.clear(); po.m_supportdata->level_ids.reserve(sslices.size()); for(int i = 0; i < int(sslices.size()); ++i) { LevelID h = sminZ + sih + i * slh; po.m_supportdata->level_ids.emplace_back(h); float fh = float(double(h) * SCALING_FACTOR); SLAPrintObject::SliceRecord& sr = po.m_slice_index[fh]; assert(sr.support_slices_idx == SLAPrintObject::SliceRecord::NONE); sr.support_slices_idx = SLAPrintObject::SliceRecord::Idx(i); } } }; // Rasterizing the model objects, and their supports auto rasterize = [this, max_objstatus]() { if(canceled()) return; // clear the rasterizer input m_printer_input.clear(); for(SLAPrintObject * o : m_objects) { auto& po = *o; std::vector& oslices = po.m_model_slices; // We need to adjust the min Z level of the slices to be zero LevelID smfirst = po.m_supportdata && !po.m_supportdata->level_ids.empty() ? po.m_supportdata->level_ids.front() : 0; LevelID mfirst = po.m_level_ids.empty()? 0 : po.m_level_ids.front(); LevelID gndlvl = -(std::min(smfirst, mfirst)); // now merge this object's support and object slices with the rest // of the print object slices for(size_t i = 0; i < oslices.size(); ++i) { auto& lyrs = m_printer_input[gndlvl + po.m_level_ids[i]]; lyrs.emplace_back(oslices[i], po.m_instances); } if(!po.m_supportdata) continue; std::vector& sslices = po.m_supportdata->support_slices; for(size_t i = 0; i < sslices.size(); ++i) { LayerRefs& lyrs = m_printer_input[gndlvl + po.m_supportdata->level_ids[i]]; lyrs.emplace_back(sslices[i], po.m_instances); } } // collect all the keys std::vector keys; keys.reserve(m_printer_input.size()); for(auto& e : m_printer_input) keys.emplace_back(e.first); // If the raster has vertical orientation, we will flip the coordinates bool flpXY = m_printer_config.display_orientation.getInt() == SLADisplayOrientation::sladoPortrait; { // create a raster printer for the current print parameters // I don't know any better auto& ocfg = m_objects.front()->m_config; auto& matcfg = m_material_config; auto& printcfg = m_printer_config; double w = printcfg.display_width.getFloat(); double h = printcfg.display_height.getFloat(); auto pw = unsigned(printcfg.display_pixels_x.getInt()); auto ph = unsigned(printcfg.display_pixels_y.getInt()); double lh = ocfg.layer_height.getFloat(); double exp_t = matcfg.exposure_time.getFloat(); double iexp_t = matcfg.initial_exposure_time.getFloat(); if(flpXY) { std::swap(w, h); std::swap(pw, ph); } m_printer.reset(new SLAPrinter(w, h, pw, ph, lh, exp_t, iexp_t, flpXY? SLAPrinter::RO_PORTRAIT : SLAPrinter::RO_LANDSCAPE)); } // Allocate space for all the layers SLAPrinter& printer = *m_printer; auto lvlcnt = unsigned(m_printer_input.size()); printer.layers(lvlcnt); // slot is the portion of 100% that is realted to rasterization unsigned slot = PRINT_STEP_LEVELS[slapsRasterize]; // ist: initial state; pst: previous state unsigned ist = max_objstatus, pst = ist; // coefficient to map the rasterization state (0-99) to the allocated // portion (slot) of the process state double sd = (100 - ist) / 100.0; SpinMutex slck; // procedure to process one height level. This will run in parallel auto lvlfn = [this, &slck, &keys, &printer, slot, sd, ist, &pst, flpXY] (unsigned level_id) { if(canceled()) return; LayerRefs& lrange = m_printer_input[keys[level_id]]; // Switch to the appropriate layer in the printer printer.begin_layer(level_id); for(auto& lyrref : lrange) { // for all layers in the current level if(canceled()) break; const Layer& sl = lyrref.lref; // get the layer reference const LayerCopies& copies = lyrref.copies; // Draw all the polygons in the slice to the actual layer. for(auto& cp : copies) { for(ExPolygon slice : sl) { // The order is important here: // apply rotation before translation... slice.rotate(double(cp.rotation)); slice.translate(cp.shift(X), cp.shift(Y)); if(flpXY) swapXY(slice); printer.draw_polygon(slice, level_id); } } } // Finish the layer for later saving it. printer.finish_layer(level_id); // Status indication guarded with the spinlock auto st = ist + unsigned(sd*level_id*slot/m_printer_input.size()); { std::lock_guard lck(slck); if( st > pst) { report_status(*this, int(st), PRINT_STEP_LABELS[slapsRasterize]); pst = st; } } }; // last minute escape if(canceled()) return; // Sequential version (for testing) // for(unsigned l = 0; l < lvlcnt; ++l) process_level(l); // Print all the layers in parallel tbb::parallel_for(0, lvlcnt, lvlfn); // Fill statistics this->fill_statistics(); // Set statistics values to the printer m_printer->set_statistics({(m_print_statistics.objects_used_material + m_print_statistics.support_used_material)/1000, 10.0, double(m_print_statistics.slow_layers_count), double(m_print_statistics.fast_layers_count) }); }; using slaposFn = std::function; using slapsFn = std::function; std::array pobj_program = { slice_model, support_points, support_tree, base_pool, slice_supports, index_slices }; std::array print_program = { rasterize, [](){} // validate }; unsigned st = min_objstatus; unsigned incr = 0; BOOST_LOG_TRIVIAL(info) << "Start slicing process."; // TODO: this loop could run in parallel but should not exhaust all the CPU // power available // Calculate the support structures first before slicing the supports, so that the preview will get displayed ASAP for all objects. std::vector step_ranges = { slaposObjectSlice, slaposSliceSupports, slaposCount }; for (size_t idx_range = 0; idx_range + 1 < step_ranges.size(); ++ idx_range) { for(SLAPrintObject * po : m_objects) { BOOST_LOG_TRIVIAL(info) << "Slicing object " << po->model_object()->name; for (int s = (int)step_ranges[idx_range]; s < (int)step_ranges[idx_range + 1]; ++s) { auto currentstep = (SLAPrintObjectStep)s; // Cancellation checking. Each step will check for cancellation // on its own and return earlier gracefully. Just after it returns // execution gets to this point and throws the canceled signal. throw_if_canceled(); st += unsigned(incr * ostepd); if(po->m_stepmask[currentstep] && po->set_started(currentstep)) { report_status(*this, int(st), OBJ_STEP_LABELS[currentstep]); pobj_program[currentstep](*po); throw_if_canceled(); po->set_done(currentstep); } incr = OBJ_STEP_LEVELS[currentstep]; } } } std::array printsteps = { slapsRasterize, slapsValidate }; // this would disable the rasterization step // m_stepmask[slapsRasterize] = false; double pstd = (100 - max_objstatus) / 100.0; st = max_objstatus; for(size_t s = 0; s < print_program.size(); ++s) { auto currentstep = printsteps[s]; throw_if_canceled(); if(m_stepmask[currentstep] && set_started(currentstep)) { report_status(*this, int(st), PRINT_STEP_LABELS[currentstep]); print_program[currentstep](); throw_if_canceled(); set_done(currentstep); } st += unsigned(PRINT_STEP_LEVELS[currentstep] * pstd); } // If everything vent well report_status(*this, 100, L("Slicing done")); } bool SLAPrint::invalidate_state_by_config_options(const std::vector &opt_keys) { if (opt_keys.empty()) return false; // Cache the plenty of parameters, which influence the final rasterization only, // or they are only notes not influencing the rasterization step. static std::unordered_set steps_rasterize = { "exposure_time", "initial_exposure_time", "material_correction_printing", "material_correction_curing", "display_width", "display_height", "display_pixels_x", "display_pixels_y", "display_orientation", "printer_correction" }; static std::unordered_set steps_ignore = { "bed_shape", "max_print_height", "printer_technology", "output_filename_format", "fast_tilt_time", "slow_tilt_time", "area_fill" }; std::vector steps; std::vector osteps; bool invalidated = false; for (const t_config_option_key &opt_key : opt_keys) { if (steps_rasterize.find(opt_key) != steps_rasterize.end()) { // These options only affect the final rasterization, or they are just notes without influence on the output, // so there is nothing to invalidate. steps.emplace_back(slapsRasterize); } else if (steps_ignore.find(opt_key) != steps_ignore.end()) { // These steps have no influence on the output. Just ignore them. } else if (opt_key == "initial_layer_height") { steps.emplace_back(slapsRasterize); osteps.emplace_back(slaposObjectSlice); } else { // All values should be covered. assert(false); } } sort_remove_duplicates(steps); for (SLAPrintStep step : steps) invalidated |= this->invalidate_step(step); sort_remove_duplicates(osteps); for (SLAPrintObjectStep ostep : osteps) for (SLAPrintObject *object : m_objects) invalidated |= object->invalidate_step(ostep); return invalidated; } void SLAPrint::fill_statistics() { const double init_layer_height = m_material_config.initial_layer_height.getFloat(); const double layer_height = m_default_object_config.layer_height.getFloat(); const double area_fill = m_printer_config.area_fill.getFloat()*0.01;// 0.5 (50%); const double fast_tilt = m_printer_config.fast_tilt_time.getFloat();// 5.0; const double slow_tilt = m_printer_config.slow_tilt_time.getFloat();// 8.0; const double init_exp_time = m_material_config.initial_exposure_time.getFloat(); const double exp_time = m_material_config.exposure_time.getFloat(); const int fade_layers_cnt = m_default_object_config.faded_layers.getInt();// 10 // [3;20] const double width = m_printer_config.display_width.getFloat() / SCALING_FACTOR; const double height = m_printer_config.display_height.getFloat() / SCALING_FACTOR; const double display_area = width*height; // get polygons for all instances in the object auto get_all_polygons = [](const ExPolygons& input_polygons, const std::vector& instances) { const size_t inst_cnt = instances.size(); size_t polygon_cnt = 0; for (const ExPolygon& polygon : input_polygons) polygon_cnt += polygon.holes.size() + 1; Polygons polygons; polygons.reserve(polygon_cnt * inst_cnt); for (const ExPolygon& polygon : input_polygons) { for (size_t i = 0; i < inst_cnt; ++i) { ExPolygon tmp = polygon; tmp.rotate(Geometry::rad2deg(instances[i].rotation)); tmp.translate(instances[i].shift.x(), instances[i].shift.y()); polygons_append(polygons, to_polygons(std::move(tmp))); } } return polygons; }; double supports_volume = 0.0; double models_volume = 0.0; double estim_time = 0.0; size_t slow_layers = 0; size_t fast_layers = 0; // find highest object // Which is a better bet? To compare by max_z or by number of layers in the index? double max_z = 0.; size_t max_layers_cnt = 0; size_t highest_obj_idx = 0; for (SLAPrintObject *&po : m_objects) { const SLAPrintObject::SliceIndex& slice_index = po->get_slice_index(); if (! slice_index.empty()) { double z = (-- slice_index.end())->first; size_t cnt = slice_index.size(); //if (z > max_z) { if (cnt > max_layers_cnt) { max_layers_cnt = cnt; max_z = z; highest_obj_idx = &po - &m_objects.front(); } } } const SLAPrintObject * highest_obj = m_objects[highest_obj_idx]; const SLAPrintObject::SliceIndex& highest_obj_slice_index = highest_obj->get_slice_index(); const double delta_fade_time = (init_exp_time - exp_time) / (fade_layers_cnt + 1); double fade_layer_time = init_exp_time; int sliced_layer_cnt = 0; for (const auto& layer : highest_obj_slice_index) { const double l_height = (layer.first == highest_obj_slice_index.begin()->first) ? init_layer_height : layer_height; // Calculation of the consumed material Polygons model_polygons; Polygons supports_polygons; for (SLAPrintObject * po : m_objects) { const SLAPrintObject::SliceRecord *record = nullptr; { const SLAPrintObject::SliceIndex& index = po->get_slice_index(); auto key = layer.first; const SLAPrintObject::SliceIndex::const_iterator it_key = index.lower_bound(key - float(EPSILON)); if (it_key == index.end() || it_key->first > key + EPSILON) continue; record = &it_key->second; } if (record->model_slices_idx != SLAPrintObject::SliceRecord::NONE) append(model_polygons, get_all_polygons(po->get_model_slices()[record->model_slices_idx], po->instances())); if (record->support_slices_idx != SLAPrintObject::SliceRecord::NONE) append(supports_polygons, get_all_polygons(po->get_support_slices()[record->support_slices_idx], po->instances())); } model_polygons = union_(model_polygons); double layer_model_area = 0; for (const Polygon& polygon : model_polygons) layer_model_area += polygon.area(); if (layer_model_area != 0) models_volume += layer_model_area * l_height; if (!supports_polygons.empty() && !model_polygons.empty()) supports_polygons = diff(supports_polygons, model_polygons); double layer_support_area = 0; for (const Polygon& polygon : supports_polygons) layer_support_area += polygon.area(); if (layer_support_area != 0) supports_volume += layer_support_area * l_height; // Calculation of the slow and fast layers to the future controlling those values on FW const bool is_fast_layer = (layer_model_area + layer_support_area) <= display_area*area_fill; const double tilt_time = is_fast_layer ? fast_tilt : slow_tilt; if (is_fast_layer) fast_layers++; else slow_layers++; // Calculation of the printing time if (sliced_layer_cnt < 3) estim_time += init_exp_time; else if (fade_layer_time > exp_time) { fade_layer_time -= delta_fade_time; estim_time += fade_layer_time; } else estim_time += exp_time; estim_time += tilt_time; sliced_layer_cnt++; } m_print_statistics.support_used_material = supports_volume * SCALING_FACTOR * SCALING_FACTOR; m_print_statistics.objects_used_material = models_volume * SCALING_FACTOR * SCALING_FACTOR; // Estimated printing time // A layers count o the highest object if (max_layers_cnt == 0) m_print_statistics.estimated_print_time = "N/A"; else m_print_statistics.estimated_print_time = get_time_dhms(float(estim_time)); m_print_statistics.fast_layers_count = fast_layers; m_print_statistics.slow_layers_count = slow_layers; } // Returns true if an object step is done on all objects and there's at least one object. bool SLAPrint::is_step_done(SLAPrintObjectStep step) const { if (m_objects.empty()) return false; tbb::mutex::scoped_lock lock(this->state_mutex()); for (const SLAPrintObject *object : m_objects) if (! object->is_step_done_unguarded(step)) return false; return true; } SLAPrintObject::SLAPrintObject(SLAPrint *print, ModelObject *model_object): Inherited(print, model_object), m_stepmask(slaposCount, true), m_transformed_rmesh( [this](TriangleMesh& obj){ obj = m_model_object->raw_mesh(); obj.transform(m_trafo); }) { } SLAPrintObject::~SLAPrintObject() {} // Called by SLAPrint::apply_config(). // This method only accepts SLAPrintObjectConfig option keys. bool SLAPrintObject::invalidate_state_by_config_options(const std::vector &opt_keys) { if (opt_keys.empty()) return false; std::vector steps; bool invalidated = false; for (const t_config_option_key &opt_key : opt_keys) { if (opt_key == "layer_height") { steps.emplace_back(slaposObjectSlice); } else if ( opt_key == "supports_enable" || opt_key == "support_points_density_relative" || opt_key == "support_points_minimal_distance") { steps.emplace_back(slaposSupportPoints); } else if ( opt_key == "support_head_front_diameter" || opt_key == "support_head_penetration" || opt_key == "support_head_width" || opt_key == "support_pillar_diameter" || opt_key == "support_pillar_connection_mode" || opt_key == "support_buildplate_only" || opt_key == "support_base_diameter" || opt_key == "support_base_height" || opt_key == "support_critical_angle" || opt_key == "support_max_bridge_length" || opt_key == "support_object_elevation") { steps.emplace_back(slaposSupportTree); } else if ( opt_key == "pad_enable" || opt_key == "pad_wall_thickness" || opt_key == "pad_wall_height" || opt_key == "pad_max_merge_distance" || opt_key == "pad_edge_radius") { steps.emplace_back(slaposBasePool); } else { // All keys should be covered. assert(false); } } sort_remove_duplicates(steps); for (SLAPrintObjectStep step : steps) invalidated |= this->invalidate_step(step); return invalidated; } bool SLAPrintObject::invalidate_step(SLAPrintObjectStep step) { bool invalidated = Inherited::invalidate_step(step); // propagate to dependent steps if (step == slaposObjectSlice) { invalidated |= this->invalidate_all_steps(); } else if (step == slaposSupportPoints) { invalidated |= this->invalidate_steps({ slaposSupportTree, slaposBasePool, slaposSliceSupports, slaposIndexSlices }); invalidated |= m_print->invalidate_step(slapsRasterize); } else if (step == slaposSupportTree) { invalidated |= this->invalidate_steps({ slaposBasePool, slaposSliceSupports, slaposIndexSlices }); invalidated |= m_print->invalidate_step(slapsRasterize); } else if (step == slaposBasePool) { invalidated |= this->invalidate_steps({slaposSliceSupports, slaposIndexSlices}); invalidated |= m_print->invalidate_step(slapsRasterize); } else if (step == slaposSliceSupports) { invalidated |= this->invalidate_step(slaposIndexSlices); invalidated |= m_print->invalidate_step(slapsRasterize); } else if(step == slaposIndexSlices) { invalidated |= m_print->invalidate_step(slapsRasterize); } return invalidated; } bool SLAPrintObject::invalidate_all_steps() { return Inherited::invalidate_all_steps() | m_print->invalidate_all_steps(); } double SLAPrintObject::get_elevation() const { bool se = m_config.supports_enable.getBool(); double ret = se? m_config.support_object_elevation.getFloat() : 0; // if the pad is enabled, then half of the pad height is its base plate if(m_config.pad_enable.getBool()) { // Normally the elevation for the pad itself would be the thickness of // its walls but currently it is half of its thickness. Whatever it // will be in the future, we provide the config to the get_pad_elevation // method and we will have the correct value sla::PoolConfig pcfg; pcfg.min_wall_height_mm = m_config.pad_wall_height.getFloat(); pcfg.min_wall_thickness_mm = m_config.pad_wall_thickness.getFloat(); pcfg.edge_radius_mm = m_config.pad_edge_radius.getFloat(); pcfg.max_merge_distance_mm = m_config.pad_max_merge_distance.getFloat(); ret += sla::get_pad_elevation(pcfg); } return ret; } double SLAPrintObject::get_current_elevation() const { bool se = m_config.supports_enable.getBool(); bool has_supports = is_step_done(slaposSupportTree); bool has_pad = is_step_done(slaposBasePool); if(!has_supports && !has_pad) return 0; else if(has_supports && !has_pad) return se ? m_config.support_object_elevation.getFloat() : 0; return get_elevation(); } namespace { // dummy empty static containers for return values in some methods const std::vector EMPTY_SLICES; const TriangleMesh EMPTY_MESH; } const std::vector& SLAPrintObject::get_support_points() const { return m_supportdata->support_points; } const std::vector &SLAPrintObject::get_support_slices() const { // assert(is_step_done(slaposSliceSupports)); if (!m_supportdata) return EMPTY_SLICES; return m_supportdata->support_slices; } const SLAPrintObject::SliceIndex &SLAPrintObject::get_slice_index() const { // assert(is_step_done(slaposIndexSlices)); return m_slice_index; } const std::vector &SLAPrintObject::get_model_slices() const { // assert(is_step_done(slaposObjectSlice)); return m_model_slices; } bool SLAPrintObject::has_mesh(SLAPrintObjectStep step) const { switch (step) { case slaposSupportTree: return ! this->support_mesh().empty(); case slaposBasePool: return ! this->pad_mesh().empty(); default: return false; } } TriangleMesh SLAPrintObject::get_mesh(SLAPrintObjectStep step) const { switch (step) { case slaposSupportTree: return this->support_mesh(); case slaposBasePool: return this->pad_mesh(); default: return TriangleMesh(); } } const TriangleMesh& SLAPrintObject::support_mesh() const { if(m_config.supports_enable.getBool() && m_supportdata && m_supportdata->support_tree_ptr) { return m_supportdata->support_tree_ptr->merged_mesh(); } return EMPTY_MESH; } const TriangleMesh& SLAPrintObject::pad_mesh() const { if(m_config.pad_enable.getBool() && m_supportdata && m_supportdata->support_tree_ptr) return m_supportdata->support_tree_ptr->get_pad(); return EMPTY_MESH; } const TriangleMesh &SLAPrintObject::transformed_mesh() const { // we need to transform the raw mesh... // currently all the instances share the same x and y rotation and scaling // so we have to extract those from e.g. the first instance and apply to the // raw mesh. This is also true for the support points. // BUT: when the support structure is spawned for each instance than it has // to omit the X, Y rotation and scaling as those have been already applied // or apply an inverse transformation on the support structure after it // has been created. return m_transformed_rmesh.get(); } std::vector SLAPrintObject::transformed_support_points() const { assert(m_model_object != nullptr); std::vector& spts = m_model_object->sla_support_points; // this could be cached as well std::vector ret; ret.reserve(spts.size()); for(sla::SupportPoint& sp : spts) { Vec3d transformed_pos = trafo() * Vec3d(sp.pos(0), sp.pos(1), sp.pos(2)); ret.emplace_back(transformed_pos(0), transformed_pos(1), transformed_pos(2), sp.head_front_radius, sp.is_new_island); } return ret; } DynamicConfig SLAPrintStatistics::config() const { DynamicConfig config; const std::string print_time = Slic3r::short_time(this->estimated_print_time); config.set_key_value("print_time", new ConfigOptionString(print_time)); config.set_key_value("objects_used_material", new ConfigOptionFloat(this->objects_used_material)); config.set_key_value("support_used_material", new ConfigOptionFloat(this->support_used_material)); config.set_key_value("total_cost", new ConfigOptionFloat(this->total_cost)); config.set_key_value("total_weight", new ConfigOptionFloat(this->total_weight)); return config; } DynamicConfig SLAPrintStatistics::placeholders() { DynamicConfig config; for (const std::string &key : { "print_time", "total_cost", "total_weight", "objects_used_material", "support_used_material" }) config.set_key_value(key, new ConfigOptionString(std::string("{") + key + "}")); return config; } std::string SLAPrintStatistics::finalize_output_path(const std::string &path_in) const { std::string final_path; try { boost::filesystem::path path(path_in); DynamicConfig cfg = this->config(); PlaceholderParser pp; std::string new_stem = pp.process(path.stem().string(), 0, &cfg); final_path = (path.parent_path() / (new_stem + path.extension().string())).string(); } catch (const std::exception &ex) { BOOST_LOG_TRIVIAL(error) << "Failed to apply the print statistics to the export file name: " << ex.what(); final_path = path_in; } return final_path; } } // namespace Slic3r