zone_filler.cpp

Go to the documentation of this file.

/*
 * This program source code file is part of KiCad, a free EDA CAD application.
 *
 * Copyright (C) 2014-2017 CERN
 * Copyright The KiCad Developers, see AUTHORS.txt for contributors.
 * @author Tomasz Włostowski <[email protected]>
 *
 * This program is free software: you can redistribute it and/or modify it
 * under the terms of the GNU General Public License as published by the
 * Free Software Foundation, either version 3 of the License, or (at your
 * option) any later version.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with this program; if not, you may find one here:
 * http://www.gnu.org/licenses/old-licenses/gpl-2.0.html
 * or you may search the http://www.gnu.org website for the version 2 license,
 * or you may write to the Free Software Foundation, Inc.,
 * 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301, USA
 */
 
#include <atomic>
#include <future>
#include <hash.h>
#include <set>
#include <unordered_map>
#include <unordered_set>
#include <core/kicad_algo.h>
#include <advanced_config.h>
#include <board.h>
#include <board_design_settings.h>
#include <drc/drc_engine.h>
#include <zone.h>
#include <footprint.h>
#include <pad.h>
#include <pcb_target.h>
#include <pcb_track.h>
#include <pcb_text.h>
#include <pcb_textbox.h>
#include <pcb_tablecell.h>
#include <pcb_table.h>
#include <pcb_dimension.h>
#include <connectivity/connectivity_data.h>
#include <convert_basic_shapes_to_polygon.h>
#include <board_commit.h>
#include <progress_reporter.h>
#include <geometry/shape_poly_set.h>
#include <geometry/convex_hull.h>
#include <geometry/geometry_utils.h>
#include <geometry/vertex_set.h>
#include <geometry/poly_ystripes_index.h>
#include <kidialog.h>
#include <thread_pool.h>
#include <math/util.h>      // for KiROUND
#include "zone_filler.h"
#include "project.h"
#include "project/project_local_settings.h"
#include "pcb_barcode.h"
 
// Helper classes for connect_nearby_polys
class RESULTS
{
public:
    RESULTS( int aOutline1, int aOutline2, int aVertex1, int aVertex2 ) :
            m_outline1( aOutline1 ), m_outline2( aOutline2 ),
            m_vertex1( aVertex1 ), m_vertex2( aVertex2 )
    {
    }
 
    bool operator<( const RESULTS& aOther ) const
    {
        if( m_outline1 != aOther.m_outline1 )
            return m_outline1 < aOther.m_outline1;
        if( m_outline2 != aOther.m_outline2 )
            return m_outline2 < aOther.m_outline2;
        if( m_vertex1 != aOther.m_vertex1 )
            return m_vertex1 < aOther.m_vertex1;
        return m_vertex2 < aOther.m_vertex2;
    }
 
    int m_outline1;
    int m_outline2;
    int m_vertex1;
    int m_vertex2;
};
 
class VERTEX_CONNECTOR : protected VERTEX_SET
{
public:
    VERTEX_CONNECTOR( const BOX2I& aBBox, const SHAPE_POLY_SET& aPolys, int aDist ) : VERTEX_SET( 0 )
    {
        SetBoundingBox( aBBox );
        VERTEX* tail = nullptr;
 
        for( int i = 0; i < aPolys.OutlineCount(); i++ )
            tail = createList( aPolys.Outline( i ), tail, (void*)( intptr_t )( i ) );
 
        if( tail )
            tail->updateList();
        m_dist = aDist;
    }
 
    VERTEX* getPoint( VERTEX* aPt ) const
    {
        // z-order range for the current point ± limit bounding box
        const uint32_t     maxZ = zOrder( aPt->x + m_dist, aPt->y + m_dist );
        const uint32_t     minZ = zOrder( aPt->x - m_dist, aPt->y - m_dist );
        const SEG::ecoord limit2 = SEG::Square( m_dist );
 
        // first look for points in increasing z-order
        SEG::ecoord min_dist = std::numeric_limits<SEG::ecoord>::max();
        VERTEX* retval = nullptr;
 
        auto check_pt = [&]( VERTEX* p )
        {
            VECTOR2D diff( p->x - aPt->x, p->y - aPt->y );
            SEG::ecoord dist2 = diff.SquaredEuclideanNorm();
 
            if( dist2 > 0 && dist2 < limit2 && dist2 < min_dist && p->isEar( true ) )
            {
                min_dist = dist2;
                retval = p;
            }
        };
 
        VERTEX* p = aPt->nextZ;
 
        while( p && p->z <= maxZ )
        {
            check_pt( p );
            p = p->nextZ;
        }
 
        p = aPt->prevZ;
 
        while( p && p->z >= minZ )
        {
            check_pt( p );
            p = p->prevZ;
        }
 
        return retval;
    }
 
    void FindResults()
    {
        if( m_vertices.empty() )
            return;
 
        VERTEX* p = m_vertices.front().next;
        std::set<VERTEX*> visited;
 
        while( p != &m_vertices.front() )
        {
            // Skip points that are concave
            if( !p->isEar() )
            {
                p = p->next;
                continue;
            }
 
            VERTEX* q = nullptr;
 
            if( ( visited.empty() || !visited.contains( p ) ) && ( q = getPoint( p ) ) )
            {
                visited.insert( p );
 
                if( !visited.contains( q ) &&
                    m_results.emplace( (intptr_t) p->GetUserData(), (intptr_t) q->GetUserData(),
                                        p->i, q->i ).second )
                {
                    // We don't want to connect multiple points in the same vicinity, so skip
                    // 2 points before and after each point and match.
                    visited.insert( p->prev );
                    visited.insert( p->prev->prev );
                    visited.insert( p->next );
                    visited.insert( p->next->next );
 
                    visited.insert( q->prev );
                    visited.insert( q->prev->prev );
                    visited.insert( q->next );
                    visited.insert( q->next->next );
 
                    visited.insert( q );
                }
            }
 
            p = p->next;
        }
    }
 
    std::set<RESULTS> GetResults() const
    {
        return m_results;
    }
 
private:
    std::set<RESULTS> m_results;
    int m_dist;
};
 

namespace
{

struct PAD_KNOCKOUT_KEY
{
    VECTOR2I  position;
    VECTOR2I  effectiveSize;   // For circular: max of drill and pad; otherwise pad size
    int       shape;           // PAD_SHAPE enum value
    EDA_ANGLE orientation;
    int       netCode;
 
    bool operator==( const PAD_KNOCKOUT_KEY& other ) const
    {
        return position == other.position && effectiveSize == other.effectiveSize
               && shape == other.shape && orientation == other.orientation
               && netCode == other.netCode;
    }
};
 
struct PAD_KNOCKOUT_KEY_HASH
{
    size_t operator()( const PAD_KNOCKOUT_KEY& key ) const
    {
        return hash_val( key.position.x, key.position.y, key.effectiveSize.x, key.effectiveSize.y,
                         key.shape, key.orientation.AsDegrees(), key.netCode );
    }
};

struct VIA_KNOCKOUT_KEY
{
    VECTOR2I position;
    int      effectiveSize;    // max of drill and via width
    int      netCode;
 
    bool operator==( const VIA_KNOCKOUT_KEY& other ) const
    {
        return position == other.position && effectiveSize == other.effectiveSize
               && netCode == other.netCode;
    }
};
 
struct VIA_KNOCKOUT_KEY_HASH
{
    size_t operator()( const VIA_KNOCKOUT_KEY& key ) const
    {
        return hash_val( key.position.x, key.position.y, key.effectiveSize, key.netCode );
    }
};

struct TRACK_KNOCKOUT_KEY
{
    VECTOR2I start;
    VECTOR2I end;
    int      width;
 
    TRACK_KNOCKOUT_KEY( const VECTOR2I& aStart, const VECTOR2I& aEnd, int aWidth ) :
            width( aWidth )
    {
        // Canonicalize endpoint order for consistent hashing
        if( aStart.x < aEnd.x || ( aStart.x == aEnd.x && aStart.y <= aEnd.y ) )
        {
            start = aStart;
            end = aEnd;
        }
        else
        {
            start = aEnd;
            end = aStart;
        }
    }
 
    bool operator==( const TRACK_KNOCKOUT_KEY& other ) const
    {
        return start == other.start && end == other.end && width == other.width;
    }
};
 
struct TRACK_KNOCKOUT_KEY_HASH
{
    size_t operator()( const TRACK_KNOCKOUT_KEY& key ) const
    {
        return hash_val( key.start.x, key.start.y, key.end.x, key.end.y, key.width );
    }
};
 
template<typename Func>
void forEachBoardAndFootprintZone( BOARD* aBoard, Func&& aFunc )
{
    for( ZONE* zone : aBoard->Zones() )
        aFunc( zone );
 
    for( FOOTPRINT* footprint : aBoard->Footprints() )
    {
        for( ZONE* zone : footprint->Zones() )
            aFunc( zone );
    }
}
 
bool isZoneFillKeepout( const ZONE* aZone, PCB_LAYER_ID aLayer, const BOX2I& aBBox )
{
    return aZone->GetIsRuleArea()
           && aZone->HasKeepoutParametersSet()
           && aZone->GetDoNotAllowZoneFills()
           && aZone->IsOnLayer( aLayer )
           && aZone->GetBoundingBox().Intersects( aBBox );
}
 
void appendZoneOutlineWithoutArcs( const ZONE* aZone, SHAPE_POLY_SET& aPolys )
{
    if( aZone->Outline()->ArcCount() == 0 )
    {
        aPolys.Append( *aZone->Outline() );
        return;
    }
 
    SHAPE_POLY_SET outline( *aZone->Outline() );
    outline.ClearArcs();
    aPolys.Append( outline );
}
 
} // anonymous namespace
 
 
ZONE_FILLER::ZONE_FILLER( BOARD* aBoard, COMMIT* aCommit ) :
        m_board( aBoard ),
        m_brdOutlinesValid( false ),
        m_commit( aCommit ),
        m_progressReporter( nullptr ),
        m_worstClearance( 0 )
{
    m_maxError = aBoard->GetDesignSettings().m_MaxError;
 
    // To enable add "DebugZoneFiller=1" to kicad_advanced settings file.
    m_debugZoneFiller = ADVANCED_CFG::GetCfg().m_DebugZoneFiller;
}
 
 
ZONE_FILLER::~ZONE_FILLER()
{
}
 
 
void ZONE_FILLER::SetProgressReporter( PROGRESS_REPORTER* aReporter )
{
    m_progressReporter = aReporter;
}
 

bool ZONE_FILLER::Fill( const std::vector<ZONE*>& aZones, bool aCheck, wxWindow* aParent )
{
    std::lock_guard<KISPINLOCK> lock( m_board->GetConnectivity()->GetLock() );
 
    std::vector<std::pair<ZONE*, PCB_LAYER_ID>>               toFill;
    std::map<std::pair<ZONE*, PCB_LAYER_ID>, HASH_128>        oldFillHashes;
    std::map<ZONE*, std::map<PCB_LAYER_ID, ISOLATED_ISLANDS>> isolatedIslandsMap;
 
    std::shared_ptr<CONNECTIVITY_DATA> connectivity = m_board->GetConnectivity();
 
    // Ensure that multiple threads don't attempt to initialize the advanced cfg global at the same
    // time.
    ADVANCED_CFG::GetCfg();
 
    // Rebuild (from scratch, ignoring dirty flags) just in case. This really needs to be reliable.
    connectivity->ClearRatsnest();
    connectivity->Build( m_board, m_progressReporter );
 
    m_worstClearance = m_board->GetMaxClearanceValue();
 
    if( m_progressReporter )
    {
        m_progressReporter->Report( aCheck ? _( "Checking zone fills..." )
                                           : _( "Building zone fills..." ) );
        m_progressReporter->SetMaxProgress( aZones.size() );
        m_progressReporter->KeepRefreshing();
    }
 
    // The board outlines is used to clip solid areas inside the board (when outlines are valid)
    m_boardOutline.RemoveAllContours();
    m_brdOutlinesValid = m_board->GetBoardPolygonOutlines( m_boardOutline, true );
 
    // Update and cache zone bounding boxes and pad effective shapes so that we don't have to
    // make them thread-safe.
    //
    for( ZONE* zone : m_board->Zones() )
        zone->CacheBoundingBox();
 
    for( FOOTPRINT* footprint : m_board->Footprints() )
    {
        for( PAD* pad : footprint->Pads() )
        {
            if( pad->IsDirty() )
            {
                pad->BuildEffectiveShapes();
                pad->BuildEffectivePolygon( ERROR_OUTSIDE );
            }
        }
 
        for( ZONE* zone : footprint->Zones() )
            zone->CacheBoundingBox();
 
        // Rules may depend on insideCourtyard() or other expressions
        footprint->BuildCourtyardCaches();
        footprint->BuildNetTieCache();
    }
 
    LSET boardCuMask = LSET::AllCuMask( m_board->GetCopperLayerCount() );
 
    // Pre-build Y-stripe spatial indices for zone outline containment queries.
    // Amortizes build cost across the thousands of via/pad flash checks below.
    std::unordered_map<const ZONE*, POLY_YSTRIPES_INDEX> zoneOutlineIndices;
 
    for( ZONE* zone : m_board->Zones() )
    {
        if( zone->GetNumCorners() <= 2 )
            continue;
 
        zoneOutlineIndices[zone].Build( *zone->Outline() );
    }
 
    auto findHighestPriorityZone =
            [&]( const BOX2I& bbox, PCB_LAYER_ID itemLayer, int netcode,
                 const std::function<bool( const ZONE* )>& testFn ) -> ZONE*
            {
                unsigned highestPriority = 0;
                ZONE*    highestPriorityZone = nullptr;
 
                for( ZONE* zone : m_board->Zones() )
                {
                    // Rule areas are not filled
                    if( zone->GetIsRuleArea() )
                        continue;
 
                    if( zone->GetAssignedPriority() < highestPriority )
                        continue;
 
                    if( !zone->IsOnLayer( itemLayer ) )
                        continue;
 
                    // Degenerate zones will cause trouble; skip them
                    if( zone->GetNumCorners() <= 2 )
                        continue;
 
                    if( !zone->GetBoundingBox().Intersects( bbox ) )
                        continue;
 
                    if( !testFn( zone ) )
                        continue;
 
                    // Prefer highest priority and matching netcode
                    if( zone->GetAssignedPriority() > highestPriority
                            || zone->GetNetCode() == netcode )
                    {
                        highestPriority = zone->GetAssignedPriority();
                        highestPriorityZone = zone;
                    }
                }
 
                return highestPriorityZone;
            };
 
    auto isInPourKeepoutArea =
            [&]( const BOX2I& bbox, PCB_LAYER_ID itemLayer, const VECTOR2I& testPoint ) -> bool
            {
                for( ZONE* zone : m_board->Zones() )
                {
                    if( !zone->GetIsRuleArea() )
                        continue;
 
                    if( !zone->HasKeepoutParametersSet() )
                        continue;
 
                    if( !zone->GetDoNotAllowZoneFills() )
                        continue;
 
                    if( !zone->IsOnLayer( itemLayer ) )
                        continue;
 
                    // Degenerate zones will cause trouble; skip them
                    if( zone->GetNumCorners() <= 2 )
                        continue;
 
                    if( !zone->GetBoundingBox().Intersects( bbox ) )
                        continue;
 
                    auto it = zoneOutlineIndices.find( zone );
 
                    if( it != zoneOutlineIndices.end() && it->second.Contains( testPoint ) )
                        return true;
                }
 
                return false;
            };
 
    // Determine state of conditional via flashing
    // This is now done completely deterministically prior to filling due to the pathological
    // case presented in https://gitlab.com/kicad/code/kicad/-/issues/12964.
    for( PCB_TRACK* track : m_board->Tracks() )
    {
        if( track->Type() == PCB_VIA_T )
        {
            PCB_VIA*  via = static_cast<PCB_VIA*>( track );
            PADSTACK& padstack = via->Padstack();
 
            via->ClearZoneLayerOverrides();
 
            if( !via->GetRemoveUnconnected() )
                continue;
 
            BOX2I    bbox = via->GetBoundingBox();
            VECTOR2I center = via->GetPosition();
            int      holeRadius = via->GetDrillValue() / 2 + 1;
            int      netcode = via->GetNetCode();
            LSET     layers = via->GetLayerSet() & boardCuMask;
 
            // Checking if the via hole touches the zone outline
            auto viaTestFn =
                    [&]( const ZONE* aZone ) -> bool
                    {
                        return aZone->Outline()->Contains( center, -1, holeRadius );
                    };
 
            for( PCB_LAYER_ID layer : layers )
            {
                if( !via->ConditionallyFlashed( layer ) )
                    continue;
 
                if( isInPourKeepoutArea( bbox, layer, center ) )
                {
                    via->SetZoneLayerOverride( layer, ZLO_FORCE_NO_ZONE_CONNECTION );
                }
                else
                {
                    ZONE* zone = findHighestPriorityZone( bbox, layer, netcode, viaTestFn );
 
                    if( zone && zone->GetNetCode() == via->GetNetCode()
                             && ( padstack.UnconnectedLayerMode() != UNCONNECTED_LAYER_MODE::START_END_ONLY
                                  || layer == padstack.Drill().start
                                  || layer == padstack.Drill().end ) )
                    {
                        via->SetZoneLayerOverride( layer, ZLO_FORCE_FLASHED );
                    }
                    else
                    {
                        via->SetZoneLayerOverride( layer, ZLO_FORCE_NO_ZONE_CONNECTION );
                    }
                }
            }
        }
    }
 
    // Determine state of conditional pad flashing
    for( FOOTPRINT* footprint : m_board->Footprints() )
    {
        for( PAD* pad : footprint->Pads() )
        {
            pad->ClearZoneLayerOverrides();
 
            if( !pad->GetRemoveUnconnected() )
                continue;
 
            BOX2I    bbox = pad->GetBoundingBox();
            VECTOR2I center = pad->GetPosition();
            int      netcode = pad->GetNetCode();
            LSET     layers = pad->GetLayerSet() & boardCuMask;
 
            auto padTestFn =
                    [&]( const ZONE* aZone ) -> bool
                    {
                        auto it = zoneOutlineIndices.find( aZone );
 
                        if( it != zoneOutlineIndices.end() )
                            return it->second.Contains( center );
 
                        return aZone->Outline()->Contains( center );
                    };
 
            for( PCB_LAYER_ID layer : layers )
            {
                if( !pad->ConditionallyFlashed( layer ) )
                    continue;
 
                if( isInPourKeepoutArea( bbox, layer, center ) )
                {
                    pad->SetZoneLayerOverride( layer, ZLO_FORCE_NO_ZONE_CONNECTION );
                }
                else
                {
                    ZONE* zone = findHighestPriorityZone( bbox, layer, netcode, padTestFn );
 
                    if( zone && zone->GetNetCode() == pad->GetNetCode() )
                        pad->SetZoneLayerOverride( layer, ZLO_FORCE_FLASHED );
                    else
                        pad->SetZoneLayerOverride( layer, ZLO_FORCE_NO_ZONE_CONNECTION );
                }
            }
        }
    }
 
    for( ZONE* zone : aZones )
    {
        // Rule areas are not filled
        if( zone->GetIsRuleArea() )
            continue;
 
        // Degenerate zones will cause trouble; skip them
        if( zone->GetNumCorners() <= 2 )
            continue;
 
        if( m_commit )
            m_commit->Modify( zone );
 
        // calculate the hash value for filled areas. it will be used later to know if the
        // current filled areas are up to date
        for( PCB_LAYER_ID layer : zone->GetLayerSet() )
        {
            zone->BuildHashValue( layer );
            oldFillHashes[ { zone, layer } ] = zone->GetHashValue( layer );
 
            // Add the zone to the list of zones to test or refill
            toFill.emplace_back( std::make_pair( zone, layer ) );
 
            isolatedIslandsMap[zone][layer] = ISOLATED_ISLANDS();
        }
 
        // Remove existing fill first to prevent drawing invalid polygons on some platforms
        zone->UnFill();
    }
 
    auto zone_fill_dependency =
            [&]( ZONE* aZone, PCB_LAYER_ID aLayer, ZONE* aOtherZone,
                 bool aRequireCompletedOtherFill ) -> bool
            {
                // Check to see if we have to knock-out the filled areas of a higher-priority
                // zone.  If so we have to wait until said zone is filled before we can fill.
 
                // If the other zone is already filled on the requested layer then we're
                // good-to-go
                if( aRequireCompletedOtherFill && aOtherZone->GetFillFlag( aLayer ) )
                    return false;
 
                // Even if keepouts exclude copper pours, the exclusion is by outline rather than
                // filled area, so we're good-to-go here too
                if( aOtherZone->GetIsRuleArea() )
                    return false;
 
                // If the other zone is never going to be filled then don't wait for it
                if( aOtherZone->GetNumCorners() <= 2 )
                    return false;
 
                // If the zones share no common layers
                if( !aOtherZone->GetLayerSet().test( aLayer ) )
                    return false;
 
                if( aZone->HigherPriority( aOtherZone ) )
                    return false;
 
                // Same-net zones always use outlines to produce determinate results
                if( aOtherZone->SameNet( aZone ) )
                    return false;
 
                // A higher priority zone is found: if we intersect and it's not filled yet
                // then we have to wait.
                BOX2I inflatedBBox = aZone->GetBoundingBox();
                inflatedBBox.Inflate( m_worstClearance );
 
                if( !inflatedBBox.Intersects( aOtherZone->GetBoundingBox() ) )
                    return false;
 
                return aZone->Outline()->Collide( aOtherZone->Outline(), m_worstClearance );
            };
 
    auto check_fill_dependency =
            [&]( ZONE* aZone, PCB_LAYER_ID aLayer, ZONE* aOtherZone ) -> bool
            {
                return zone_fill_dependency( aZone, aLayer, aOtherZone, true );
            };
 
    auto fill_item_dependency =
            [&]( const std::pair<ZONE*, PCB_LAYER_ID>& aWaiter,
                 const std::pair<ZONE*, PCB_LAYER_ID>& aDependency ) -> bool
            {
                if( aWaiter.first == aDependency.first || aWaiter.second != aDependency.second )
                    return false;
 
                return check_fill_dependency( aWaiter.first, aWaiter.second, aDependency.first );
            };
 
    auto fill_lambda =
            [&]( std::pair<ZONE*, PCB_LAYER_ID> aFillItem ) -> int
            {
                if( m_progressReporter && m_progressReporter->IsCancelled() )
                    return 0;
 
                PCB_LAYER_ID layer = aFillItem.second;
                ZONE*        zone = aFillItem.first;
 
                SHAPE_POLY_SET fillPolys;
 
                if( !fillSingleZone( zone, layer, fillPolys ) )
                    return 0;
 
                zone->SetFilledPolysList( layer, fillPolys );
 
                if( m_progressReporter )
                    m_progressReporter->AdvanceProgress();
 
                return 1;
            };
 
    auto tesselate_lambda =
            [&]( std::pair<ZONE*, PCB_LAYER_ID> aFillItem ) -> int
            {
                if( m_progressReporter && m_progressReporter->IsCancelled() )
                    return 0;
 
                PCB_LAYER_ID layer = aFillItem.second;
                ZONE*        zone = aFillItem.first;
 
                zone->CacheTriangulation( layer );
                zone->SetFillFlag( layer, true );
 
                return 1;
            };
 
    thread_pool&      tp = GetKiCadThreadPool();
    std::atomic<bool> cancelled = false;
 
    auto waitForFutures =
            [&]( std::vector<std::future<int>>& aFutures, std::vector<int>* aResults = nullptr )
            {
                if( aResults )
                    aResults->clear();
 
                for( auto& future : aFutures )
                {
                    while( future.wait_for( std::chrono::milliseconds( 100 ) )
                            != std::future_status::ready )
                    {
                        if( m_progressReporter )
                        {
                            m_progressReporter->KeepRefreshing();
 
                            if( m_progressReporter->IsCancelled() )
                                cancelled = true;
                        }
                    }
 
                    int result = future.get();
 
                    if( aResults )
                        aResults->push_back( result );
                }
            };
 
    struct FILL_DAG
    {
        std::vector<std::vector<size_t>> successors;
        std::vector<int>                 inDegree;
        std::vector<size_t>              currentWave;
    };
 
    auto build_fill_dag =
            [&]( const std::vector<std::pair<ZONE*, PCB_LAYER_ID>>& aFillItems,
                 auto&& aHasDependency ) -> FILL_DAG
            {
                FILL_DAG dag;
 
                dag.successors.resize( aFillItems.size() );
                dag.inDegree.assign( aFillItems.size(), 0 );
                dag.currentWave.reserve( aFillItems.size() );
 
                for( size_t i = 0; i < aFillItems.size(); ++i )
                {
                    for( size_t j = 0; j < aFillItems.size(); ++j )
                    {
                        if( i == j )
                            continue;
 
                        if( aHasDependency( aFillItems[j], aFillItems[i] ) )
                        {
                            dag.successors[i].push_back( j );
                            dag.inDegree[j]++;
                        }
                    }
                }
 
                for( size_t i = 0; i < aFillItems.size(); ++i )
                {
                    if( dag.inDegree[i] == 0 )
                        dag.currentWave.push_back( i );
                }
 
                return dag;
            };
 
    auto run_fill_waves =
            [&]( const std::vector<std::pair<ZONE*, PCB_LAYER_ID>>& aFillItems, auto&& aFillFn,
                 auto&& aTessFn, auto&& aHasDependency )
            {
                FILL_DAG dag = build_fill_dag( aFillItems, aHasDependency );
 
                while( !dag.currentWave.empty() && !cancelled.load() )
                {
                    std::vector<std::future<int>> fillFutures;
                    std::vector<int>              fillResults;
 
                    fillFutures.reserve( dag.currentWave.size() );
 
                    for( size_t idx : dag.currentWave )
                    {
                        fillFutures.emplace_back( tp.submit_task(
                                [&aFillFn, &aFillItems, idx]()
                                {
                                    return aFillFn( aFillItems[idx] );
                                } ) );
                    }
 
                    waitForFutures( fillFutures, &fillResults );
 
                    std::vector<std::future<int>> tessFutures;
 
                    tessFutures.reserve( dag.currentWave.size() );
 
                    for( size_t ii = 0; ii < fillResults.size(); ++ii )
                    {
                        if( fillResults[ii] == 0 )
                            continue;
 
                        size_t idx = dag.currentWave[ii];
 
                        tessFutures.emplace_back( tp.submit_task(
                                [&aTessFn, &aFillItems, idx]()
                                {
                                    return aTessFn( aFillItems[idx] );
                                } ) );
                    }
 
                    waitForFutures( tessFutures );
 
                    if( cancelled.load() )
                        break;
 
                    std::vector<size_t> nextWave;
 
                    for( size_t idx : dag.currentWave )
                    {
                        for( size_t succ : dag.successors[idx] )
                        {
                            if( --dag.inDegree[succ] == 0 )
                                nextWave.push_back( succ );
                        }
                    }
 
                    dag.currentWave = std::move( nextWave );
                }
            };
 
    run_fill_waves( toFill, fill_lambda, tesselate_lambda, fill_item_dependency );
 
    // Now update the connectivity to check for isolated copper islands
    // (NB: FindIsolatedCopperIslands() is multi-threaded)
    if( m_progressReporter )
    {
        if( m_progressReporter->IsCancelled() )
            return false;
 
        m_progressReporter->AdvancePhase();
        m_progressReporter->Report( _( "Removing isolated copper islands..." ) );
        m_progressReporter->KeepRefreshing();
    }
 
    connectivity->SetProgressReporter( m_progressReporter );
    connectivity->FillIsolatedIslandsMap( isolatedIslandsMap );
    connectivity->SetProgressReporter( nullptr );
 
    if( m_progressReporter && m_progressReporter->IsCancelled() )
        return false;
 
    for( ZONE* zone : aZones )
    {
        // Keepout zones are not filled
        if( zone->GetIsRuleArea() )
            continue;
 
        zone->SetIsFilled( true );
    }
 
    // Now remove isolated copper islands according to the isolated islands strategy assigned
    // by the user (always, never, below-certain-size).
    //
    // Track zones that had islands removed for potential iterative refill
    std::set<ZONE*> zonesWithRemovedIslands;
 
    for( const auto& [ zone, zoneIslands ] : isolatedIslandsMap )
    {
        // If *all* the polygons are islands, do not remove any of them
        bool allIslands = true;
 
        for( const auto& [ layer, layerIslands ] : zoneIslands )
        {
            if( layerIslands.m_IsolatedOutlines.size()
                    != static_cast<size_t>( zone->GetFilledPolysList( layer )->OutlineCount() ) )
            {
                allIslands = false;
                break;
            }
        }
 
        if( allIslands )
            continue;
 
        for( const auto& [ layer, layerIslands ] : zoneIslands )
        {
            if( m_debugZoneFiller && LSET::InternalCuMask().Contains( layer ) )
                continue;
 
            if( layerIslands.m_IsolatedOutlines.empty() )
                continue;
 
            std::vector<int> islands = layerIslands.m_IsolatedOutlines;
 
            // The list of polygons to delete must be explored from last to first in list,
            // to allow deleting a polygon from list without breaking the remaining of the list
            std::sort( islands.begin(), islands.end(), std::greater<int>() );
 
            std::shared_ptr<SHAPE_POLY_SET> poly = zone->GetFilledPolysList( layer );
            long long int                   minArea = zone->GetMinIslandArea();
            ISLAND_REMOVAL_MODE             mode = zone->GetIslandRemovalMode();
 
            for( int idx : islands )
            {
                SHAPE_LINE_CHAIN& outline = poly->Outline( idx );
 
                if( mode == ISLAND_REMOVAL_MODE::ALWAYS )
                {
                    poly->DeletePolygonAndTriangulationData( idx, false );
                    zonesWithRemovedIslands.insert( zone );
                }
                else if ( mode == ISLAND_REMOVAL_MODE::AREA && outline.Area( true ) < minArea )
                {
                    poly->DeletePolygonAndTriangulationData( idx, false );
                    zonesWithRemovedIslands.insert( zone );
                }
                else
                {
                    zone->SetIsIsland( layer, idx );
                }
            }
 
            poly->UpdateTriangulationDataHash();
            zone->CalculateFilledArea();
 
            if( m_progressReporter && m_progressReporter->IsCancelled() )
                return false;
        }
    }
 
    // Iterative refill: If islands were removed from higher-priority zones, lower-priority zones
    // may need to be refilled to occupy the now-available space (issue 21746).
    const bool iterativeRefill = ADVANCED_CFG::GetCfg().m_ZoneFillIterativeRefill;
 
    if( iterativeRefill && !zonesWithRemovedIslands.empty() )
    {
        // Find lower-priority zones that may need refilling.
        // A zone needs refilling if it overlaps with a zone that had islands removed
        // and has lower priority than that zone.
        std::vector<std::pair<ZONE*, PCB_LAYER_ID>> zonesToRefill;
 
        for( ZONE* zoneWithIsland : zonesWithRemovedIslands )
        {
            BOX2I islandZoneBBox = zoneWithIsland->GetBoundingBox();
            islandZoneBBox.Inflate( m_worstClearance );
 
            for( ZONE* zone : aZones )
            {
                // Skip the zone that had islands removed
                if( zone == zoneWithIsland )
                    continue;
 
                // Skip keepout zones
                if( zone->GetIsRuleArea() )
                    continue;
 
                // Only refill zones with lower priority than the zone that had islands removed
                if( !zoneWithIsland->HigherPriority( zone ) )
                    continue;
 
                // Check for layer overlap
                LSET commonLayers = zone->GetLayerSet() & zoneWithIsland->GetLayerSet();
 
                if( commonLayers.none() )
                    continue;
 
                // Check for bounding box overlap
                if( !zone->GetBoundingBox().Intersects( islandZoneBBox ) )
                    continue;
 
                // Add zone/layer pairs for refilling
                for( PCB_LAYER_ID layer : commonLayers )
                {
                    auto fillItem = std::make_pair( zone, layer );
 
                    if( std::find( zonesToRefill.begin(), zonesToRefill.end(), fillItem ) == zonesToRefill.end() )
                    {
                        zonesToRefill.push_back( fillItem );
                    }
                }
            }
        }
 
        if( !zonesToRefill.empty() )
        {
            if( m_progressReporter )
            {
                m_progressReporter->AdvancePhase();
                m_progressReporter->Report( _( "Refilling zones after island removal..." ) );
                m_progressReporter->KeepRefreshing();
            }
 
            // Refill using cached pre-knockout fills - much faster than full refill
            // since we only need to re-apply the higher-priority zone knockout
            auto cached_refill_fill_lambda =
                    [&]( const std::pair<ZONE*, PCB_LAYER_ID>& aFillItem ) -> int
                    {
                        ZONE*          zone = aFillItem.first;
                        PCB_LAYER_ID   layer = aFillItem.second;
                        SHAPE_POLY_SET fillPolys;
 
                        if( !refillZoneFromCache( zone, layer, fillPolys ) )
                            return 0;
 
                        zone->SetFilledPolysList( layer, fillPolys );
                        zone->SetFillFlag( layer, false );
                        return 1;
                    };
 
            auto cached_refill_tessellate_lambda =
                    [&]( const std::pair<ZONE*, PCB_LAYER_ID>& aFillItem ) -> int
                    {
                        ZONE*        zone = aFillItem.first;
                        PCB_LAYER_ID layer = aFillItem.second;
 
                        zone->CacheTriangulation( layer );
                        zone->SetFillFlag( layer, true );
                        return 1;
                    };
 
            auto refill_item_dependency =
                    [&]( const std::pair<ZONE*, PCB_LAYER_ID>& aWaiter,
                         const std::pair<ZONE*, PCB_LAYER_ID>& aDependency ) -> bool
                    {
                        if( aWaiter.first == aDependency.first || aWaiter.second != aDependency.second )
                            return false;
 
                        return zone_fill_dependency( aWaiter.first, aWaiter.second,
                                                     aDependency.first, false );
                    };
 
            run_fill_waves( zonesToRefill, cached_refill_fill_lambda,
                            cached_refill_tessellate_lambda, refill_item_dependency );
 
            // Re-run island detection for refilled zones
            std::map<ZONE*, std::map<PCB_LAYER_ID, ISOLATED_ISLANDS>> refillIslandsMap;
            std::set<ZONE*> refillZones;
 
            for( const auto& [zone, layer] : zonesToRefill )
                refillZones.insert( zone );
 
            for( ZONE* zone : refillZones )
            {
                refillIslandsMap[zone] = std::map<PCB_LAYER_ID, ISOLATED_ISLANDS>();
 
                for( PCB_LAYER_ID layer : zone->GetLayerSet() )
                    refillIslandsMap[zone][layer] = ISOLATED_ISLANDS();
            }
 
            connectivity->FillIsolatedIslandsMap( refillIslandsMap );
 
            // Remove islands from refilled zones
            for( const auto& [ zone, zoneIslands ] : refillIslandsMap )
            {
                bool allIslands = true;
 
                for( const auto& [ layer, layerIslands ] : zoneIslands )
                {
                    if( layerIslands.m_IsolatedOutlines.size()
                            != static_cast<size_t>( zone->GetFilledPolysList( layer )->OutlineCount() ) )
                    {
                        allIslands = false;
                        break;
                    }
                }
 
                if( allIslands )
                    continue;
 
                for( const auto& [ layer, layerIslands ] : zoneIslands )
                {
                    if( m_debugZoneFiller && LSET::InternalCuMask().Contains( layer ) )
                        continue;
 
                    if( layerIslands.m_IsolatedOutlines.empty() )
                        continue;
 
                    std::vector<int> islands = layerIslands.m_IsolatedOutlines;
                    std::sort( islands.begin(), islands.end(), std::greater<int>() );
 
                    std::shared_ptr<SHAPE_POLY_SET> poly = zone->GetFilledPolysList( layer );
                    long long int                   minArea = zone->GetMinIslandArea();
                    ISLAND_REMOVAL_MODE             mode = zone->GetIslandRemovalMode();
 
                    for( int idx : islands )
                    {
                        SHAPE_LINE_CHAIN& outline = poly->Outline( idx );
 
                        if( mode == ISLAND_REMOVAL_MODE::ALWAYS )
                            poly->DeletePolygonAndTriangulationData( idx, false );
                        else if( mode == ISLAND_REMOVAL_MODE::AREA && outline.Area( true ) < minArea )
                            poly->DeletePolygonAndTriangulationData( idx, false );
                        else
                            zone->SetIsIsland( layer, idx );
                    }
 
                    poly->UpdateTriangulationDataHash();
                    zone->CalculateFilledArea();
                }
            }
 
        }
    }
 
    // Now remove islands which are either outside the board edge or fail to meet the minimum
    // area requirements
    using island_check_return = std::vector<std::pair<std::shared_ptr<SHAPE_POLY_SET>, int>>;
 
    std::vector<std::pair<std::shared_ptr<SHAPE_POLY_SET>, double>> polys_to_check;
 
    // rough estimate to save re-allocation time
    polys_to_check.reserve( m_board->GetCopperLayerCount() * aZones.size() );
 
    for( ZONE* zone : aZones )
    {
        // Don't check for connections on layers that only exist in the zone but
        // were disabled in the board
        BOARD* board = zone->GetBoard();
        LSET zoneCopperLayers = zone->GetLayerSet() & LSET::AllCuMask( board->GetCopperLayerCount() );
 
        // Min-thickness is the web thickness.  On the other hand, a blob min-thickness by
        // min-thickness is not useful.  Since there's no obvious definition of web vs. blob, we
        // arbitrarily choose "at least 3X the area".
        double minArea = (double) zone->GetMinThickness() * zone->GetMinThickness() * 3;
 
        for( PCB_LAYER_ID layer : zoneCopperLayers )
        {
            if( m_debugZoneFiller && LSET::InternalCuMask().Contains( layer ) )
                continue;
 
            polys_to_check.emplace_back( zone->GetFilledPolysList( layer ), minArea );
        }
    }
 
    auto island_lambda =
            [&]( int aStart, int aEnd ) -> island_check_return
            {
                island_check_return retval;
 
                for( int ii = aStart; ii < aEnd && !cancelled.load(); ++ii )
                {
                    auto [poly, minArea] = polys_to_check[ii];
 
                    for( int jj = poly->OutlineCount() - 1; jj >= 0; jj-- )
                    {
                        SHAPE_POLY_SET island;
                        SHAPE_POLY_SET intersection;
                        const SHAPE_LINE_CHAIN& test_poly = poly->Polygon( jj ).front();
                        double island_area = test_poly.Area();
 
                        if( island_area < minArea )
                            continue;
 
 
                        island.AddOutline( test_poly );
                        intersection.BooleanIntersection( m_boardOutline, island );
 
                        // Nominally, all of these areas should be either inside or outside the
                        // board outline.  So this test should be able to just compare areas (if
                        // they are equal, you are inside).  But in practice, we sometimes have
                        // slight overlap at the edges, so testing against half-size area acts as
                        // a fail-safe.
                        if( intersection.Area() < island_area / 2.0 )
                            retval.emplace_back( poly, jj );
                    }
                }
 
                return retval;
            };
 
    auto island_returns = tp.submit_blocks( 0, polys_to_check.size(), island_lambda );
    cancelled = false;
 
    // Allow island removal threads to finish
    for( size_t ii = 0; ii < island_returns.size(); ++ii )
    {
        std::future<island_check_return>& ret = island_returns[ii];
 
        if( ret.valid() )
        {
            std::future_status status = ret.wait_for( std::chrono::seconds( 0 ) );
 
            while( status != std::future_status::ready )
            {
                if( m_progressReporter )
                {
                    m_progressReporter->KeepRefreshing();
 
                    if( m_progressReporter->IsCancelled() )
                        cancelled = true;
                }
 
                status = ret.wait_for( std::chrono::milliseconds( 100 ) );
            }
        }
    }
 
    if( cancelled.load() )
        return false;
 
    for( size_t ii = 0; ii < island_returns.size(); ++ii )
    {
        std::future<island_check_return>& ret = island_returns[ii];
 
        if( ret.valid() )
        {
            for( auto& action_item : ret.get() )
                action_item.first->DeletePolygonAndTriangulationData( action_item.second, true );
        }
    }
 
    for( ZONE* zone : aZones )
        zone->CalculateFilledArea();
 
    // Second pass: Re-evaluate via flashing based on actual filled polygons.
    // The first pass (before filling) marks vias as ZLO_FORCE_FLASHED if they're within the
    // zone outline. However, if the fill doesn't actually reach the via (due to obstacles like
    // tracks), we should not flash the via. See https://gitlab.com/kicad/code/kicad/-/issues/22010
    //
    // Build a spatial index per filled zone-layer for O(log V) containment queries instead of
    // O(V) ray-casting. This is critical for boards with large zone fills (many vertices) and
    // many vias/pads.
    struct INDEXED_ZONE
    {
        BOX2I                                       bbox;
        std::unique_ptr<POLY_YSTRIPES_INDEX>        index;
    };
 
    struct NET_LAYER_HASH
    {
        size_t operator()( const std::pair<int, PCB_LAYER_ID>& k ) const
        {
            return std::hash<int>()( k.first ) ^ ( std::hash<int>()( k.second ) << 16 );
        }
    };
 
    std::unordered_map<std::pair<int, PCB_LAYER_ID>, std::vector<INDEXED_ZONE>, NET_LAYER_HASH>
            filledZonesByNetLayer;
 
    for( ZONE* zone : m_board->Zones() )
    {
        if( zone->GetIsRuleArea() )
            continue;
 
        for( PCB_LAYER_ID layer : zone->GetLayerSet() )
        {
            if( !zone->HasFilledPolysForLayer( layer ) )
                continue;
 
            const std::shared_ptr<SHAPE_POLY_SET>& fill = zone->GetFilledPolysList( layer );
 
            if( fill->IsEmpty() )
                continue;
 
            INDEXED_ZONE iz;
            iz.bbox = fill->BBox();
            iz.index = std::make_unique<POLY_YSTRIPES_INDEX>();
            iz.index->Build( *fill );
            filledZonesByNetLayer[{ zone->GetNetCode(), layer }].push_back( std::move( iz ) );
        }
    }
 
    auto zoneReachesPoint =
            [&]( int aNetcode, PCB_LAYER_ID aLayer, const VECTOR2I& aCenter, int aRadius ) -> bool
            {
                auto it = filledZonesByNetLayer.find( { aNetcode, aLayer } );
 
                if( it == filledZonesByNetLayer.end() )
                    return false;
 
                for( const INDEXED_ZONE& iz : it->second )
                {
                    if( !iz.bbox.GetInflated( aRadius ).Contains( aCenter ) )
                        continue;
 
                    if( iz.index->Contains( aCenter, aRadius ) )
                        return true;
                }
 
                return false;
            };
 
    for( PCB_TRACK* track : m_board->Tracks() )
    {
        if( track->Type() != PCB_VIA_T )
            continue;
 
        PCB_VIA*  via = static_cast<PCB_VIA*>( track );
        VECTOR2I  center = via->GetPosition();
        int       holeRadius = via->GetDrillValue() / 2;
        int       netcode = via->GetNetCode();
        LSET      layers = via->GetLayerSet() & boardCuMask;
 
        for( PCB_LAYER_ID layer : layers )
        {
            if( via->GetZoneLayerOverride( layer ) != ZLO_FORCE_FLASHED )
                continue;
 
            if( !zoneReachesPoint( netcode, layer, center, holeRadius ) )
                via->SetZoneLayerOverride( layer, ZLO_FORCE_NO_ZONE_CONNECTION );
        }
    }
 
    for( FOOTPRINT* footprint : m_board->Footprints() )
    {
        for( PAD* pad : footprint->Pads() )
        {
            VECTOR2I center = pad->GetPosition();
            int      netcode = pad->GetNetCode();
            LSET     layers = pad->GetLayerSet() & boardCuMask;
 
            int holeRadius = 0;
 
            if( pad->HasHole() )
                holeRadius = std::min( pad->GetDrillSizeX(), pad->GetDrillSizeY() ) / 2;
 
            for( PCB_LAYER_ID layer : layers )
            {
                if( pad->GetZoneLayerOverride( layer ) != ZLO_FORCE_FLASHED )
                    continue;
 
                if( !zoneReachesPoint( netcode, layer, center, holeRadius ) )
                    pad->SetZoneLayerOverride( layer, ZLO_FORCE_NO_ZONE_CONNECTION );
            }
        }
    }
 
    if( aCheck )
    {
        bool outOfDate = false;
 
        for( ZONE* zone : aZones )
        {
            // Keepout zones are not filled
            if( zone->GetIsRuleArea() )
                continue;
 
            for( PCB_LAYER_ID layer : zone->GetLayerSet() )
            {
                zone->BuildHashValue( layer );
 
                if( oldFillHashes[ { zone, layer } ] != zone->GetHashValue( layer ) )
                    outOfDate = true;
            }
        }
 
        if( ( m_board->GetProject()
              && m_board->GetProject()->GetLocalSettings().m_PrototypeZoneFill ) )
        {
            KIDIALOG dlg( aParent, _( "Prototype zone fill enabled. Disable setting and refill?" ), _( "Confirmation" ),
                          wxOK | wxCANCEL | wxICON_WARNING );
            dlg.SetOKCancelLabels( _( "Disable and refill" ), _( "Continue without Refill" ) );
            dlg.DoNotShowCheckbox( __FILE__, __LINE__ );
 
            if( dlg.ShowModal() == wxID_OK )
            {
                m_board->GetProject()->GetLocalSettings().m_PrototypeZoneFill = false;
            }
            else if( !outOfDate )
            {
                return false;
            }
        }
 
        if( outOfDate )
        {
            KIDIALOG dlg( aParent, _( "Zone fills are out-of-date. Refill?" ), _( "Confirmation" ),
                          wxOK | wxCANCEL | wxICON_WARNING );
            dlg.SetOKCancelLabels( _( "Refill" ), _( "Continue without Refill" ) );
            dlg.DoNotShowCheckbox( __FILE__, __LINE__ );
 
            if( dlg.ShowModal() == wxID_CANCEL )
                return false;
        }
        else
        {
            // No need to commit something that hasn't changed (and committing will set
            // the modified flag).
            return false;
        }
    }
 
    if( m_progressReporter )
    {
        if( m_progressReporter->IsCancelled() )
            return false;
 
        m_progressReporter->AdvancePhase();
        m_progressReporter->KeepRefreshing();
    }
 
    return true;
}
 

void ZONE_FILLER::addKnockout( BOARD_ITEM* aItem, PCB_LAYER_ID aLayer, int aGap, SHAPE_POLY_SET& aHoles )
{
    if( aItem->Type() == PCB_PAD_T && static_cast<PAD*>( aItem )->GetShape( aLayer ) == PAD_SHAPE::CUSTOM )
    {
        PAD* pad = static_cast<PAD*>( aItem );
        SHAPE_POLY_SET poly;
        pad->TransformShapeToPolygon( poly, aLayer, aGap, m_maxError, ERROR_OUTSIDE );
 
        // the pad shape in zone can be its convex hull or the shape itself
        if( pad->GetCustomShapeInZoneOpt() == CUSTOM_SHAPE_ZONE_MODE::CONVEXHULL )
        {
            std::vector<VECTOR2I> convex_hull;
            BuildConvexHull( convex_hull, poly );
 
            aHoles.NewOutline();
 
            for( const VECTOR2I& pt : convex_hull )
                aHoles.Append( pt );
        }
        else
        {
            aHoles.Append( poly );
        }
    }
    else
    {
        aItem->TransformShapeToPolygon( aHoles, aLayer, aGap, m_maxError, ERROR_OUTSIDE );
    }
}
 

void ZONE_FILLER::addHoleKnockout( PAD* aPad, int aGap, SHAPE_POLY_SET& aHoles )
{
    aPad->TransformHoleToPolygon( aHoles, aGap, m_maxError, ERROR_OUTSIDE );
}
 
 

void ZONE_FILLER::addKnockout( BOARD_ITEM* aItem, PCB_LAYER_ID aLayer, int aGap,
                               bool aIgnoreLineWidth, SHAPE_POLY_SET& aHoles )
{
    switch( aItem->Type() )
    {
    case PCB_FIELD_T:
    case PCB_TEXT_T:
    {
        PCB_TEXT* text = static_cast<PCB_TEXT*>( aItem );
 
        if( text->IsVisible() )
        {
            if( text->IsKnockout() )
            {
                // Knockout text should only leave holes where the text is, not where the copper fill
                // around it would be.
                PCB_TEXT textCopy = *text;
                textCopy.SetIsKnockout( false );
                textCopy.TransformTextToPolySet( aHoles, 0, m_maxError, ERROR_INSIDE );
            }
            else
            {
                text->TransformShapeToPolygon( aHoles, aLayer, aGap, m_maxError, ERROR_OUTSIDE );
            }
        }
 
        break;
    }
 
    case PCB_TEXTBOX_T:
    case PCB_TABLE_T:
    case PCB_SHAPE_T:
    case PCB_TARGET_T:
        aItem->TransformShapeToPolygon( aHoles, aLayer, aGap, m_maxError, ERROR_OUTSIDE, aIgnoreLineWidth );
        break;
 
    case PCB_BARCODE_T:
    {
        PCB_BARCODE* barcode = static_cast<PCB_BARCODE*>( aItem );
        barcode->GetBoundingHull( aHoles, aLayer, aGap, m_maxError, ERROR_OUTSIDE );
        break;
    }
 
    case PCB_DIM_ALIGNED_T:
    case PCB_DIM_LEADER_T:
    case PCB_DIM_CENTER_T:
    case PCB_DIM_RADIAL_T:
    case PCB_DIM_ORTHOGONAL_T:
    {
        PCB_DIMENSION_BASE* dim = static_cast<PCB_DIMENSION_BASE*>( aItem );
 
        dim->TransformShapeToPolygon( aHoles, aLayer, aGap, m_maxError, ERROR_OUTSIDE, false );
        dim->PCB_TEXT::TransformShapeToPolygon( aHoles, aLayer, aGap, m_maxError, ERROR_OUTSIDE );
        break;
    }
 
    default:
        break;
    }
}
 

void ZONE_FILLER::knockoutThermalReliefs( const ZONE* aZone, PCB_LAYER_ID aLayer,
                                          SHAPE_POLY_SET& aFill,
                                          std::vector<BOARD_ITEM*>& aThermalConnectionPads,
                                          std::vector<PAD*>& aNoConnectionPads )
{
    BOARD_DESIGN_SETTINGS& bds = m_board->GetDesignSettings();
    ZONE_CONNECTION        connection;
    DRC_CONSTRAINT         constraint;
    int                    padClearance;
    std::shared_ptr<SHAPE> padShape;
    int                    holeClearance;
    SHAPE_POLY_SET         holes;
 
    // Deduplication sets for coincident pads and vias
    std::unordered_set<PAD_KNOCKOUT_KEY, PAD_KNOCKOUT_KEY_HASH> processedPads;
    std::unordered_set<VIA_KNOCKOUT_KEY, VIA_KNOCKOUT_KEY_HASH> processedVias;
 
    for( FOOTPRINT* footprint : m_board->Footprints() )
    {
        for( PAD* pad : footprint->Pads() )
        {
            // NPTH pads with a drill hole affect all copper layers even when they carry no copper
            // on that layer (e.g. layers limited to "*.Mask"). The physical hole still requires
            // a clearance knockout, so skip only pads that are truly irrelevant to this layer.
            bool npthWithHole = pad->GetAttribute() == PAD_ATTRIB::NPTH
                                && pad->GetDrillSize().x > 0;
 
            if( !pad->IsOnLayer( aLayer ) && !npthWithHole )
                continue;
 
            BOX2I padBBox = pad->GetBoundingBox();
            padBBox.Inflate( m_worstClearance );
 
            if( !padBBox.Intersects( aZone->GetBoundingBox() ) )
                continue;
 
            // Deduplicate coincident pads (skip custom pads - they have complex shapes)
            PAD_SHAPE padShapeType = pad->GetShape( aLayer );
 
            if( padShapeType != PAD_SHAPE::CUSTOM )
            {
                // For circular pads: use max of drill and pad size; otherwise just pad size
                VECTOR2I padSize = pad->GetSize( aLayer );
                VECTOR2I effectiveSize;
 
                if( padShapeType == PAD_SHAPE::CIRCLE )
                {
                    int drill = std::max( pad->GetDrillSize().x, pad->GetDrillSize().y );
                    int maxDim = std::max( { padSize.x, padSize.y, drill } );
                    effectiveSize = VECTOR2I( maxDim, maxDim );
                }
                else
                {
                    effectiveSize = padSize;
                }
 
                PAD_KNOCKOUT_KEY padKey{ pad->GetPosition(), effectiveSize,
                                         static_cast<int>( padShapeType ),
                                         pad->GetOrientation(), pad->GetNetCode() };
 
                if( !processedPads.insert( padKey ).second )
                    continue;
            }
 
            bool noConnection = pad->GetNetCode() != aZone->GetNetCode();
 
            if( !aZone->IsTeardropArea() )
            {
                if( aZone->GetNetCode() == 0
                    || pad->GetZoneLayerOverride( aLayer ) == ZLO_FORCE_NO_ZONE_CONNECTION )
                {
                    noConnection = true;
                }
            }
 
            // Check if the pad is backdrilled or post-machined on this layer
            if( pad->IsBackdrilledOrPostMachined( aLayer ) )
                noConnection = true;
 
            if( noConnection )
            {
                // collect these for knockout in buildCopperItemClearances()
                aNoConnectionPads.push_back( pad );
                continue;
            }
 
            // For hatch zones, respect the zone connection type just like solid zones
            // Pads with THERMAL connection get thermal rings; FULL connections get no knockout;
            // NONE connections get handled later in buildCopperItemClearances.
            if( aZone->GetFillMode() == ZONE_FILL_MODE::HATCH_PATTERN )
            {
                constraint = bds.m_DRCEngine->EvalZoneConnection( pad, aZone, aLayer );
                connection = constraint.m_ZoneConnection;
 
                if( connection == ZONE_CONNECTION::THERMAL && !pad->CanFlashLayer( aLayer ) )
                    connection = ZONE_CONNECTION::NONE;
 
                switch( connection )
                {
                case ZONE_CONNECTION::THERMAL:
                {
                    padShape = pad->GetEffectiveShape( aLayer, FLASHING::ALWAYS_FLASHED );
 
                    if( aFill.Collide( padShape.get(), 0 ) )
                    {
                        // Get the thermal relief gap
                        constraint = bds.m_DRCEngine->EvalRules( THERMAL_RELIEF_GAP_CONSTRAINT, pad,
                                                                  aZone, aLayer );
                        int thermalGap = constraint.GetValue().Min();
 
                        // Knock out the thermal gap only - the thermal ring will be added separately
                        aThermalConnectionPads.push_back( pad );
                        addKnockout( pad, aLayer, thermalGap, holes );
                    }
 
                    break;
                }
 
                case ZONE_CONNECTION::NONE:
                    // Will be handled by buildCopperItemClearances
                    aNoConnectionPads.push_back( pad );
                    break;
 
                case ZONE_CONNECTION::FULL:
                default:
                    // No knockout - pad connects directly to the hatch
                    break;
                }
 
                continue;
            }
 
            if( aZone->IsTeardropArea() )
            {
                connection = ZONE_CONNECTION::FULL;
            }
            else
            {
                constraint = bds.m_DRCEngine->EvalZoneConnection( pad, aZone, aLayer );
                connection = constraint.m_ZoneConnection;
            }
 
            if( connection == ZONE_CONNECTION::THERMAL && !pad->CanFlashLayer( aLayer ) )
                connection = ZONE_CONNECTION::NONE;
 
            switch( connection )
            {
            case ZONE_CONNECTION::THERMAL:
                padShape = pad->GetEffectiveShape( aLayer, FLASHING::ALWAYS_FLASHED );
 
                if( aFill.Collide( padShape.get(), 0 ) )
                {
                    constraint = bds.m_DRCEngine->EvalRules( THERMAL_RELIEF_GAP_CONSTRAINT, pad, aZone, aLayer );
                    padClearance = constraint.GetValue().Min();
 
                    aThermalConnectionPads.push_back( pad );
                    addKnockout( pad, aLayer, padClearance, holes );
                }
 
                break;
 
            case ZONE_CONNECTION::NONE:
                constraint = bds.m_DRCEngine->EvalRules( PHYSICAL_CLEARANCE_CONSTRAINT, pad, aZone, aLayer );
 
                if( constraint.GetValue().Min() > aZone->GetLocalClearance().value() )
                    padClearance = constraint.GetValue().Min();
                else
                    padClearance = aZone->GetLocalClearance().value();
 
                if( pad->FlashLayer( aLayer ) )
                {
                    addKnockout( pad, aLayer, padClearance, holes );
                }
                else if( pad->GetDrillSize().x > 0 )
                {
                    constraint = bds.m_DRCEngine->EvalRules( PHYSICAL_HOLE_CLEARANCE_CONSTRAINT, pad, aZone, aLayer );
 
                    if( constraint.GetValue().Min() > padClearance )
                        holeClearance = constraint.GetValue().Min();
                    else
                        holeClearance = padClearance;
 
                    pad->TransformHoleToPolygon( holes, holeClearance, m_maxError, ERROR_OUTSIDE );
                }
 
                break;
 
            default:
                // No knockout
                continue;
            }
        }
    }
 
    // For hatch zones, vias also need thermal treatment to prevent isolation inside hatch holes.
    // We respect the zone connection type just like pads: THERMAL gets a relief knockout,
    // FULL connects directly to the webbing, NONE is handled in buildCopperItemClearances.
    if( aZone->GetFillMode() == ZONE_FILL_MODE::HATCH_PATTERN )
    {
        for( PCB_TRACK* track : m_board->Tracks() )
        {
            if( track->Type() != PCB_VIA_T )
                continue;
 
            PCB_VIA* via = static_cast<PCB_VIA*>( track );
 
            if( !via->IsOnLayer( aLayer ) )
                continue;
 
            BOX2I viaBBox = via->GetBoundingBox();
            viaBBox.Inflate( m_worstClearance );
 
            if( !viaBBox.Intersects( aZone->GetBoundingBox() ) )
                continue;
 
            // Deduplicate coincident vias (circular, so use max of drill and width)
            int viaEffectiveSize = std::max( via->GetDrillValue(), via->GetWidth( aLayer ) );
            VIA_KNOCKOUT_KEY viaKey{ via->GetPosition(), viaEffectiveSize, via->GetNetCode() };
 
            if( !processedVias.insert( viaKey ).second )
                continue;
 
            bool noConnection = via->GetNetCode() != aZone->GetNetCode()
                    || ( via->Padstack().UnconnectedLayerMode() == UNCONNECTED_LAYER_MODE::START_END_ONLY
                         && aLayer != via->Padstack().Drill().start
                         && aLayer != via->Padstack().Drill().end );
 
            if( via->GetZoneLayerOverride( aLayer ) == ZLO_FORCE_NO_ZONE_CONNECTION )
                noConnection = true;
 
            // Check if this layer is affected by backdrill or post-machining
            if( via->IsBackdrilledOrPostMachined( aLayer ) )
            {
                noConnection = true;
 
                // Add knockout for backdrill/post-machining hole
                int pmSize = 0;
                int bdSize = 0;
 
                const PADSTACK::POST_MACHINING_PROPS& frontPM = via->Padstack().FrontPostMachining();
                const PADSTACK::POST_MACHINING_PROPS& backPM = via->Padstack().BackPostMachining();
 
                if( frontPM.mode != PAD_DRILL_POST_MACHINING_MODE::NOT_POST_MACHINED
                    && frontPM.mode != PAD_DRILL_POST_MACHINING_MODE::UNKNOWN )
                {
                    pmSize = std::max( pmSize, frontPM.size );
                }
 
                if( backPM.mode != PAD_DRILL_POST_MACHINING_MODE::NOT_POST_MACHINED
                    && backPM.mode != PAD_DRILL_POST_MACHINING_MODE::UNKNOWN )
                {
                    pmSize = std::max( pmSize, backPM.size );
                }
 
                const PADSTACK::DRILL_PROPS& secDrill = via->Padstack().SecondaryDrill();
 
                if( secDrill.start != UNDEFINED_LAYER && secDrill.end != UNDEFINED_LAYER )
                    bdSize = secDrill.size.x;
 
                int knockoutSize = std::max( pmSize, bdSize );
 
                if( knockoutSize > 0 )
                {
                    int clearance = aZone->GetLocalClearance().value_or( 0 );
 
                    TransformCircleToPolygon( holes, via->GetPosition(), knockoutSize / 2 + clearance,
                                              m_maxError, ERROR_OUTSIDE );
                }
            }
 
            if( noConnection )
                continue;
 
            constraint = bds.m_DRCEngine->EvalZoneConnection( via, aZone, aLayer );
            connection = constraint.m_ZoneConnection;
 
            switch( connection )
            {
            case ZONE_CONNECTION::THERMAL:
            {
                constraint = bds.m_DRCEngine->EvalRules( THERMAL_RELIEF_GAP_CONSTRAINT, via,
                                                          aZone, aLayer );
                int thermalGap = constraint.GetValue().Min();
 
                // Only force thermal if the via is small enough to be isolated in a hatch hole.
                // A via wider than the hole width will always touch the webbing naturally.
                if( thermalGap > 0 )
                {
                    aThermalConnectionPads.push_back( via );
                    addKnockout( via, aLayer, thermalGap, holes );
                }
 
                break;
            }
 
            case ZONE_CONNECTION::NONE:
                // Will be handled by buildCopperItemClearances
                break;
 
            case ZONE_CONNECTION::FULL:
            default:
                // No knockout - via connects directly to the hatch webbing
                break;
            }
        }
    }
 
    aFill.BooleanSubtract( holes );
}
 

void ZONE_FILLER::buildCopperItemClearances( const ZONE* aZone, PCB_LAYER_ID aLayer,
                                             const std::vector<PAD*>& aNoConnectionPads,
                                             SHAPE_POLY_SET& aHoles,
                                             bool aIncludeZoneClearances )
{
    BOARD_DESIGN_SETTINGS& bds = m_board->GetDesignSettings();
    long                   ticker = 0;
 
    // Deduplication sets for coincident items
    std::unordered_set<PAD_KNOCKOUT_KEY, PAD_KNOCKOUT_KEY_HASH>     processedPads;
    std::unordered_set<VIA_KNOCKOUT_KEY, VIA_KNOCKOUT_KEY_HASH>     processedVias;
    std::unordered_set<TRACK_KNOCKOUT_KEY, TRACK_KNOCKOUT_KEY_HASH> processedTracks;
 
    auto checkForCancel =
            [&ticker]( PROGRESS_REPORTER* aReporter ) -> bool
            {
                return aReporter && ( ticker++ % 50 ) == 0 && aReporter->IsCancelled();
            };
 
    // A small extra clearance to be sure actual track clearances are not smaller than
    // requested clearance due to many approximations in calculations, like arc to segment
    // approx, rounding issues, etc.
    BOX2I zone_boundingbox = aZone->GetBoundingBox();
    int   extra_margin = pcbIUScale.mmToIU( ADVANCED_CFG::GetCfg().m_ExtraClearance );
 
    // Items outside the zone bounding box are skipped, so it needs to be inflated by the
    // largest clearance value found in the netclasses and rules
    zone_boundingbox.Inflate( m_worstClearance + extra_margin );
 
    auto evalRulesForItems =
            [&bds]( DRC_CONSTRAINT_T aConstraint, const BOARD_ITEM* a, const BOARD_ITEM* b,
                    PCB_LAYER_ID aEvalLayer ) -> int
            {
                DRC_CONSTRAINT c = bds.m_DRCEngine->EvalRules( aConstraint, a, b, aEvalLayer );
 
                if( c.IsNull() )
                    return -1;
                else
                    return c.GetValue().Min();
            };
 
    // Add non-connected pad clearances
    //
    auto knockoutPadClearance =
            [&]( PAD* aPad )
            {
                int  init_gap = evalRulesForItems( PHYSICAL_CLEARANCE_CONSTRAINT, aZone, aPad, aLayer );
                int  gap = init_gap;
                bool hasHole = aPad->GetDrillSize().x > 0;
                bool flashLayer = aPad->FlashLayer( aLayer );
                bool platedHole = hasHole && aPad->GetAttribute() == PAD_ATTRIB::PTH;
 
                if( flashLayer || platedHole )
                {
                    gap = std::max( gap, evalRulesForItems( CLEARANCE_CONSTRAINT, aZone, aPad, aLayer ) );
                }
 
                if( flashLayer && gap >= 0 )
                    addKnockout( aPad, aLayer, gap + extra_margin, aHoles );
 
                if( hasHole )
                {
                    // NPTH do not need copper clearance gaps to their holes
                    if( aPad->GetAttribute() == PAD_ATTRIB::NPTH )
                        gap = init_gap;
 
                    gap = std::max( gap, evalRulesForItems( PHYSICAL_HOLE_CLEARANCE_CONSTRAINT, aZone, aPad, aLayer ) );
 
                    gap = std::max( gap, evalRulesForItems( HOLE_CLEARANCE_CONSTRAINT, aZone, aPad, aLayer ) );
 
                    // Oblong NPTH holes are milled rather than drilled, so they need
                    // edge clearance in addition to hole clearance
                    if( aPad->GetAttribute() == PAD_ATTRIB::NPTH
                        && aPad->GetDrillSize().x != aPad->GetDrillSize().y )
                    {
                        gap = std::max( gap, evalRulesForItems( EDGE_CLEARANCE_CONSTRAINT, aZone,
                                                                aPad, aLayer ) );
                    }
 
                    if( gap >= 0 )
                        addHoleKnockout( aPad, gap + extra_margin, aHoles );
                }
 
                // Handle backdrill and post-machining knockouts
                if( aPad->IsBackdrilledOrPostMachined( aLayer ) )
                {
                    int pmSize = 0;
                    int bdSize = 0;
 
                    const PADSTACK::POST_MACHINING_PROPS& frontPM = aPad->Padstack().FrontPostMachining();
                    const PADSTACK::POST_MACHINING_PROPS& backPM = aPad->Padstack().BackPostMachining();
 
                    if( frontPM.mode != PAD_DRILL_POST_MACHINING_MODE::NOT_POST_MACHINED
                        && frontPM.mode != PAD_DRILL_POST_MACHINING_MODE::UNKNOWN )
                    {
                        pmSize = std::max( pmSize, frontPM.size );
                    }
 
                    if( backPM.mode != PAD_DRILL_POST_MACHINING_MODE::NOT_POST_MACHINED
                        && backPM.mode != PAD_DRILL_POST_MACHINING_MODE::UNKNOWN )
                    {
                        pmSize = std::max( pmSize, backPM.size );
                    }
 
                    const PADSTACK::DRILL_PROPS& secDrill = aPad->Padstack().SecondaryDrill();
 
                    if( secDrill.start != UNDEFINED_LAYER && secDrill.end != UNDEFINED_LAYER )
                        bdSize = secDrill.size.x;
 
                    int knockoutSize = std::max( pmSize, bdSize );
 
                    if( knockoutSize > 0 )
                    {
                        int clearance = std::max( gap, 0 ) + extra_margin;
 
                        TransformCircleToPolygon( aHoles, aPad->GetPosition(), knockoutSize / 2 + clearance,
                                                  m_maxError, ERROR_OUTSIDE );
                    }
                }
            };
 
    for( PAD* pad : aNoConnectionPads )
    {
        if( checkForCancel( m_progressReporter ) )
            return;
 
        // Deduplicate coincident pads (skip custom pads - they have complex shapes)
        PAD_SHAPE padShape = pad->GetShape( aLayer );
 
        if( padShape != PAD_SHAPE::CUSTOM )
        {
            // For circular pads: use max of drill and pad size; otherwise just pad size
            VECTOR2I padSize = pad->GetSize( aLayer );
            VECTOR2I effectiveSize;
 
            if( padShape == PAD_SHAPE::CIRCLE )
            {
                int drill = std::max( pad->GetDrillSize().x, pad->GetDrillSize().y );
                int maxDim = std::max( { padSize.x, padSize.y, drill } );
                effectiveSize = VECTOR2I( maxDim, maxDim );
            }
            else
            {
                effectiveSize = padSize;
            }
 
            PAD_KNOCKOUT_KEY padKey{ pad->GetPosition(), effectiveSize,
                                     static_cast<int>( padShape ), pad->GetOrientation(),
                                     pad->GetNetCode() };
 
            if( !processedPads.insert( padKey ).second )
                continue;
        }
 
        knockoutPadClearance( pad );
    }
 
    // Add non-connected track clearances
    //
    auto knockoutTrackClearance =
            [&]( PCB_TRACK* aTrack )
            {
                if( aTrack->GetBoundingBox().Intersects( zone_boundingbox ) )
                {
                    bool sameNet = aTrack->GetNetCode() == aZone->GetNetCode();
 
                    if( !aZone->IsTeardropArea() && aZone->GetNetCode() == 0 )
                        sameNet = false;
 
                    int  gap = evalRulesForItems( PHYSICAL_CLEARANCE_CONSTRAINT, aZone, aTrack, aLayer );
 
                    if( aTrack->Type() == PCB_VIA_T )
                    {
                        PCB_VIA* via = static_cast<PCB_VIA*>( aTrack );
 
                        if( via->GetZoneLayerOverride( aLayer ) == ZLO_FORCE_NO_ZONE_CONNECTION )
                            sameNet = false;
                    }
 
                    if( !sameNet )
                        gap = std::max( gap, evalRulesForItems( CLEARANCE_CONSTRAINT, aZone, aTrack, aLayer ) );
 
                    if( aTrack->Type() == PCB_VIA_T )
                    {
                        PCB_VIA* via = static_cast<PCB_VIA*>( aTrack );
 
                        if( via->FlashLayer( aLayer ) && gap > 0 )
                        {
                            via->TransformShapeToPolygon( aHoles, aLayer, gap + extra_margin, m_maxError,
                                                          ERROR_OUTSIDE );
                        }
 
                        gap = std::max( gap, evalRulesForItems( PHYSICAL_HOLE_CLEARANCE_CONSTRAINT, aZone, via,
                                                                aLayer ) );
 
                        if( !sameNet )
                            gap = std::max( gap, evalRulesForItems( HOLE_CLEARANCE_CONSTRAINT, aZone, via, aLayer ) );
 
                        if( gap >= 0 )
                        {
                            int radius = via->GetDrillValue() / 2;
 
                            TransformCircleToPolygon( aHoles, via->GetPosition(), radius + gap + extra_margin,
                                                      m_maxError, ERROR_OUTSIDE );
                        }
 
                        // Handle backdrill and post-machining knockouts
                        if( via->IsBackdrilledOrPostMachined( aLayer ) )
                        {
                            int pmSize = 0;
                            int bdSize = 0;
 
                            const PADSTACK::POST_MACHINING_PROPS& frontPM = via->Padstack().FrontPostMachining();
                            const PADSTACK::POST_MACHINING_PROPS& backPM = via->Padstack().BackPostMachining();
 
                            if( frontPM.mode != PAD_DRILL_POST_MACHINING_MODE::NOT_POST_MACHINED
                                && frontPM.mode != PAD_DRILL_POST_MACHINING_MODE::UNKNOWN )
                            {
                                pmSize = std::max( pmSize, frontPM.size );
                            }
 
                            if( backPM.mode != PAD_DRILL_POST_MACHINING_MODE::NOT_POST_MACHINED
                                && backPM.mode != PAD_DRILL_POST_MACHINING_MODE::UNKNOWN )
                            {
                                pmSize = std::max( pmSize, backPM.size );
                            }
 
                            const PADSTACK::DRILL_PROPS& secDrill = via->Padstack().SecondaryDrill();
 
                            if( secDrill.start != UNDEFINED_LAYER && secDrill.end != UNDEFINED_LAYER )
                                bdSize = secDrill.size.x;
 
                            int knockoutSize = std::max( pmSize, bdSize );
 
                            if( knockoutSize > 0 )
                            {
                                int clearance = std::max( gap, 0 ) + extra_margin;
 
                                TransformCircleToPolygon( aHoles, via->GetPosition(), knockoutSize / 2 + clearance,
                                                          m_maxError, ERROR_OUTSIDE );
                            }
                        }
                    }
                    else
                    {
                        if( gap >= 0 )
                        {
                            aTrack->TransformShapeToPolygon( aHoles, aLayer, gap + extra_margin, m_maxError,
                                                             ERROR_OUTSIDE );
                        }
                    }
                }
            };
 
    for( PCB_TRACK* track : m_board->Tracks() )
    {
        if( !track->IsOnLayer( aLayer ) )
            continue;
 
        if( checkForCancel( m_progressReporter ) )
            return;
 
        // Deduplicate coincident tracks and vias
        if( track->Type() == PCB_VIA_T )
        {
            PCB_VIA* via = static_cast<PCB_VIA*>( track );
            int viaEffectiveSize = std::max( via->GetDrillValue(), via->GetWidth( aLayer ) );
            VIA_KNOCKOUT_KEY viaKey{ via->GetPosition(), viaEffectiveSize, via->GetNetCode() };
 
            if( !processedVias.insert( viaKey ).second )
                continue;
        }
        else
        {
            TRACK_KNOCKOUT_KEY trackKey( track->GetStart(), track->GetEnd(), track->GetWidth() );
 
            if( !processedTracks.insert( trackKey ).second )
                continue;
        }
 
        knockoutTrackClearance( track );
    }
 
    // Add graphic item clearances.
    //
    auto knockoutGraphicClearance =
            [&]( BOARD_ITEM* aItem )
            {
                int shapeNet = -1;
 
                if( aItem->Type() == PCB_SHAPE_T )
                    shapeNet = static_cast<PCB_SHAPE*>( aItem )->GetNetCode();
 
                bool sameNet = shapeNet == aZone->GetNetCode();
 
                if( !aZone->IsTeardropArea() && aZone->GetNetCode() == 0 )
                    sameNet = false;
 
                // A item on the Edge_Cuts or Margin is always seen as on any layer:
                if( aItem->IsOnLayer( aLayer )
                        || aItem->IsOnLayer( Edge_Cuts )
                        || aItem->IsOnLayer( Margin ) )
                {
                    if( aItem->GetBoundingBox().Intersects( zone_boundingbox ) )
                    {
                        bool ignoreLineWidths = false;
                        int  gap = evalRulesForItems( PHYSICAL_CLEARANCE_CONSTRAINT, aZone, aItem, aLayer );
 
                        if( aItem->IsOnLayer( aLayer ) && !sameNet )
                        {
                            gap = std::max( gap, evalRulesForItems( CLEARANCE_CONSTRAINT, aZone, aItem, aLayer ) );
                        }
                        else if( aItem->IsOnLayer( Edge_Cuts ) )
                        {
                            gap = std::max( gap, evalRulesForItems( EDGE_CLEARANCE_CONSTRAINT, aZone, aItem, aLayer ) );
                            ignoreLineWidths = true;
                        }
                        else if( aItem->IsOnLayer( Margin ) )
                        {
                            gap = std::max( gap, evalRulesForItems( EDGE_CLEARANCE_CONSTRAINT, aZone, aItem, aLayer ) );
                        }
 
                        if( gap >= 0 )
                        {
                            gap += extra_margin;
                            addKnockout( aItem, aLayer, gap, ignoreLineWidths, aHoles );
                        }
                    }
                }
            };
 
    auto knockoutCourtyardClearance =
            [&]( FOOTPRINT* aFootprint )
            {
                if( aFootprint->GetBoundingBox().Intersects( zone_boundingbox ) )
                {
                    int gap = evalRulesForItems( PHYSICAL_CLEARANCE_CONSTRAINT, aZone, aFootprint, aLayer );
 
                    // For internal copper layers, GetCourtyard( aLayer ) always returns the
                    // front courtyard because IsBackLayer() is false for all internal layers.
                    // Use the footprint's own layer to select the correct courtyard instead.
                    PCB_LAYER_ID courtyardSide = IsInnerCopperLayer( aLayer ) ? aFootprint->GetLayer() : aLayer;
 
                    if( gap == 0 )
                    {
                        aHoles.Append( aFootprint->GetCourtyard( courtyardSide ) );
                    }
                    else if( gap > 0 )
                    {
                        SHAPE_POLY_SET hole = aFootprint->GetCourtyard( courtyardSide );
                        hole.Inflate( gap, CORNER_STRATEGY::ROUND_ALL_CORNERS, m_maxError );
                        aHoles.Append( hole );
                    }
                }
            };
 
    for( FOOTPRINT* footprint : m_board->Footprints() )
    {
        knockoutCourtyardClearance( footprint );
        knockoutGraphicClearance( &footprint->Reference() );
        knockoutGraphicClearance( &footprint->Value() );
 
        std::set<PAD*> allowedNetTiePads;
 
        // Don't knock out holes for graphic items which implement a net-tie to the zone's net
        // on the layer being filled.
        if( footprint->IsNetTie() )
        {
            for( PAD* pad : footprint->Pads() )
            {
                bool sameNet = pad->GetNetCode() == aZone->GetNetCode();
 
                if( !aZone->IsTeardropArea() && aZone->GetNetCode() == 0 )
                    sameNet = false;
 
                if( sameNet )
                {
                    if( pad->IsOnLayer( aLayer ) )
                        allowedNetTiePads.insert( pad );
 
                    for( PAD* other : footprint->GetNetTiePads( pad ) )
                    {
                        if( other->IsOnLayer( aLayer ) )
                            allowedNetTiePads.insert( other );
                    }
                }
            }
        }
 
        for( BOARD_ITEM* item : footprint->GraphicalItems() )
        {
            if( checkForCancel( m_progressReporter ) )
                return;
 
            BOX2I itemBBox = item->GetBoundingBox();
 
            if( !zone_boundingbox.Intersects( itemBBox ) )
                continue;
 
            bool skipItem = false;
 
            if( item->IsOnLayer( aLayer ) )
            {
                std::shared_ptr<SHAPE> itemShape = item->GetEffectiveShape();
 
                for( PAD* pad : allowedNetTiePads )
                {
                    if( pad->GetBoundingBox().Intersects( itemBBox )
                            && pad->GetEffectiveShape( aLayer )->Collide( itemShape.get() ) )
                    {
                        skipItem = true;
                        break;
                    }
                }
            }
 
            if( !skipItem )
                knockoutGraphicClearance( item );
        }
    }
 
    for( BOARD_ITEM* item : m_board->Drawings() )
    {
        if( checkForCancel( m_progressReporter ) )
            return;
 
        knockoutGraphicClearance( item );
    }
 
    // Add non-connected zone clearances
    //
    auto knockoutZoneClearance =
            [&]( ZONE* aKnockout )
            {
                // If the zones share no common layers
                if( !aKnockout->GetLayerSet().test( aLayer ) )
                    return;
 
                if( aKnockout->GetBoundingBox().Intersects( zone_boundingbox ) )
                {
                    if( aKnockout->GetIsRuleArea() )
                    {
                        if( aKnockout->GetDoNotAllowZoneFills() && !aZone->IsTeardropArea() )
                        {
                            // Keepouts use outline with no clearance
                            aKnockout->TransformSmoothedOutlineToPolygon( aHoles, 0, m_maxError, ERROR_OUTSIDE,
                                                                          nullptr );
                        }
                    }
                    else
                    {
                        if( aKnockout->HigherPriority( aZone ) && !aKnockout->SameNet( aZone ) )
                        {
                            int gap = std::max( 0, evalRulesForItems( PHYSICAL_CLEARANCE_CONSTRAINT, aZone, aKnockout,
                                                                      aLayer ) );
 
                            gap = std::max( gap, evalRulesForItems( CLEARANCE_CONSTRAINT, aZone, aKnockout, aLayer ) );
 
                            // Negative clearance permits zones to short
                            if( gap < 0 )
                                return;
 
                            SHAPE_POLY_SET poly;
                            aKnockout->TransformShapeToPolygon( poly, aLayer, gap + extra_margin, m_maxError,
                                                                ERROR_OUTSIDE );
                            aHoles.Append( poly );
                        }
                    }
                }
            };
 
    if( aIncludeZoneClearances )
    {
        for( ZONE* otherZone : m_board->Zones() )
        {
            if( checkForCancel( m_progressReporter ) )
                return;
 
            knockoutZoneClearance( otherZone );
        }
 
        for( FOOTPRINT* footprint : m_board->Footprints() )
        {
            for( ZONE* otherZone : footprint->Zones() )
            {
                if( checkForCancel( m_progressReporter ) )
                    return;
 
                knockoutZoneClearance( otherZone );
            }
        }
    }
 
    aHoles.Simplify();
}
 

void ZONE_FILLER::buildDifferentNetZoneClearances( const ZONE* aZone, PCB_LAYER_ID aLayer, SHAPE_POLY_SET& aHoles )
{
    BOARD_DESIGN_SETTINGS& bds = m_board->GetDesignSettings();
    int extra_margin = pcbIUScale.mmToIU( ADVANCED_CFG::GetCfg().m_ExtraClearance );
 
    BOX2I zone_boundingbox = aZone->GetBoundingBox();
    zone_boundingbox.Inflate( m_worstClearance + extra_margin );
 
    auto evalRulesForItems =
            [&bds]( DRC_CONSTRAINT_T aConstraint, const BOARD_ITEM* a, const BOARD_ITEM* b,
                    PCB_LAYER_ID aEvalLayer ) -> int
            {
                DRC_CONSTRAINT c = bds.m_DRCEngine->EvalRules( aConstraint, a, b, aEvalLayer );
 
                if( c.IsNull() )
                    return -1;
                else
                    return c.GetValue().Min();
            };
 
    // Keepout zones (rule areas) are excluded here because they are subtracted earlier in the
    // fill process, before the deflate/inflate min-width cycle.  Subtracting them here would
    // trigger a second deflate/inflate pass that creates artifacts along curved keepout
    // boundaries (issue 23515).
    auto knockoutZoneClearance =
            [&]( ZONE* aKnockout )
            {
                if( aKnockout->GetIsRuleArea() )
                    return;
 
                if( !aKnockout->GetLayerSet().test( aLayer ) )
                    return;
 
                if( aKnockout->GetBoundingBox().Intersects( zone_boundingbox ) )
                {
                    if( aKnockout->HigherPriority( aZone ) && !aKnockout->SameNet( aZone ) )
                    {
                        int gap = std::max( 0, evalRulesForItems( PHYSICAL_CLEARANCE_CONSTRAINT,
                                                                   aZone, aKnockout, aLayer ) );
 
                        gap = std::max( gap, evalRulesForItems( CLEARANCE_CONSTRAINT, aZone,
                                                                 aKnockout, aLayer ) );
 
                        if( gap < 0 )
                            return;
 
                        SHAPE_POLY_SET poly;
                        aKnockout->TransformShapeToPolygon( poly, aLayer, gap + extra_margin,
                                                             m_maxError, ERROR_OUTSIDE );
                        aHoles.Append( poly );
                    }
                }
            };
 
    forEachBoardAndFootprintZone( m_board, knockoutZoneClearance );
 
    aHoles.Simplify();
}
 

void ZONE_FILLER::subtractHigherPriorityZones( const ZONE* aZone, PCB_LAYER_ID aLayer,
                                                SHAPE_POLY_SET& aRawFill )
{
    BOX2I          zoneBBox = aZone->GetBoundingBox();
    SHAPE_POLY_SET knockouts;
 
    auto collectZoneOutline =
            [&]( ZONE* aKnockout )
            {
                if( !aKnockout->GetLayerSet().test( aLayer ) )
                    return;
 
                if( aKnockout->GetBoundingBox().Intersects( zoneBBox ) )
                    appendZoneOutlineWithoutArcs( aKnockout, knockouts );
            };
 
    forEachBoardAndFootprintZone(
            m_board,
            [&]( ZONE* otherZone )
            {
                // Don't use `HigherPriority()` here because we only want explicitly-higher
                // priorities, not equal-priority zones.
                bool higherPrioritySameNet =
                        otherZone->SameNet( aZone )
                        && otherZone->GetAssignedPriority() > aZone->GetAssignedPriority();
 
                if( higherPrioritySameNet && !otherZone->IsTeardropArea() )
                    collectZoneOutline( otherZone );
            } );
 
    if( knockouts.OutlineCount() > 0 )
        aRawFill.BooleanSubtract( knockouts );
}
 
 
void ZONE_FILLER::connect_nearby_polys( SHAPE_POLY_SET& aPolys, double aDistance )
{
    if( aPolys.OutlineCount() < 1 )
        return;
 
    VERTEX_CONNECTOR vs( aPolys.BBoxFromCaches(), aPolys, aDistance );
 
    vs.FindResults();
 
    // This cannot be a reference because we need to do the comparison below while
    // changing the values
    std::map<int, std::vector<std::pair<int, VECTOR2I>>> insertion_points;
 
    for( const RESULTS& result : vs.GetResults() )
    {
        SHAPE_LINE_CHAIN& line1 = aPolys.Outline( result.m_outline1 );
        SHAPE_LINE_CHAIN& line2 = aPolys.Outline( result.m_outline2 );
 
        VECTOR2I pt1 = line1.CPoint( result.m_vertex1 );
        VECTOR2I pt2 = line2.CPoint( result.m_vertex2 );
 
        // We want to insert the existing point first so that we can place the new point
        // between the two points at the same location.
        insertion_points[result.m_outline1].push_back( { result.m_vertex1, pt1 } );
        insertion_points[result.m_outline1].push_back( { result.m_vertex1, pt2 } );
    }
 
    for( auto& [outline, vertices] : insertion_points )
    {
        SHAPE_LINE_CHAIN& line = aPolys.Outline( outline );
 
        // Stable sort here because we want to make sure that we are inserting pt1 first and
        // pt2 second but still sorting the rest of the indices from highest to lowest.
        // This allows us to insert into the existing polygon without modifying the future
        // insertion points.
        std::stable_sort( vertices.begin(), vertices.end(),
                  []( const std::pair<int, VECTOR2I>& a, const std::pair<int, VECTOR2I>& b )
                  {
                      return a.first > b.first;
                  } );
 
        for( const auto& [vertex, pt] : vertices )
            line.Insert( vertex + 1, pt );
    }
}
 
 
void ZONE_FILLER::postKnockoutMinWidthPrune( const ZONE* aZone, SHAPE_POLY_SET& aFillPolys )
{
    int half_min_width = aZone->GetMinThickness() / 2;
    int epsilon = pcbIUScale.mmToIU( 0.001 );
 
    if( half_min_width - epsilon <= epsilon )
        return;
 
    SHAPE_POLY_SET preDeflate = aFillPolys.CloneDropTriangulation();
 
    aFillPolys.Deflate( half_min_width - epsilon, CORNER_STRATEGY::CHAMFER_ALL_CORNERS,
                        m_maxError );
 
    aFillPolys.Fracture();
    connect_nearby_polys( aFillPolys, aZone->GetMinThickness() );
 
    for( int ii = aFillPolys.OutlineCount() - 1; ii >= 0; ii-- )
    {
        std::vector<SHAPE_LINE_CHAIN>& island = aFillPolys.Polygon( ii );
        BOX2I                          islandExtents;
 
        for( const VECTOR2I& pt : island.front().CPoints() )
        {
            islandExtents.Merge( pt );
 
            if( islandExtents.GetSizeMax() > aZone->GetMinThickness() )
                break;
        }
 
        if( islandExtents.GetSizeMax() < aZone->GetMinThickness() )
            aFillPolys.DeletePolygon( ii );
    }
 
    aFillPolys.Inflate( half_min_width - epsilon, CORNER_STRATEGY::ROUND_ALL_CORNERS, m_maxError,
                        true );
    aFillPolys.BooleanIntersection( preDeflate );
}
 
 
#define DUMP_POLYS_TO_COPPER_LAYER( a, b, c ) \
    { if( m_debugZoneFiller && aDebugLayer == b ) \
        { \
            m_board->SetLayerName( b, c ); \
            SHAPE_POLY_SET d = a; \
            d.Fracture(); \
            aFillPolys = d; \
            return false; \
        } \
    }
 
 
/*
 * Note that aSmoothedOutline is larger than the zone where it intersects with other, same-net
 * zones.  This is to prevent the re-inflation post min-width trimming from createing divots
 * between adjacent zones.  The final aMaxExtents trimming will remove these areas from the final
 * fill.
 */
bool ZONE_FILLER::fillCopperZone( const ZONE* aZone, PCB_LAYER_ID aLayer, PCB_LAYER_ID aDebugLayer,
                                  const SHAPE_POLY_SET& aSmoothedOutline,
                                  const SHAPE_POLY_SET& aMaxExtents, SHAPE_POLY_SET& aFillPolys )
{
    // m_maxError is initialized in the constructor. Don't reassign here to avoid data races
    // when multiple threads call this function concurrently.
 
    // Features which are min_width should survive pruning; features that are *less* than
    // min_width should not.  Therefore we subtract epsilon from the min_width when
    // deflating/inflating.
    int half_min_width = aZone->GetMinThickness() / 2;
    int epsilon = pcbIUScale.mmToIU( 0.001 );
 
    // Solid polygons are deflated and inflated during calculations.  Deflating doesn't cause
    // issues, but inflate is tricky as it can create excessively long and narrow spikes for
    // acute angles.
    // ALLOW_ACUTE_CORNERS cannot be used due to the spike problem.
    // CHAMFER_ACUTE_CORNERS is tempting, but can still produce spikes in some unusual
    // circumstances (https://gitlab.com/kicad/code/kicad/-/issues/5581).
    // It's unclear if ROUND_ACUTE_CORNERS would have the same issues, but is currently avoided
    // as a "less-safe" option.
    // ROUND_ALL_CORNERS produces the uniformly nicest shapes, but also a lot of segments.
    // CHAMFER_ALL_CORNERS improves the segment count.
    CORNER_STRATEGY fastCornerStrategy = CORNER_STRATEGY::CHAMFER_ALL_CORNERS;
    CORNER_STRATEGY cornerStrategy = CORNER_STRATEGY::ROUND_ALL_CORNERS;
 
    std::vector<BOARD_ITEM*>     thermalConnectionPads;
    std::vector<PAD*>            noConnectionPads;
    std::deque<SHAPE_LINE_CHAIN> thermalSpokes;
    SHAPE_POLY_SET               clearanceHoles;
 
    aFillPolys = aSmoothedOutline;
    DUMP_POLYS_TO_COPPER_LAYER( aFillPolys, In1_Cu, wxT( "smoothed-outline" ) );
 
    if( m_progressReporter && m_progressReporter->IsCancelled() )
        return false;
 
    /* -------------------------------------------------------------------------------------
     * Knockout thermal reliefs.
     */
 
    knockoutThermalReliefs( aZone, aLayer, aFillPolys, thermalConnectionPads, noConnectionPads );
    DUMP_POLYS_TO_COPPER_LAYER( aFillPolys, In2_Cu, wxT( "minus-thermal-reliefs" ) );
 
    if( m_progressReporter && m_progressReporter->IsCancelled() )
        return false;
 
    /* -------------------------------------------------------------------------------------
     * For hatch zones, add thermal rings around pads with thermal relief.
     * The rings are clipped to the zone boundary and provide the connection point
     * for the hatch webbing instead of connecting directly to the pad.
     */
 
    SHAPE_POLY_SET thermalRings;
 
    if( aZone->GetFillMode() == ZONE_FILL_MODE::HATCH_PATTERN )
    {
        buildHatchZoneThermalRings( aZone, aLayer, aSmoothedOutline, thermalConnectionPads,
                                    aFillPolys, thermalRings );
        DUMP_POLYS_TO_COPPER_LAYER( aFillPolys, In2_Cu, wxT( "plus-thermal-rings" ) );
    }
 
    if( m_progressReporter && m_progressReporter->IsCancelled() )
        return false;
 
    /* -------------------------------------------------------------------------------------
     * Knockout electrical clearances.
     */
 
    // When iterative refill is enabled, we build zone-to-zone clearances separately so we can
    // cache the fill before zone knockouts are applied (issue 21746).  Keepout zones are always
    // included in clearanceHoles regardless of the iterative refill setting so they are
    // subtracted before the deflate/inflate min-width cycle.  Subtracting keepouts after that
    // cycle and running a second deflate/inflate pass creates artifacts along curved keepout
    // boundaries (issue 23515).
    const bool iterativeRefill = ADVANCED_CFG::GetCfg().m_ZoneFillIterativeRefill;
 
    buildCopperItemClearances( aZone, aLayer, noConnectionPads, clearanceHoles,
                               !iterativeRefill /* include zone clearances only if not iterative */ );
 
    if( iterativeRefill )
    {
        BOX2I zone_boundingbox = aZone->GetBoundingBox();
        bool  addedKeepoutHoles = false;
 
        auto collectKeepoutHoles =
                [&]( ZONE* candidate )
                {
                    if( aZone->IsTeardropArea() )
                        return;
 
                    if( !isZoneFillKeepout( candidate, aLayer, zone_boundingbox ) )
                        return;
 
                    candidate->TransformSmoothedOutlineToPolygon( clearanceHoles, 0, m_maxError,
                                                                  ERROR_OUTSIDE, nullptr );
                    addedKeepoutHoles = true;
                };
 
        forEachBoardAndFootprintZone( m_board, collectKeepoutHoles );
 
        if( addedKeepoutHoles )
            clearanceHoles.Simplify();
    }
 
    DUMP_POLYS_TO_COPPER_LAYER( clearanceHoles, In3_Cu, wxT( "clearance-holes" ) );
 
    if( m_progressReporter && m_progressReporter->IsCancelled() )
        return false;
 
    /* -------------------------------------------------------------------------------------
     * Add thermal relief spokes.
     */
 
    buildThermalSpokes( aZone, aLayer, thermalConnectionPads, thermalSpokes );
 
    if( m_progressReporter && m_progressReporter->IsCancelled() )
        return false;
 
    // Create a temporary zone that we can hit-test spoke-ends against.  It's only temporary
    // because the "real" subtract-clearance-holes has to be done after the spokes are added.
    SHAPE_POLY_SET testAreas = aFillPolys.CloneDropTriangulation();
    testAreas.BooleanSubtract( clearanceHoles );
 
    // When iterative refill is enabled, zone-to-zone clearances are not included in
    // clearanceHoles (they're applied later to allow pre-knockout caching).  But we still
    // need to account for them when testing spoke endpoints, otherwise spokes will be kept
    // that point into areas that will be knocked out by higher-priority zones.
    SHAPE_POLY_SET zoneClearances;
 
    if( iterativeRefill )
    {
        buildDifferentNetZoneClearances( aZone, aLayer, zoneClearances );
 
        if( zoneClearances.OutlineCount() > 0 )
            testAreas.BooleanSubtract( zoneClearances );
    }
 
    DUMP_POLYS_TO_COPPER_LAYER( testAreas, In4_Cu, wxT( "minus-clearance-holes" ) );
 
    // Prune features that don't meet minimum-width criteria
    if( half_min_width - epsilon > epsilon )
    {
        testAreas.Deflate( half_min_width - epsilon, fastCornerStrategy, m_maxError );
        DUMP_POLYS_TO_COPPER_LAYER( testAreas, In5_Cu, wxT( "spoke-test-deflated" ) );
 
        testAreas.Inflate( half_min_width - epsilon, fastCornerStrategy, m_maxError );
        DUMP_POLYS_TO_COPPER_LAYER( testAreas, In6_Cu, wxT( "spoke-test-reinflated" ) );
    }
 
    if( m_progressReporter && m_progressReporter->IsCancelled() )
        return false;
 
    // Build a Y-stripe spatial index for O(sqrt(V)) spoke endpoint containment queries
    // instead of O(V) brute-force ray-casting with bbox caches.
    POLY_YSTRIPES_INDEX spokeTestIndex;
    spokeTestIndex.Build( testAreas );
    int interval = 0;
 
    SHAPE_POLY_SET debugSpokes;
 
    for( const SHAPE_LINE_CHAIN& spoke : thermalSpokes )
    {
        const VECTOR2I& testPt = spoke.CPoint( 3 );
 
        // Hit-test against zone body
        if( spokeTestIndex.Contains( testPt, 1 ) )
        {
            if( m_debugZoneFiller )
                debugSpokes.AddOutline( spoke );
 
            aFillPolys.AddOutline( spoke );
            continue;
        }
 
        if( interval++ > 400 )
        {
            if( m_progressReporter && m_progressReporter->IsCancelled() )
                return false;
 
            interval = 0;
        }
 
        // Hit-test against other spokes
        for( const SHAPE_LINE_CHAIN& other : thermalSpokes )
        {
            // Hit test in both directions to avoid interactions with round-off errors.
            // (See https://gitlab.com/kicad/code/kicad/-/issues/13316.)
            if( &other != &spoke
                && other.PointInside( testPt, 1 )
                && spoke.PointInside( other.CPoint( 3 ), 1 ) )
            {
                if( m_debugZoneFiller )
                    debugSpokes.AddOutline( spoke );
 
                aFillPolys.AddOutline( spoke );
                break;
            }
        }
    }
 
    DUMP_POLYS_TO_COPPER_LAYER( debugSpokes, In7_Cu, wxT( "spokes" ) );
 
    if( m_progressReporter && m_progressReporter->IsCancelled() )
        return false;
 
    aFillPolys.BooleanSubtract( clearanceHoles );
    DUMP_POLYS_TO_COPPER_LAYER( aFillPolys, In8_Cu, wxT( "after-spoke-trimming" ) );
 
    /* -------------------------------------------------------------------------------------
     * Prune features that don't meet minimum-width criteria
     */
 
    if( half_min_width - epsilon > epsilon )
    {
        aFillPolys.Deflate( half_min_width - epsilon, fastCornerStrategy, m_maxError );
 
        // Also deflate thermal rings to match, for correct hatch hole notching
        if( thermalRings.OutlineCount() > 0 )
            thermalRings.Deflate( half_min_width - epsilon, fastCornerStrategy, m_maxError );
    }
 
    // Min-thickness is the web thickness.  On the other hand, a blob min-thickness by
    // min-thickness is not useful.  Since there's no obvious definition of web vs. blob, we
    // arbitrarily choose "at least 2X min-thickness on one axis".  (Since we're doing this
    // during the deflated state, that means we test for "at least min-thickness".)
    for( int ii = aFillPolys.OutlineCount() - 1; ii >= 0; ii-- )
    {
        std::vector<SHAPE_LINE_CHAIN>& island = aFillPolys.Polygon( ii );
        BOX2I                          islandExtents;
 
        for( const VECTOR2I& pt : island.front().CPoints() )
        {
            islandExtents.Merge( pt );
 
            if( islandExtents.GetSizeMax() > aZone->GetMinThickness() )
                break;
        }
 
        if( islandExtents.GetSizeMax() < aZone->GetMinThickness() )
            aFillPolys.DeletePolygon( ii );
    }
 
    DUMP_POLYS_TO_COPPER_LAYER( aFillPolys, In9_Cu, wxT( "deflated" ) );
 
    if( m_progressReporter && m_progressReporter->IsCancelled() )
        return false;
 
    /* -------------------------------------------------------------------------------------
     * Process the hatch pattern (note that we do this while deflated)
     */
 
    if( aZone->GetFillMode() == ZONE_FILL_MODE::HATCH_PATTERN
        && ( !m_board->GetProject()
             || !m_board->GetProject()->GetLocalSettings().m_PrototypeZoneFill ) )
    {
        // Combine thermal rings with clearance holes (non-connected pad clearances) so that
        // the hatch hole-dropping logic considers both types of rings
        SHAPE_POLY_SET ringsToProtect = thermalRings;
        ringsToProtect.BooleanAdd( clearanceHoles );
 
        if( !addHatchFillTypeOnZone( aZone, aLayer, aDebugLayer, aFillPolys, ringsToProtect ) )
            return false;
    }
    else
    {
        /* ---------------------------------------------------------------------------------
         * Connect nearby polygons with zero-width lines in order to ensure correct
         * re-inflation.
         */
        aFillPolys.Fracture();
        connect_nearby_polys( aFillPolys, aZone->GetMinThickness() );
 
        DUMP_POLYS_TO_COPPER_LAYER( aFillPolys, In10_Cu, wxT( "connected-nearby-polys" ) );
    }
 
    if( m_progressReporter && m_progressReporter->IsCancelled() )
        return false;
 
    /* -------------------------------------------------------------------------------------
     * Finish minimum-width pruning by re-inflating
     */
 
    if( half_min_width - epsilon > epsilon )
        aFillPolys.Inflate( half_min_width - epsilon, cornerStrategy, m_maxError, true );
 
    // The deflation/inflation process can leave notches in the outline.  Remove these by
    // doing a union with the original ring
    aFillPolys.BooleanAdd( thermalRings );
 
    DUMP_POLYS_TO_COPPER_LAYER( aFillPolys, In15_Cu, wxT( "after-reinflating" ) );
 
    /* -------------------------------------------------------------------------------------
     * Ensure additive changes (thermal stubs and inflating acute corners) do not add copper
     * outside the zone boundary, inside the clearance holes, or between otherwise isolated
     * islands
     */
 
    for( BOARD_ITEM* item : thermalConnectionPads )
    {
        if( item->Type() == PCB_PAD_T )
            addHoleKnockout( static_cast<PAD*>( item ), 0, clearanceHoles );
    }
 
    aFillPolys.BooleanIntersection( aMaxExtents );
    DUMP_POLYS_TO_COPPER_LAYER( aFillPolys, In16_Cu, wxT( "after-trim-to-outline" ) );
    aFillPolys.BooleanSubtract( clearanceHoles );
    DUMP_POLYS_TO_COPPER_LAYER( aFillPolys, In17_Cu, wxT( "after-trim-to-clearance-holes" ) );
 
    // Cache the pre-knockout fill for iterative refill optimization (issue 21746).
    // The cache stores the fill BEFORE zone-to-zone knockouts so the iterative refill can
    // reclaim space when higher-priority zones have islands removed.
    bool knockoutsApplied = false;
 
    if( iterativeRefill )
    {
        {
            std::lock_guard<std::mutex> lock( m_cacheMutex );
            m_preKnockoutFillCache[{ aZone, aLayer }] = aFillPolys;
        }
 
        // Reuse the zone clearances already computed for spoke endpoint testing
        if( zoneClearances.OutlineCount() > 0 )
        {
            aFillPolys.BooleanSubtract( zoneClearances );
            knockoutsApplied = true;
        }
    }
 
    /* -------------------------------------------------------------------------------------
     * Re-prune minimum-width violations introduced by different-net zone knockouts.
     *
     * This must run BEFORE subtracting same-net higher-priority zones.  At this point the
     * fill still extends into overlapping same-net zone areas, which provides a natural
     * buffer that prevents the deflate/inflate cycle from creating divots at same-net
     * zone boundaries (the same role aSmoothedOutline plays in the initial min-width pass).
     */
 
    if( knockoutsApplied )
        postKnockoutMinWidthPrune( aZone, aFillPolys );
 
    DUMP_POLYS_TO_COPPER_LAYER( aFillPolys, In18_Cu, wxT( "after-post-knockout-min-width" ) );
 
    /* -------------------------------------------------------------------------------------
     * Lastly give any same-net but higher-priority zones control over their own area.
     */
 
    subtractHigherPriorityZones( aZone, aLayer, aFillPolys );
    DUMP_POLYS_TO_COPPER_LAYER( aFillPolys, In19_Cu, wxT( "minus-higher-priority-zones" ) );
 
    aFillPolys.Fracture();
    return true;
}
 
 
bool ZONE_FILLER::fillNonCopperZone( const ZONE* aZone, PCB_LAYER_ID aLayer,
                                     const SHAPE_POLY_SET& aSmoothedOutline,
                                     SHAPE_POLY_SET& aFillPolys )
{
    BOX2I                  zone_boundingbox = aZone->GetBoundingBox();
    SHAPE_POLY_SET         clearanceHoles;
    long                   ticker = 0;
 
    auto checkForCancel =
            [&ticker]( PROGRESS_REPORTER* aReporter ) -> bool
            {
                return aReporter && ( ticker++ % 50 ) == 0 && aReporter->IsCancelled();
            };
 
    auto knockoutGraphicItem =
            [&]( BOARD_ITEM* aItem )
            {
                if( aItem->IsKnockout() && aItem->IsOnLayer( aLayer )
                        && aItem->GetBoundingBox().Intersects( zone_boundingbox ) )
                {
                    addKnockout( aItem, aLayer, 0, true, clearanceHoles );
                }
            };
 
    for( FOOTPRINT* footprint : m_board->Footprints() )
    {
        if( checkForCancel( m_progressReporter ) )
            return false;
 
        knockoutGraphicItem( &footprint->Reference() );
        knockoutGraphicItem( &footprint->Value() );
 
        for( BOARD_ITEM* item : footprint->GraphicalItems() )
            knockoutGraphicItem( item );
    }
 
    for( BOARD_ITEM* item : m_board->Drawings() )
    {
        if( checkForCancel( m_progressReporter ) )
            return false;
 
        knockoutGraphicItem( item );
    }
 
    aFillPolys = aSmoothedOutline;
    aFillPolys.BooleanSubtract( clearanceHoles );
 
    SHAPE_POLY_SET keepoutHoles;
 
    auto collectKeepout =
            [&]( ZONE* candidate )
            {
                if( !isZoneFillKeepout( candidate, aLayer, zone_boundingbox ) )
                    return;
 
                appendZoneOutlineWithoutArcs( candidate, keepoutHoles );
            };
 
    bool cancelledKeepoutScan = false;
 
    forEachBoardAndFootprintZone(
            m_board,
            [&]( ZONE* keepout )
            {
                if( cancelledKeepoutScan )
                    return;
 
                if( checkForCancel( m_progressReporter ) )
                {
                    cancelledKeepoutScan = true;
                    return;
                }
 
                collectKeepout( keepout );
            } );
 
    if( cancelledKeepoutScan )
        return false;
 
    if( keepoutHoles.OutlineCount() > 0 )
        aFillPolys.BooleanSubtract( keepoutHoles );
 
    // Features which are min_width should survive pruning; features that are *less* than
    // min_width should not.  Therefore we subtract epsilon from the min_width when
    // deflating/inflating.
    int half_min_width = aZone->GetMinThickness() / 2;
    int epsilon = pcbIUScale.mmToIU( 0.001 );
 
    aFillPolys.Deflate( half_min_width - epsilon, CORNER_STRATEGY::CHAMFER_ALL_CORNERS, m_maxError );
 
    // Remove the non filled areas due to the hatch pattern
    if( aZone->GetFillMode() == ZONE_FILL_MODE::HATCH_PATTERN )
    {
        SHAPE_POLY_SET noThermalRings;  // Non-copper zones have no thermal reliefs
 
        if( !addHatchFillTypeOnZone( aZone, aLayer, aLayer, aFillPolys, noThermalRings ) )
            return false;
    }
 
    // Re-inflate after pruning of areas that don't meet minimum-width criteria
    if( half_min_width - epsilon > epsilon )
        aFillPolys.Inflate( half_min_width - epsilon, CORNER_STRATEGY::ROUND_ALL_CORNERS, m_maxError );
 
    aFillPolys.Fracture();
    return true;
}
 
 
/*
 * Build the filled solid areas data from real outlines (stored in m_Poly)
 * The solid areas can be more than one on copper layers, and do not have holes
 * ( holes are linked by overlapping segments to the main outline)
 */
bool ZONE_FILLER::fillSingleZone( ZONE* aZone, PCB_LAYER_ID aLayer, SHAPE_POLY_SET& aFillPolys )
{
    SHAPE_POLY_SET* boardOutline = m_brdOutlinesValid ? &m_boardOutline : nullptr;
    SHAPE_POLY_SET  maxExtents;
    SHAPE_POLY_SET  smoothedPoly;
    PCB_LAYER_ID    debugLayer = UNDEFINED_LAYER;
 
    if( m_debugZoneFiller && LSET::InternalCuMask().Contains( aLayer ) )
    {
        debugLayer = aLayer;
        aLayer = F_Cu;
    }
 
    if( !aZone->BuildSmoothedPoly( maxExtents, aLayer, boardOutline, &smoothedPoly ) )
        return false;
 
    if( m_progressReporter && m_progressReporter->IsCancelled() )
        return false;
 
    if( aZone->IsOnCopperLayer() )
    {
        if( fillCopperZone( aZone, aLayer, debugLayer, smoothedPoly, maxExtents, aFillPolys ) )
            aZone->SetNeedRefill( false );
    }
    else
    {
        if( fillNonCopperZone( aZone, aLayer, smoothedPoly, aFillPolys ) )
            aZone->SetNeedRefill( false );
    }
 
    return true;
}
 

void ZONE_FILLER::buildThermalSpokes( const ZONE* aZone, PCB_LAYER_ID aLayer,
                                      const std::vector<BOARD_ITEM*>& aSpokedPadsList,
                                      std::deque<SHAPE_LINE_CHAIN>& aSpokesList )
{
    BOARD_DESIGN_SETTINGS& bds = m_board->GetDesignSettings();
    BOX2I                  zoneBB = aZone->GetBoundingBox();
    DRC_CONSTRAINT         constraint;
    int                    zone_half_width = aZone->GetMinThickness() / 2;
 
    if( aZone->GetFillMode() == ZONE_FILL_MODE::HATCH_PATTERN )
        zone_half_width = aZone->GetHatchThickness() / 2;
 
    zoneBB.Inflate( std::max( bds.GetBiggestClearanceValue(), aZone->GetLocalClearance().value() ) );
 
    // Is a point on the boundary of the polygon inside or outside?
    // The boundary may be off by MaxError
    int epsilon = bds.m_MaxError;
 
    for( BOARD_ITEM* item : aSpokedPadsList )
    {
        // We currently only connect to pads, not pad holes
        if( !item->IsOnLayer( aLayer ) )
            continue;
 
        int       thermalReliefGap = 0;
        int       spoke_w = 0;
        PAD*      pad = nullptr;
        PCB_VIA*  via = nullptr;
        bool      circular = false;
 
        if( item->Type() == PCB_PAD_T )
        {
            pad = static_cast<PAD*>( item );
            VECTOR2I padSize = pad->GetSize( aLayer );
 
            if( pad->GetShape( aLayer) == PAD_SHAPE::CIRCLE
                    || ( pad->GetShape( aLayer ) == PAD_SHAPE::OVAL && padSize.x == padSize.y ) )
            {
                circular = true;
            }
        }
        else if( item->Type() == PCB_VIA_T )
        {
            via = static_cast<PCB_VIA*>( item );
            circular = true;
        }
 
        // For hatch zones, use proper DRC constraints for thermal gap and spoke width,
        // just like solid zones. This ensures consistent thermal relief appearance and
        // respects pad-specific thermal spoke settings.
        if( aZone->GetFillMode() == ZONE_FILL_MODE::HATCH_PATTERN )
        {
            if( pad )
            {
                constraint = bds.m_DRCEngine->EvalRules( THERMAL_RELIEF_GAP_CONSTRAINT, pad,
                                                        aZone, aLayer );
                thermalReliefGap = constraint.GetValue().Min();
 
                constraint = bds.m_DRCEngine->EvalRules( THERMAL_SPOKE_WIDTH_CONSTRAINT, pad,
                                                        aZone, aLayer );
                spoke_w = constraint.GetValue().Opt();
 
                int spoke_max_allowed_w = std::min( pad->GetSize( aLayer ).x, pad->GetSize( aLayer ).y );
                spoke_w = std::clamp( spoke_w, constraint.Value().Min(), constraint.Value().Max() );
                spoke_w = std::min( spoke_w, spoke_max_allowed_w );
 
                if( spoke_w < aZone->GetMinThickness() )
                    continue;
            }
            else if( via )
            {
                constraint = bds.m_DRCEngine->EvalRules( THERMAL_RELIEF_GAP_CONSTRAINT, via,
                                                        aZone, aLayer );
                thermalReliefGap = constraint.GetValue().Min();
 
                constraint = bds.m_DRCEngine->EvalRules( THERMAL_SPOKE_WIDTH_CONSTRAINT, via,
                                                        aZone, aLayer );
                spoke_w = constraint.GetValue().Opt();
 
                spoke_w = std::min( spoke_w, via->GetWidth( aLayer ) );
 
                if( spoke_w < aZone->GetMinThickness() )
                    continue;
            }
            else
            {
                continue;
            }
        }
        else if( pad )
        {
            constraint = bds.m_DRCEngine->EvalRules( THERMAL_RELIEF_GAP_CONSTRAINT, pad, aZone, aLayer );
            thermalReliefGap = constraint.GetValue().Min();
 
            constraint = bds.m_DRCEngine->EvalRules( THERMAL_SPOKE_WIDTH_CONSTRAINT, pad, aZone, aLayer );
            spoke_w = constraint.GetValue().Opt();
 
            // Spoke width should ideally be smaller than the pad minor axis.
            // Otherwise the thermal shape is not really a thermal relief,
            // and the algo to count the actual number of spokes can fail
            int spoke_max_allowed_w = std::min( pad->GetSize( aLayer ).x, pad->GetSize( aLayer ).y );
 
            spoke_w = std::clamp( spoke_w, constraint.Value().Min(), constraint.Value().Max() );
 
            // ensure the spoke width is smaller than the pad minor size
            spoke_w = std::min( spoke_w, spoke_max_allowed_w );
 
            // Cannot create stubs having a width < zone min thickness
            if( spoke_w < aZone->GetMinThickness() )
                continue;
        }
        else
        {
            // We don't currently support via thermal connections *except* in a hatched zone.
            continue;
        }
 
        int spoke_half_w = spoke_w / 2;
 
        // Quick test here to possibly save us some work
        BOX2I itemBB = item->GetBoundingBox();
        itemBB.Inflate( thermalReliefGap + epsilon );
 
        if( !( itemBB.Intersects( zoneBB ) ) )
            continue;
 
        bool customSpokes = false;
 
        if( pad && pad->GetShape( aLayer ) == PAD_SHAPE::CUSTOM )
        {
            for( const std::shared_ptr<PCB_SHAPE>& primitive : pad->GetPrimitives( aLayer ) )
            {
                if( primitive->IsProxyItem() && primitive->GetShape() == SHAPE_T::SEGMENT )
                {
                    customSpokes = true;
                    break;
                }
            }
        }
 
        // Thermal spokes consist of square-ended segments from the pad center to points just
        // outside the thermal relief.  The outside end has an extra center point (which must be
        // at idx 3) which is used for testing whether or not the spoke connects to copper in the
        // parent zone.
 
        auto buildSpokesFromOrigin =
                [&]( const BOX2I& box, EDA_ANGLE angle )
                {
                    VECTOR2I center = box.GetCenter();
                    VECTOR2I half_size = KiROUND( box.GetWidth() / 2.0, box.GetHeight() / 2.0 );
 
                    // Function to find intersection of line with box edge
                    auto intersectBBox =
                            [&]( const EDA_ANGLE& spokeAngle, VECTOR2I* spoke_side ) -> VECTOR2I
                            {
                                double dx = spokeAngle.Cos();
                                double dy = spokeAngle.Sin();
 
                                // Short-circuit the axis cases because they will be degenerate in the
                                // intersection test
                                if( dx == 0 )
                                {
                                    *spoke_side = VECTOR2I( spoke_half_w, 0 );
                                    return KiROUND( 0.0, dy * half_size.y );
                                }
                                else if( dy == 0 )
                                {
                                    *spoke_side = VECTOR2I( 0, spoke_half_w );
                                    return KiROUND( dx * half_size.x, 0.0 );
                                }
 
                                // We are going to intersect with one side or the other.  Whichever
                                // we hit first is the fraction of the spoke length we keep
                                double dist_x = half_size.x / std::abs( dx );
                                double dist_y = half_size.y / std::abs( dy );
 
                                if( dist_x < dist_y )
                                {
                                    *spoke_side = KiROUND( 0.0, spoke_half_w / ( ANGLE_90 - spokeAngle ).Sin() );
                                    return KiROUND( dx * dist_x, dy * dist_x );
                                }
                                else
                                {
                                    *spoke_side = KiROUND( spoke_half_w / spokeAngle.Sin(), 0.0 );
                                    return KiROUND( dx * dist_y, dy * dist_y );
                                }
                            };
 
                    // Precalculate angles for four cardinal directions
                    const EDA_ANGLE angles[4] = {
                        EDA_ANGLE(  0.0, DEGREES_T ) + angle,  // Right
                        EDA_ANGLE( 90.0, DEGREES_T ) + angle,  // Up
                        EDA_ANGLE( 180.0, DEGREES_T ) + angle, // Left
                        EDA_ANGLE( 270.0, DEGREES_T ) + angle  // Down
                    };
 
                    // Generate four spokes in cardinal directions
                    for( const EDA_ANGLE& spokeAngle : angles )
                    {
                        VECTOR2I spoke_side;
                        VECTOR2I intersection = intersectBBox( spokeAngle, &spoke_side );
 
                        SHAPE_LINE_CHAIN spoke;
                        spoke.Append( center + spoke_side );
                        spoke.Append( center - spoke_side );
                        spoke.Append( center + intersection - spoke_side );
                        spoke.Append( center + intersection ); // test pt
                        spoke.Append( center + intersection + spoke_side );
                        spoke.SetClosed( true );
                        aSpokesList.push_back( std::move( spoke ) );
                    }
                };
 
        if( customSpokes )
        {
            SHAPE_POLY_SET   thermalPoly;
            SHAPE_LINE_CHAIN thermalOutline;
 
            pad->TransformShapeToPolygon( thermalPoly, aLayer, thermalReliefGap + epsilon, m_maxError, ERROR_OUTSIDE );
 
            if( thermalPoly.OutlineCount() )
                thermalOutline = thermalPoly.Outline( 0 );
 
            SHAPE_LINE_CHAIN padOutline = pad->GetEffectivePolygon( aLayer, ERROR_OUTSIDE )->Outline( 0 );
 
            auto trimToOutline = [&]( SEG& aSegment )
            {
                SHAPE_LINE_CHAIN::INTERSECTIONS intersections;
 
                if( padOutline.Intersect( aSegment, intersections ) )
                {
                    intersections.clear();
 
                    // Trim the segment to the thermal outline
                    if( thermalOutline.Intersect( aSegment, intersections ) )
                    {
                        aSegment.B = intersections.front().p;
                        return true;
                    }
                }
                return false;
            };
 
            for( const std::shared_ptr<PCB_SHAPE>& primitive : pad->GetPrimitives( aLayer ) )
            {
                if( primitive->IsProxyItem() && primitive->GetShape() == SHAPE_T::SEGMENT )
                {
                    SEG seg( primitive->GetStart(), primitive->GetEnd() );
                    SHAPE_LINE_CHAIN::INTERSECTIONS intersections;
 
                    RotatePoint( seg.A, pad->GetOrientation() );
                    RotatePoint( seg.B, pad->GetOrientation() );
                    seg.A += pad->ShapePos( aLayer );
                    seg.B += pad->ShapePos( aLayer );
 
                    // Make sure seg.A is the origin
                    if( !pad->GetEffectivePolygon( aLayer, ERROR_OUTSIDE )->Contains( seg.A ) )
                    {
                        // Do not create this spoke if neither point is in the pad.
                        if( !pad->GetEffectivePolygon( aLayer, ERROR_OUTSIDE )->Contains( seg.B ) )
                            continue;
 
                        seg.Reverse();
                    }
 
                    // Trim segment to pad and thermal outline polygon.
                    // If there is no intersection with the pad, don't create the spoke.
                    if( trimToOutline( seg ) )
                    {
                        VECTOR2I direction = ( seg.B - seg.A ).Resize( spoke_half_w );
                        VECTOR2I offset = direction.Perpendicular().Resize( spoke_half_w );
                        // Extend the spoke edges by half the spoke width to capture convex pad shapes
                        // with a maximum of 45 degrees.
                        SEG segL( seg.A - direction - offset, seg.B + direction - offset );
                        SEG segR( seg.A - direction + offset, seg.B + direction + offset );
 
                        // Only create this spoke if both edges intersect the pad and thermal outline
                        if( trimToOutline( segL ) && trimToOutline( segR ) )
                        {
                            // Extend the spoke by the minimum thickness for the zone to ensure full
                            // connection width
                            direction = direction.Resize( aZone->GetMinThickness() );
 
                            SHAPE_LINE_CHAIN spoke;
 
                            spoke.Append( seg.A + offset );
                            spoke.Append( seg.A - offset );
 
                            spoke.Append( segL.B + direction );
                            spoke.Append( seg.B + direction ); // test pt at index 3.
                            spoke.Append( segR.B + direction );
 
                            spoke.SetClosed( true );
                            aSpokesList.push_back( std::move( spoke ) );
                        }
                    }
                }
            }
        }
        else
        {
            EDA_ANGLE thermalSpokeAngle;
 
            // Use pad's thermal spoke angle for both solid and hatch zones.
            // This ensures custom thermal spoke templates are respected.
            if( pad )
                thermalSpokeAngle = pad->GetThermalSpokeAngle();
 
            BOX2I     spokesBox;
            VECTOR2I  position;
            EDA_ANGLE orientation;
 
            // Since the bounding-box needs to be correclty rotated we use a dummy pad to keep
            // from dirtying the real pad's cached shapes.
            if( pad )
            {
                PAD dummy_pad( *pad );
                dummy_pad.SetOrientation( ANGLE_0 );
 
                // Spokes are from center of pad shape, not from hole. So the dummy pad has no shape
                // offset and is at position 0,0
                dummy_pad.SetPosition( VECTOR2I( 0, 0 ) );
                dummy_pad.SetOffset( aLayer, VECTOR2I( 0, 0 ) );
 
                spokesBox = dummy_pad.GetBoundingBox( aLayer );
                position = pad->ShapePos( aLayer );
                orientation = pad->GetOrientation();
            }
            else if( via )
            {
                PCB_VIA dummy_via( *via );
                dummy_via.SetPosition( VECTOR2I( 0, 0 ) );
 
                spokesBox = dummy_via.GetBoundingBox( aLayer );
                position = via->GetPosition();
            }
 
            // Add half the zone mininum width to the inflate amount to account for the fact that
            // the deflation procedure will shrink the results by half the half the zone min width.
            spokesBox.Inflate( thermalReliefGap + epsilon + zone_half_width );
 
            // Yet another wrinkle: the bounding box for circles will overshoot the mark considerably
            // when the spokes are near a 45 degree increment.  So we build the spokes at 0 degrees
            // and then rotate them to the correct position.
            if( circular )
            {
                buildSpokesFromOrigin( spokesBox, ANGLE_0 );
 
                if( thermalSpokeAngle != ANGLE_0 )
                {
                    // Rotate the last four elements of aspokeslist
                    for( auto it = aSpokesList.rbegin(); it != aSpokesList.rbegin() + 4; ++it )
                        it->Rotate( thermalSpokeAngle );
                }
            }
            else
            {
                buildSpokesFromOrigin( spokesBox, thermalSpokeAngle );
            }
 
            auto spokeIter = aSpokesList.rbegin();
 
            for( int ii = 0; ii < 4; ++ii, ++spokeIter )
            {
                spokeIter->Rotate( orientation );
                spokeIter->Move( position );
            }
        }
    }
 
    for( size_t ii = 0; ii < aSpokesList.size(); ++ii )
        aSpokesList[ii].GenerateBBoxCache();
}
 
 
void ZONE_FILLER::buildHatchZoneThermalRings( const ZONE* aZone, PCB_LAYER_ID aLayer,
                                              const SHAPE_POLY_SET& aSmoothedOutline,
                                              const std::vector<BOARD_ITEM*>& aThermalConnectionPads,
                                              SHAPE_POLY_SET& aFillPolys,
                                              SHAPE_POLY_SET& aThermalRings )
{
    BOARD_DESIGN_SETTINGS& bds = m_board->GetDesignSettings();
    DRC_CONSTRAINT         constraint;
 
    for( BOARD_ITEM* item : aThermalConnectionPads )
    {
        if( !item->IsOnLayer( aLayer ) )
            continue;
 
        PAD*     pad = nullptr;
        PCB_VIA* via = nullptr;
        bool     isCircular = false;
        int      thermalGap = 0;
        int      spokeWidth = 0;
        VECTOR2I position;
        int      padRadius = 0;
 
        if( item->Type() == PCB_PAD_T )
        {
            pad = static_cast<PAD*>( item );
            VECTOR2I padSize = pad->GetSize( aLayer );
            position = pad->ShapePos( aLayer );
 
            isCircular = ( pad->GetShape( aLayer ) == PAD_SHAPE::CIRCLE
                           || ( pad->GetShape( aLayer ) == PAD_SHAPE::OVAL && padSize.x == padSize.y ) );
 
            if( isCircular )
                padRadius = std::max( padSize.x, padSize.y ) / 2;
 
            constraint = bds.m_DRCEngine->EvalRules( THERMAL_RELIEF_GAP_CONSTRAINT, pad, aZone, aLayer );
            thermalGap = constraint.GetValue().Min();
 
            constraint = bds.m_DRCEngine->EvalRules( THERMAL_SPOKE_WIDTH_CONSTRAINT, pad, aZone, aLayer );
            spokeWidth = constraint.GetValue().Opt();
 
            // Clamp spoke width to pad size
            int spokeMaxWidth = std::min( padSize.x, padSize.y );
            spokeWidth = std::min( spokeWidth, spokeMaxWidth );
        }
        else if( item->Type() == PCB_VIA_T )
        {
            via = static_cast<PCB_VIA*>( item );
            position = via->GetPosition();
            isCircular = true;
            padRadius = via->GetWidth( aLayer ) / 2;
 
            constraint = bds.m_DRCEngine->EvalRules( THERMAL_RELIEF_GAP_CONSTRAINT, via, aZone, aLayer );
            thermalGap = constraint.GetValue().Min();
 
            constraint = bds.m_DRCEngine->EvalRules( THERMAL_SPOKE_WIDTH_CONSTRAINT, via, aZone, aLayer );
            spokeWidth = constraint.GetValue().Opt();
 
            // Clamp spoke width to via diameter
            spokeWidth = std::min( spokeWidth, padRadius * 2 );
        }
        else
        {
            continue;
        }
 
        // Don't create a ring if spoke width is too small
        if( spokeWidth < aZone->GetMinThickness() )
            continue;
 
        SHAPE_POLY_SET thermalRing;
 
        if( isCircular )
        {
            // For circular pads/vias: create an arc ring
            // Ring inner radius = pad radius + thermal gap
            // Ring width = spoke width
            int ringInnerRadius = padRadius + thermalGap;
            int ringWidth = spokeWidth;
 
            TransformRingToPolygon( thermalRing, position, ringInnerRadius + ringWidth / 2,
                                    ringWidth, m_maxError, ERROR_OUTSIDE );
        }
        else
        {
            // For non-circular pads: create ring by inflating pad to outer radius,
            // then subtracting pad inflated to inner radius
            SHAPE_POLY_SET outerShape;
            SHAPE_POLY_SET innerShape;
 
            // Outer ring edge = pad + thermal gap + spoke width
            pad->TransformShapeToPolygon( outerShape, aLayer, thermalGap + spokeWidth,
                                          m_maxError, ERROR_OUTSIDE );
 
            // Inner ring edge = pad + thermal gap (this is already knocked out)
            pad->TransformShapeToPolygon( innerShape, aLayer, thermalGap,
                                          m_maxError, ERROR_OUTSIDE );
 
            thermalRing = outerShape;
            thermalRing.BooleanSubtract( innerShape );
        }
 
        // Clip the thermal ring to the zone boundary so it doesn't overflow
        thermalRing.BooleanIntersection( aSmoothedOutline );
 
        // Add the thermal ring to the fill
        aFillPolys.BooleanAdd( thermalRing );
 
        // Also collect thermal rings for hatch hole notching to ensure connectivity
        aThermalRings.BooleanAdd( thermalRing );
    }
}
 
 
bool ZONE_FILLER::addHatchFillTypeOnZone( const ZONE* aZone, PCB_LAYER_ID aLayer,
                                          PCB_LAYER_ID aDebugLayer, SHAPE_POLY_SET& aFillPolys,
                                          const SHAPE_POLY_SET& aThermalRings )
{
    // Build grid:
 
    // obviously line thickness must be > zone min thickness.
    // It can happens if a board file was edited by hand by a python script
    // Use 1 micron margin to be *sure* there is no issue in Gerber files
    // (Gbr file unit = 1 or 10 nm) due to some truncation in coordinates or calculations
    // This margin also avoid problems due to rounding coordinates in next calculations
    // that can create incorrect polygons
    int thickness = std::max( aZone->GetHatchThickness(),
                              aZone->GetMinThickness() + pcbIUScale.mmToIU( 0.001 ) );
 
    int gridsize = thickness + aZone->GetHatchGap();
    int maxError = m_board->GetDesignSettings().m_MaxError;
 
    SHAPE_POLY_SET filledPolys = aFillPolys.CloneDropTriangulation();
    // Use a area that contains the rotated bbox by orientation, and after rotate the result
    // by -orientation.
    if( !aZone->GetHatchOrientation().IsZero() )
        filledPolys.Rotate( - aZone->GetHatchOrientation() );
 
    BOX2I bbox = filledPolys.BBox( 0 );
 
    // Build hole shape
    // the hole size is aZone->GetHatchGap(), but because the outline thickness
    // is aZone->GetMinThickness(), the hole shape size must be larger
    SHAPE_LINE_CHAIN hole_base;
    int hole_size = aZone->GetHatchGap() + aZone->GetMinThickness();
    VECTOR2I corner( 0, 0 );;
    hole_base.Append( corner );
    corner.x += hole_size;
    hole_base.Append( corner );
    corner.y += hole_size;
    hole_base.Append( corner );
    corner.x = 0;
    hole_base.Append( corner );
    hole_base.SetClosed( true );
 
    // Calculate minimal area of a grid hole.
    // All holes smaller than a threshold will be removed
    double minimal_hole_area = hole_base.Area() * aZone->GetHatchHoleMinArea();
 
    // Now convert this hole to a smoothed shape:
    if( aZone->GetHatchSmoothingLevel() > 0 )
    {
        // the actual size of chamfer, or rounded corner radius is the half size
        // of the HatchFillTypeGap scaled by aZone->GetHatchSmoothingValue()
        // aZone->GetHatchSmoothingValue() = 1.0 is the max value for the chamfer or the
        // radius of corner (radius = half size of the hole)
        int smooth_value = KiROUND( aZone->GetHatchGap()
                                    * aZone->GetHatchSmoothingValue() / 2 );
 
        // Minimal optimization:
        // make smoothing only for reasonable smooth values, to avoid a lot of useless segments
        // and if the smooth value is small, use chamfer even if fillet is requested
        #define SMOOTH_MIN_VAL_MM 0.02
        #define SMOOTH_SMALL_VAL_MM 0.04
 
        if( smooth_value > pcbIUScale.mmToIU( SMOOTH_MIN_VAL_MM ) )
        {
            SHAPE_POLY_SET smooth_hole;
            smooth_hole.AddOutline( hole_base );
            int smooth_level = aZone->GetHatchSmoothingLevel();
 
            if( smooth_value < pcbIUScale.mmToIU( SMOOTH_SMALL_VAL_MM ) && smooth_level > 1 )
                smooth_level = 1;
 
            // Use a larger smooth_value to compensate the outline tickness
            // (chamfer is not visible is smooth value < outline thickess)
            smooth_value += aZone->GetMinThickness() / 2;
 
            // smooth_value cannot be bigger than the half size oh the hole:
            smooth_value = std::min( smooth_value, aZone->GetHatchGap() / 2 );
 
            // the error to approximate a circle by segments when smoothing corners by a arc
            maxError = std::max( maxError * 2, smooth_value / 20 );
 
            switch( smooth_level )
            {
            case 1:
                // Chamfer() uses the distance from a corner to create a end point
                // for the chamfer.
                hole_base = smooth_hole.Chamfer( smooth_value ).Outline( 0 );
                break;
 
            default:
                if( aZone->GetHatchSmoothingLevel() > 2 )
                    maxError /= 2;    // Force better smoothing
 
                hole_base = smooth_hole.Fillet( smooth_value, maxError ).Outline( 0 );
                break;
 
            case 0:
                break;
            };
        }
    }
 
    // Build holes
    SHAPE_POLY_SET holes;
 
    auto& defaultOffsets = m_board->GetDesignSettings().m_ZoneLayerProperties;
    auto& localOffsets = aZone->LayerProperties();
 
    VECTOR2I offset = defaultOffsets[aLayer].hatching_offset.value_or( VECTOR2I() );
 
    if( localOffsets.contains( aLayer ) && localOffsets.at( aLayer ).hatching_offset.has_value() )
        offset = localOffsets.at( aLayer ).hatching_offset.value();
 
    int x_offset = bbox.GetX() - ( bbox.GetX() ) % gridsize - gridsize;
    int y_offset = bbox.GetY() - ( bbox.GetY() ) % gridsize - gridsize;
 
 
    for( int xx = x_offset; xx <= bbox.GetRight(); xx += gridsize )
    {
        for( int yy = y_offset; yy <= bbox.GetBottom(); yy += gridsize )
        {
            // Generate hole
            SHAPE_LINE_CHAIN hole( hole_base );
            hole.Move( VECTOR2I( xx, yy ) );
 
            if( !aZone->GetHatchOrientation().IsZero() )
            {
                hole.Rotate( aZone->GetHatchOrientation() );
            }
 
            hole.Move( VECTOR2I( offset.x % gridsize, offset.y % gridsize ) );
 
            holes.AddOutline( hole );
        }
    }
 
    holes.ClearArcs();
 
    DUMP_POLYS_TO_COPPER_LAYER( holes, In10_Cu, wxT( "hatch-holes" ) );
 
    int deflated_thickness = aZone->GetHatchThickness() - aZone->GetMinThickness();
 
    // Don't let thickness drop below maxError * 2 or it might not get reinflated.
    deflated_thickness = std::max( deflated_thickness, maxError * 2 );
 
    // The fill has already been deflated to ensure GetMinThickness() so we just have to
    // account for anything beyond that.
    SHAPE_POLY_SET deflatedFilledPolys = aFillPolys.CloneDropTriangulation();
    deflatedFilledPolys.ClearArcs();
    deflatedFilledPolys.Deflate( deflated_thickness, CORNER_STRATEGY::CHAMFER_ALL_CORNERS, maxError );
    holes.BooleanIntersection( deflatedFilledPolys );
    DUMP_POLYS_TO_COPPER_LAYER( holes, In11_Cu, wxT( "fill-clipped-hatch-holes" ) );
 
    SHAPE_POLY_SET deflatedOutline = aZone->Outline()->CloneDropTriangulation();
    deflatedOutline.ClearArcs();
    deflatedOutline.Deflate( aZone->GetMinThickness(), CORNER_STRATEGY::CHAMFER_ALL_CORNERS, maxError );
    holes.BooleanIntersection( deflatedOutline );
    DUMP_POLYS_TO_COPPER_LAYER( holes, In12_Cu, wxT( "outline-clipped-hatch-holes" ) );
 
    // Now filter truncated holes to avoid small holes in pattern
    // It happens for holes near the zone outline
    for( int ii = 0; ii < holes.OutlineCount(); )
    {
        double area = holes.Outline( ii ).Area();
 
        if( area < minimal_hole_area ) // The current hole is too small: remove it
            holes.DeletePolygon( ii );
        else
            ++ii;
    }
 
    // Drop any holes that completely enclose a thermal ring to ensure thermal reliefs
    // stay connected to the hatch webbing. Only drop holes where the thermal ring is
    // entirely inside the hole; partial overlaps are kept to preserve the hatch pattern.
    if( aThermalRings.OutlineCount() > 0 )
    {
        BOX2I thermalBBox = aThermalRings.BBox();
 
        // Iterate through holes (backwards since we may delete)
        for( int holeIdx = holes.OutlineCount() - 1; holeIdx >= 0; holeIdx-- )
        {
            const SHAPE_LINE_CHAIN& hole = holes.Outline( holeIdx );
            BOX2I                   holeBBox = hole.BBox();
 
            // Quick rejection: skip if hole bbox doesn't intersect thermal rings bbox
            if( !holeBBox.Intersects( thermalBBox ) )
                continue;
 
            // Check if ANY thermal ring is completely enclosed by this hole
            for( int ringIdx = 0; ringIdx < aThermalRings.OutlineCount(); ringIdx++ )
            {
                const SHAPE_LINE_CHAIN& ring = aThermalRings.Outline( ringIdx );
                BOX2I                   ringBBox = ring.BBox();
                VECTOR2I                ringCenter = ringBBox.Centre();
 
                // Quick rejection: hole bbox must contain ring bbox
                if( !holeBBox.Contains( ringBBox ) )
                    continue;
 
                // Check 1: Is the ring center inside the hole?
                if( !hole.PointInside( ringCenter ) )
                    continue;
 
                // Check 2: Is at least one point on the ring inside the hole?
                if( ring.PointCount() == 0 || !hole.PointInside( ring.CPoint( 0 ) ) )
                    continue;
 
                // Check 3: Does the ring outline NOT intersect the hole outline?
                // If there's no intersection, the ring is fully enclosed (not touching edges)
                SHAPE_LINE_CHAIN::INTERSECTIONS intersections;
                ring.Intersect( hole, intersections );
 
                if( intersections.empty() )
                {
                    // This hole completely encloses a ring - drop it
                    holes.DeletePolygon( holeIdx );
                    break;  // Move to next hole
                }
            }
        }
    }
 
    // create grid. Useto
    // generate strictly simple polygons needed by Gerber files and Fracture()
    aFillPolys.BooleanSubtract( aFillPolys, holes );
    DUMP_POLYS_TO_COPPER_LAYER( aFillPolys, In14_Cu, wxT( "after-hatching" ) );
 
    return true;
}
 
 
bool ZONE_FILLER::refillZoneFromCache( ZONE* aZone, PCB_LAYER_ID aLayer, SHAPE_POLY_SET& aFillPolys )
{
    auto cacheKey = std::make_pair( static_cast<const ZONE*>( aZone ), aLayer );
 
    {
        std::lock_guard<std::mutex> lock( m_cacheMutex );
        auto it = m_preKnockoutFillCache.find( cacheKey );
 
        if( it == m_preKnockoutFillCache.end() )
            return false;
 
        // Restore the cached pre-knockout fill
        aFillPolys = it->second;
    }
 
    // Subtract the FILLED area of higher-priority zones (with clearance for different nets).
    // For same-net zones: subtract the filled area directly.
    // For different-net zones: subtract the filled area with DRC-evaluated clearance plus
    // extra_margin and m_maxError to match the margins used in the initial fill. Without these
    // margins, polygon approximation error can produce fills that violate clearance (issue 23053).
    BOARD_DESIGN_SETTINGS& bds = m_board->GetDesignSettings();
    int   extra_margin = pcbIUScale.mmToIU( ADVANCED_CFG::GetCfg().m_ExtraClearance );
    BOX2I zoneBBox = aZone->GetBoundingBox();
    zoneBBox.Inflate( m_worstClearance + extra_margin );
 
    auto evalRulesForItems =
            [&bds]( DRC_CONSTRAINT_T aConstraint, const BOARD_ITEM* a, const BOARD_ITEM* b,
                    PCB_LAYER_ID aEvalLayer ) -> int
            {
                DRC_CONSTRAINT c = bds.m_DRCEngine->EvalRules( aConstraint, a, b, aEvalLayer );
 
                if( c.IsNull() )
                    return -1;
                else
                    return c.GetValue().Min();
            };
 
    bool           knockoutsApplied = false;
    SHAPE_POLY_SET diffNetKnockouts;
    SHAPE_POLY_SET sameNetKnockouts;
 
    auto collectZoneKnockout =
            [&]( ZONE* otherZone )
            {
                if( otherZone == aZone )
                    return;
 
                if( !otherZone->GetLayerSet().test( aLayer ) )
                    return;
 
                if( otherZone->IsTeardropArea() && otherZone->SameNet( aZone ) )
                    return;
 
                if( !otherZone->HigherPriority( aZone ) )
                    return;
 
                if( !otherZone->GetBoundingBox().Intersects( zoneBBox ) )
                    return;
 
                if( !otherZone->HasFilledPolysForLayer( aLayer ) )
                    return;
 
                std::shared_ptr<SHAPE_POLY_SET> otherFill = otherZone->GetFilledPolysList( aLayer );
 
                if( !otherFill || otherFill->OutlineCount() == 0 )
                    return;
 
                if( otherZone->SameNet( aZone ) )
                {
                    sameNetKnockouts.Append( *otherFill );
                }
                else
                {
                    int gap = std::max( 0, evalRulesForItems( PHYSICAL_CLEARANCE_CONSTRAINT,
                                                              aZone, otherZone, aLayer ) );
 
                    gap = std::max( gap, evalRulesForItems( CLEARANCE_CONSTRAINT, aZone,
                                                            otherZone, aLayer ) );
 
                    if( gap < 0 )
                        return;
 
                    SHAPE_POLY_SET inflatedFill = *otherFill;
                    inflatedFill.Inflate( gap + extra_margin + m_maxError,
                                          CORNER_STRATEGY::ROUND_ALL_CORNERS, m_maxError );
                    diffNetKnockouts.Append( inflatedFill );
                    knockoutsApplied = true;
                }
            };
 
    forEachBoardAndFootprintZone( m_board, collectZoneKnockout );
 
    // Keepout zones are not collected here because they are already baked into the cached
    // pre-knockout fill.  They were subtracted before the initial deflate/inflate min-width
    // cycle so the cached fill already reflects keepout boundaries (issue 23515).
 
    // Subtract different-net knockouts first, then re-prune min-width
    // violations BEFORE subtracting same-net knockouts.  The fill still extends into
    // overlapping same-net zone areas at this point, which provides a natural buffer
    // that prevents the deflate/inflate cycle from creating divots at same-net
    // zone boundaries.
    if( diffNetKnockouts.OutlineCount() > 0 )
        aFillPolys.BooleanSubtract( diffNetKnockouts );
 
    if( knockoutsApplied )
        postKnockoutMinWidthPrune( aZone, aFillPolys );
 
    if( sameNetKnockouts.OutlineCount() > 0 )
        aFillPolys.BooleanSubtract( sameNetKnockouts );
 
    aFillPolys.Fracture();
 
    return true;
}