kicad/pcbnew/router/pns_node.cpp

1336 lines
32 KiB
C++

/*
* KiRouter - a push-and-(sometimes-)shove PCB router
*
* Copyright (C) 2013-2014 CERN
* Copyright (C) 2016 KiCad Developers, see AUTHORS.txt for contributors.
* Author: Tomasz Wlostowski <tomasz.wlostowski@cern.ch>
*
* 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, see <http://www.gnu.org/licenses/>.
*/
#include <vector>
#include <cassert>
#include <math/vector2d.h>
#include <geometry/seg.h>
#include <geometry/shape.h>
#include <geometry/shape_line_chain.h>
#include <geometry/shape_index.h>
#include "pns_item.h"
#include "pns_line.h"
#include "pns_node.h"
#include "pns_via.h"
#include "pns_solid.h"
#include "pns_joint.h"
#include "pns_index.h"
#include "pns_router.h"
using boost::unordered_set;
using boost::unordered_map;
namespace PNS {
#ifdef DEBUG
static boost::unordered_set<NODE*> allocNodes;
#endif
NODE::NODE()
{
wxLogTrace( "PNS", "NODE::create %p", this );
m_depth = 0;
m_root = this;
m_parent = NULL;
m_maxClearance = 800000; // fixme: depends on how thick traces are.
m_ruleResolver = NULL;
m_index = new INDEX;
#ifdef DEBUG
allocNodes.insert( this );
#endif
}
NODE::~NODE()
{
wxLogTrace( "PNS", "NODE::delete %p", this );
if( !m_children.empty() )
{
wxLogTrace( "PNS", "attempting to free a node that has kids." );
assert( false );
}
#ifdef DEBUG
if( allocNodes.find( this ) == allocNodes.end() )
{
wxLogTrace( "PNS", "attempting to free an already-free'd node." );
assert( false );
}
allocNodes.erase( this );
#endif
m_joints.clear();
for( INDEX::ITEM_SET::iterator i = m_index->begin(); i != m_index->end(); ++i )
{
if( (*i)->BelongsTo( this ) )
delete *i;
}
releaseGarbage();
unlinkParent();
delete m_index;
}
int NODE::GetClearance( const ITEM* aA, const ITEM* aB ) const
{
if( !m_ruleResolver )
return 100000;
return m_ruleResolver->Clearance( aA, aB );
}
NODE* NODE::Branch()
{
NODE* child = new NODE;
wxLogTrace( "PNS", "NODE::branch %p (parent %p)", child, this );
m_children.insert( child );
child->m_depth = m_depth + 1;
child->m_parent = this;
child->m_ruleResolver = m_ruleResolver;
child->m_root = isRoot() ? this : m_root;
// immmediate offspring of the root branch needs not copy anything.
// For the rest, deep-copy joints, overridden item map and pointers
// to stored items.
if( !isRoot() )
{
JOINT_MAP::iterator j;
for( INDEX::ITEM_SET::iterator i = m_index->begin(); i != m_index->end(); ++i )
child->m_index->Add( *i );
child->m_joints = m_joints;
child->m_override = m_override;
}
wxLogTrace( "PNS", "%d items, %d joints, %d overrides",
child->m_index->Size(), (int) child->m_joints.size(), (int) child->m_override.size() );
return child;
}
void NODE::unlinkParent()
{
if( isRoot() )
return;
m_parent->m_children.erase( this );
}
OBSTACLE_VISITOR::OBSTACLE_VISITOR( const ITEM* aItem ) :
m_item( aItem ),
m_node( NULL ),
m_override( NULL ),
m_extraClearance( 0 )
{
}
void OBSTACLE_VISITOR::SetWorld( const NODE* aNode, const NODE* aOverride )
{
m_node = aNode;
m_override = aOverride;
}
bool OBSTACLE_VISITOR::visit( ITEM* aCandidate )
{
// check if there is a more recent branch with a newer
// (possibily modified) version of this item.
if( m_override && m_override->Overrides( aCandidate ) )
return true;
return false;
}
// function object that visits potential obstacles and performs
// the actual collision refining
struct NODE::DEFAULT_OBSTACLE_VISITOR : public OBSTACLE_VISITOR
{
///> list of encountered obstacles
OBSTACLES& m_tab;
///> acccepted kinds of colliding items (solids, vias, segments, etc...)
int m_kindMask;
///> max number of hits
int m_limitCount;
///> number of items found so far
int m_matchCount;
///> additional clearance
int m_extraClearance;
bool m_differentNetsOnly;
int m_forceClearance;
DEFAULT_OBSTACLE_VISITOR( NODE::OBSTACLES& aTab, const ITEM* aItem, int aKindMask, bool aDifferentNetsOnly ) :
OBSTACLE_VISITOR( aItem ),
m_tab( aTab ),
m_kindMask( aKindMask ),
m_limitCount( -1 ),
m_matchCount( 0 ),
m_extraClearance( 0 ),
m_differentNetsOnly( aDifferentNetsOnly ),
m_forceClearance( -1 )
{
if( aItem && aItem->Kind() == ITEM::LINE_T )
{
m_extraClearance += static_cast<const LINE*>( aItem )->Width() / 2;
}
}
void SetCountLimit( int aLimit )
{
m_limitCount = aLimit;
}
bool operator()( ITEM* aCandidate ) override
{
if( !aCandidate->OfKind( m_kindMask ) )
return true;
if( visit( aCandidate ) )
return true;
int clearance = m_extraClearance + m_node->GetClearance( aCandidate, m_item );
if( aCandidate->Kind() == ITEM::LINE_T ) // this should never happen.
{
assert( false );
clearance += static_cast<LINE*>( aCandidate )->Width() / 2;
}
if( m_forceClearance >= 0 )
clearance = m_forceClearance;
if( !aCandidate->Collide( m_item, clearance, m_differentNetsOnly ) )
return true;
OBSTACLE obs;
obs.m_item = aCandidate;
obs.m_head = m_item;
m_tab.push_back( obs );
m_matchCount++;
if( m_limitCount > 0 && m_matchCount >= m_limitCount )
return false;
return true;
};
};
int NODE::QueryColliding( const ITEM* aItem, OBSTACLE_VISITOR& aVisitor )
{
aVisitor.SetWorld( this, NULL );
m_index->Query( aItem, m_maxClearance, aVisitor );
// if we haven't found enough items, look in the root branch as well.
if( !isRoot() )
{
aVisitor.SetWorld( m_root, this );
m_root->m_index->Query( aItem, m_maxClearance, aVisitor );
}
return 0;
}
int NODE::QueryColliding( const ITEM* aItem,
NODE::OBSTACLES& aObstacles, int aKindMask, int aLimitCount, bool aDifferentNetsOnly, int aForceClearance )
{
DEFAULT_OBSTACLE_VISITOR visitor( aObstacles, aItem, aKindMask, aDifferentNetsOnly );
#ifdef DEBUG
assert( allocNodes.find( this ) != allocNodes.end() );
#endif
visitor.SetCountLimit( aLimitCount );
visitor.SetWorld( this, NULL );
visitor.m_forceClearance = aForceClearance;
// first, look for colliding items in the local index
m_index->Query( aItem, m_maxClearance, visitor );
// if we haven't found enough items, look in the root branch as well.
if( !isRoot() && ( visitor.m_matchCount < aLimitCount || aLimitCount < 0 ) )
{
visitor.SetWorld( m_root, this );
m_root->m_index->Query( aItem, m_maxClearance, visitor );
}
return aObstacles.size();
}
NODE::OPT_OBSTACLE NODE::NearestObstacle( const LINE* aItem, int aKindMask,
const std::set<ITEM*>* aRestrictedSet )
{
OBSTACLES obs_list;
bool found_isects = false;
const SHAPE_LINE_CHAIN& line = aItem->CLine();
obs_list.reserve( 100 );
int n = 0;
for( int i = 0; i < line.SegmentCount(); i++ )
{
const SEGMENT s( *aItem, line.CSegment( i ) );
n += QueryColliding( &s, obs_list, aKindMask );
}
if( aItem->EndsWithVia() )
n += QueryColliding( &aItem->Via(), obs_list, aKindMask );
if( !n )
return OPT_OBSTACLE();
LINE& aLine = (LINE&) *aItem;
OBSTACLE nearest;
nearest.m_item = NULL;
nearest.m_distFirst = INT_MAX;
for( OBSTACLE obs : obs_list )
{
VECTOR2I ip_first, ip_last;
int dist_max = INT_MIN;
if( aRestrictedSet && aRestrictedSet->find( obs.m_item ) == aRestrictedSet->end() )
continue;
std::vector<SHAPE_LINE_CHAIN::INTERSECTION> isect_list;
int clearance = GetClearance( obs.m_item, &aLine );
SHAPE_LINE_CHAIN hull = obs.m_item->Hull( clearance, aItem->Width() );
if( aLine.EndsWithVia() )
{
clearance = GetClearance( obs.m_item, &aLine.Via() );
SHAPE_LINE_CHAIN viaHull = aLine.Via().Hull( clearance, aItem->Width() );
viaHull.Intersect( hull, isect_list );
for( SHAPE_LINE_CHAIN::INTERSECTION isect : isect_list )
{
int dist = aLine.CLine().Length() +
( isect.p - aLine.Via().Pos() ).EuclideanNorm();
if( dist < nearest.m_distFirst )
{
found_isects = true;
nearest.m_distFirst = dist;
nearest.m_ipFirst = isect.p;
nearest.m_item = obs.m_item;
nearest.m_hull = hull;
}
if( dist > dist_max )
{
dist_max = dist;
ip_last = isect.p;
}
}
}
isect_list.clear();
hull.Intersect( aLine.CLine(), isect_list );
for( SHAPE_LINE_CHAIN::INTERSECTION isect : isect_list )
{
int dist = aLine.CLine().PathLength( isect.p );
if( dist < nearest.m_distFirst )
{
found_isects = true;
nearest.m_distFirst = dist;
nearest.m_ipFirst = isect.p;
nearest.m_item = obs.m_item;
nearest.m_hull = hull;
}
if( dist > dist_max )
{
dist_max = dist;
ip_last = isect.p;
}
}
nearest.m_ipLast = ip_last;
nearest.m_distLast = dist_max;
}
if( !found_isects )
nearest.m_item = obs_list[0].m_item;
return nearest;
}
NODE::OPT_OBSTACLE NODE::CheckColliding( const ITEM_SET& aSet, int aKindMask )
{
for( const ITEM* item : aSet.CItems() )
{
OPT_OBSTACLE obs = CheckColliding( item, aKindMask );
if( obs )
return obs;
}
return OPT_OBSTACLE();
}
NODE::OPT_OBSTACLE NODE::CheckColliding( const ITEM* aItemA, int aKindMask )
{
OBSTACLES obs;
obs.reserve( 100 );
if( aItemA->Kind() == ITEM::LINE_T )
{
int n = 0;
const LINE* line = static_cast<const LINE*>( aItemA );
const SHAPE_LINE_CHAIN& l = line->CLine();
for( int i = 0; i < l.SegmentCount(); i++ )
{
const SEGMENT s( *line, l.CSegment( i ) );
n += QueryColliding( &s, obs, aKindMask, 1 );
if( n )
return OPT_OBSTACLE( obs[0] );
}
if( line->EndsWithVia() )
{
n += QueryColliding( &line->Via(), obs, aKindMask, 1 );
if( n )
return OPT_OBSTACLE( obs[0] );
}
}
else if( QueryColliding( aItemA, obs, aKindMask, 1 ) > 0 )
return OPT_OBSTACLE( obs[0] );
return OPT_OBSTACLE();
}
bool NODE::CheckColliding( const ITEM* aItemA, const ITEM* aItemB, int aKindMask, int aForceClearance )
{
assert( aItemB );
int clearance;
if( aForceClearance >= 0 )
clearance = aForceClearance;
else
clearance = GetClearance( aItemA, aItemB );
// fixme: refactor
if( aItemA->Kind() == ITEM::LINE_T )
clearance += static_cast<const LINE*>( aItemA )->Width() / 2;
if( aItemB->Kind() == ITEM::LINE_T )
clearance += static_cast<const LINE*>( aItemB )->Width() / 2;
return aItemA->Collide( aItemB, clearance );
}
struct HIT_VISITOR : public OBSTACLE_VISITOR
{
ITEM_SET& m_items;
const VECTOR2I& m_point;
HIT_VISITOR( ITEM_SET& aTab, const VECTOR2I& aPoint ) :
OBSTACLE_VISITOR( NULL ),
m_items( aTab ), m_point( aPoint )
{}
bool operator()( ITEM* aItem ) override
{
SHAPE_CIRCLE cp( m_point, 0 );
int cl = 0;
if( aItem->Shape()->Collide( &cp, cl ) )
m_items.Add( aItem );
return true;
}
};
const ITEM_SET NODE::HitTest( const VECTOR2I& aPoint ) const
{
ITEM_SET items;
// fixme: we treat a point as an infinitely small circle - this is inefficient.
SHAPE_CIRCLE s( aPoint, 0 );
HIT_VISITOR visitor( items, aPoint );
visitor.SetWorld( this, NULL );
m_index->Query( &s, m_maxClearance, visitor );
if( !isRoot() ) // fixme: could be made cleaner
{
ITEM_SET items_root;
visitor.SetWorld( m_root, NULL );
HIT_VISITOR visitor_root( items_root, aPoint );
m_root->m_index->Query( &s, m_maxClearance, visitor_root );
for( ITEM* item : items_root.Items() )
{
if( !Overrides( item ) )
items.Add( item );
}
}
return items;
}
void NODE::addSolid( SOLID* aSolid )
{
linkJoint( aSolid->Pos(), aSolid->Layers(), aSolid->Net(), aSolid );
m_index->Add( aSolid );
}
void NODE::Add( std::unique_ptr< SOLID > aSolid )
{
aSolid->SetOwner( this );
addSolid( aSolid.release() );
}
void NODE::addVia( VIA* aVia )
{
linkJoint( aVia->Pos(), aVia->Layers(), aVia->Net(), aVia );
m_index->Add( aVia );
}
void NODE::Add( std::unique_ptr< VIA > aVia )
{
aVia->SetOwner( this );
addVia( aVia.release() );
}
void NODE::Add( LINE& aLine, bool aAllowRedundant )
{
assert( !aLine.IsLinked() );
SHAPE_LINE_CHAIN& l = aLine.Line();
for( int i = 0; i < l.SegmentCount(); i++ )
{
SEG s = l.CSegment( i );
if( s.A != s.B )
{
SEGMENT* rseg;
if( !aAllowRedundant &&
(rseg = findRedundantSegment( s.A, s.B, aLine.Layers(), aLine.Net() )) )
{
// another line could be referencing this segment too :(
aLine.LinkSegment( rseg );
}
else
{
std::unique_ptr< SEGMENT > newseg( new SEGMENT( aLine, s ) );
aLine.LinkSegment( newseg.get() );
Add( std::move( newseg ), true );
}
}
}
}
void NODE::addSegment( SEGMENT* aSeg )
{
linkJoint( aSeg->Seg().A, aSeg->Layers(), aSeg->Net(), aSeg );
linkJoint( aSeg->Seg().B, aSeg->Layers(), aSeg->Net(), aSeg );
m_index->Add( aSeg );
}
void NODE::Add( std::unique_ptr< SEGMENT > aSegment, bool aAllowRedundant )
{
if( aSegment->Seg().A == aSegment->Seg().B )
{
wxLogTrace( "PNS", "attempting to add a segment with same end coordinates, ignoring." );
return;
}
if( !aAllowRedundant && findRedundantSegment( aSegment.get() ) )
return;
aSegment->SetOwner( this );
addSegment( aSegment.release() );
}
void NODE::Add( std::unique_ptr< ITEM > aItem, bool aAllowRedundant )
{
switch( aItem->Kind() )
{
case ITEM::SOLID_T:
Add( ItemCast<SOLID>( std::move( aItem ) ) );
break;
case ITEM::SEGMENT_T:
Add( ItemCast<SEGMENT>( std::move( aItem ) ), aAllowRedundant );
break;
case ITEM::LINE_T:
assert( false );
break;
case ITEM::VIA_T:
Add( ItemCast<VIA>( std::move( aItem ) ) );
break;
default:
assert( false );
}
}
void NODE::doRemove( ITEM* aItem )
{
// case 1: removing an item that is stored in the root node from any branch:
// mark it as overridden, but do not remove
if( aItem->BelongsTo( m_root ) && !isRoot() )
m_override.insert( aItem );
// case 2: the item belongs to this branch or a parent, non-root branch,
// or the root itself and we are the root: remove from the index
else if( !aItem->BelongsTo( m_root ) || isRoot() )
m_index->Remove( aItem );
// the item belongs to this particular branch: un-reference it
if( aItem->BelongsTo( this ) )
{
aItem->SetOwner( NULL );
m_root->m_garbageItems.insert( aItem );
}
}
void NODE::removeSegmentIndex( SEGMENT* aSeg )
{
unlinkJoint( aSeg->Seg().A, aSeg->Layers(), aSeg->Net(), aSeg );
unlinkJoint( aSeg->Seg().B, aSeg->Layers(), aSeg->Net(), aSeg );
}
void NODE::removeViaIndex( VIA* aVia )
{
// We have to split a single joint (associated with a via, binding together multiple layers)
// into multiple independent joints. As I'm a lazy bastard, I simply delete the via and all its links and re-insert them.
JOINT::HASH_TAG tag;
VECTOR2I p( aVia->Pos() );
LAYER_RANGE vLayers( aVia->Layers() );
int net = aVia->Net();
JOINT* jt = FindJoint( p, vLayers.Start(), net );
JOINT::LINKED_ITEMS links( jt->LinkList() );
tag.net = net;
tag.pos = p;
bool split;
do
{
split = false;
std::pair<JOINT_MAP::iterator, JOINT_MAP::iterator> range = m_joints.equal_range( tag );
if( range.first == m_joints.end() )
break;
// find and remove all joints containing the via to be removed
for( JOINT_MAP::iterator f = range.first; f != range.second; ++f )
{
if( aVia->LayersOverlap( &f->second ) )
{
m_joints.erase( f );
split = true;
break;
}
}
} while( split );
// and re-link them, using the former via's link list
for(ITEM* item : links)
{
if( item != aVia )
linkJoint( p, item->Layers(), net, item );
}
}
void NODE::removeSolidIndex( SOLID* aSolid )
{
// fixme: this fucks up the joints, but it's only used for marking colliding obstacles for the moment, so we don't care.
}
void NODE::Replace( ITEM* aOldItem, std::unique_ptr< ITEM > aNewItem )
{
Remove( aOldItem );
Add( std::move( aNewItem ) );
}
void NODE::Replace( LINE& aOldLine, LINE& aNewLine )
{
Remove( aOldLine );
Add( aNewLine );
}
void NODE::Remove( SOLID* aSolid )
{
removeSolidIndex( aSolid );
doRemove( aSolid );
}
void NODE::Remove( VIA* aVia )
{
removeViaIndex( aVia );
doRemove( aVia );
}
void NODE::Remove( SEGMENT* aSegment )
{
removeSegmentIndex( aSegment );
doRemove( aSegment );
}
void NODE::Remove( ITEM* aItem )
{
switch( aItem->Kind() )
{
case ITEM::SOLID_T:
Remove( static_cast<SOLID*>( aItem ) );
break;
case ITEM::SEGMENT_T:
Remove( static_cast<SEGMENT*>( aItem ) );
break;
case ITEM::LINE_T:
assert( false );
break;
case ITEM::VIA_T:
Remove( static_cast<VIA*>( aItem ) );
break;
default:
break;
}
}
void NODE::Remove( LINE& aLine )
{
// LINE does not have a seperate remover, as LINEs are never truly a member of the tree
std::vector<SEGMENT*>& segRefs = aLine.LinkedSegments();
for( SEGMENT* seg : segRefs )
{
Remove( seg );
}
aLine.SetOwner( nullptr );
aLine.ClearSegmentLinks();
}
void NODE::followLine( SEGMENT* aCurrent, bool aScanDirection, int& aPos,
int aLimit, VECTOR2I* aCorners, SEGMENT** aSegments, bool& aGuardHit,
bool aStopAtLockedJoints )
{
bool prevReversed = false;
const VECTOR2I guard = aScanDirection ? aCurrent->Seg().B : aCurrent->Seg().A;
for( int count = 0 ; ; ++count )
{
const VECTOR2I p =
( aScanDirection ^ prevReversed ) ? aCurrent->Seg().B : aCurrent->Seg().A;
const JOINT* jt = FindJoint( p, aCurrent );
assert( jt );
aCorners[aPos] = jt->Pos();
aSegments[aPos] = aCurrent;
aPos += ( aScanDirection ? 1 : -1 );
if( count && guard == p)
{
aSegments[aPos] = NULL;
aGuardHit = true;
break;
}
bool locked = aStopAtLockedJoints ? jt->IsLocked() : false;
if( locked || !jt->IsLineCorner() || aPos < 0 || aPos == aLimit )
break;
aCurrent = jt->NextSegment( aCurrent );
prevReversed =
( jt->Pos() == ( aScanDirection ? aCurrent->Seg().B : aCurrent->Seg().A ) );
}
}
const LINE NODE::AssembleLine( SEGMENT* aSeg, int* aOriginSegmentIndex, bool aStopAtLockedJoints )
{
const int MaxVerts = 1024 * 16;
VECTOR2I corners[MaxVerts + 1];
SEGMENT* segs[MaxVerts + 1];
LINE pl;
bool guardHit = false;
int i_start = MaxVerts / 2, i_end = i_start + 1;
pl.SetWidth( aSeg->Width() );
pl.SetLayers( aSeg->Layers() );
pl.SetNet( aSeg->Net() );
pl.SetOwner( this );
followLine( aSeg, false, i_start, MaxVerts, corners, segs, guardHit, aStopAtLockedJoints );
if( !guardHit )
followLine( aSeg, true, i_end, MaxVerts, corners, segs, guardHit, aStopAtLockedJoints );
int n = 0;
SEGMENT* prev_seg = NULL;
bool originSet = false;
for( int i = i_start + 1; i < i_end; i++ )
{
const VECTOR2I& p = corners[i];
pl.Line().Append( p );
if( segs[i] && prev_seg != segs[i] )
{
pl.LinkSegment( segs[i] );
// latter condition to avoid loops
if( segs[i] == aSeg && aOriginSegmentIndex && !originSet )
{
*aOriginSegmentIndex = n;
originSet = true;
}
n++;
}
prev_seg = segs[i];
}
assert( pl.SegmentCount() != 0 );
return pl;
}
void NODE::FindLineEnds( const LINE& aLine, JOINT& aA, JOINT& aB )
{
aA = *FindJoint( aLine.CPoint( 0 ), &aLine );
aB = *FindJoint( aLine.CPoint( -1 ), &aLine );
}
#if 0
void NODE::MapConnectivity( JOINT* aStart, std::vector<JOINT*>& aFoundJoints )
{
std::deque<JOINT*> searchQueue;
std::set<JOINT*> processed;
searchQueue.push_back( aStart );
processed.insert( aStart );
while( !searchQueue.empty() )
{
JOINT* current = searchQueue.front();
searchQueue.pop_front();
for( ITEM* item : current->LinkList() )
{
if( item->OfKind( ITEM::SEGMENT_T ) )
{
SEGMENT* seg = static_cast<SEGMENT *>( item );
JOINT* a = FindJoint( seg->Seg().A, seg );
JOINT* b = FindJoint( seg->Seg().B, seg );
JOINT* next = ( *a == *current ) ? b : a;
if( processed.find( next ) == processed.end() )
{
processed.insert( next );
searchQueue.push_back( next );
}
}
}
}
for(JOINT* jt : processed)
aFoundJoints.push_back( jt );
}
#endif
int NODE::FindLinesBetweenJoints( JOINT& aA, JOINT& aB, std::vector<LINE>& aLines )
{
for( ITEM* item : aA.LinkList() )
{
if( item->Kind() == ITEM::SEGMENT_T )
{
SEGMENT* seg = static_cast<SEGMENT*>( item );
LINE line = AssembleLine( seg );
if( !line.Layers().Overlaps( aB.Layers() ) )
continue;
JOINT j_start, j_end;
FindLineEnds( line, j_start, j_end );
int id_start = line.CLine().Find( aA.Pos() );
int id_end = line.CLine().Find( aB.Pos() );
if( id_end < id_start )
std::swap( id_end, id_start );
if( id_start >= 0 && id_end >= 0 )
{
line.ClipVertexRange( id_start, id_end );
aLines.push_back( line );
}
}
}
return 0;
}
JOINT* NODE::FindJoint( const VECTOR2I& aPos, int aLayer, int aNet )
{
JOINT::HASH_TAG tag;
tag.net = aNet;
tag.pos = aPos;
JOINT_MAP::iterator f = m_joints.find( tag ), end = m_joints.end();
if( f == end && !isRoot() )
{
end = m_root->m_joints.end();
f = m_root->m_joints.find( tag ); // m_root->FindJoint(aPos, aLayer, aNet);
}
if( f == end )
return NULL;
while( f != end )
{
if( f->second.Layers().Overlaps( aLayer ) )
return &f->second;
++f;
}
return NULL;
}
void NODE::LockJoint( const VECTOR2I& aPos, const ITEM* aItem, bool aLock )
{
JOINT& jt = touchJoint( aPos, aItem->Layers(), aItem->Net() );
jt.Lock( aLock );
}
JOINT& NODE::touchJoint( const VECTOR2I& aPos, const LAYER_RANGE& aLayers, int aNet )
{
JOINT::HASH_TAG tag;
tag.pos = aPos;
tag.net = aNet;
// try to find the joint in this node.
JOINT_MAP::iterator f = m_joints.find( tag );
std::pair<JOINT_MAP::iterator, JOINT_MAP::iterator> range;
// not found and we are not root? find in the root and copy results here.
if( f == m_joints.end() && !isRoot() )
{
range = m_root->m_joints.equal_range( tag );
for( f = range.first; f != range.second; ++f )
m_joints.insert( *f );
}
// now insert and combine overlapping joints
JOINT jt( aPos, aLayers, aNet );
bool merged;
do
{
merged = false;
range = m_joints.equal_range( tag );
if( range.first == m_joints.end() )
break;
for( f = range.first; f != range.second; ++f )
{
if( aLayers.Overlaps( f->second.Layers() ) )
{
jt.Merge( f->second );
m_joints.erase( f );
merged = true;
break;
}
}
}
while( merged );
return m_joints.insert( TagJointPair( tag, jt ) )->second;
}
void JOINT::Dump() const
{
wxLogTrace( "PNS", "joint layers %d-%d, net %d, pos %s, links: %d", m_layers.Start(),
m_layers.End(), m_tag.net, m_tag.pos.Format().c_str(), LinkCount() );
}
void NODE::linkJoint( const VECTOR2I& aPos, const LAYER_RANGE& aLayers,
int aNet, ITEM* aWhere )
{
JOINT& jt = touchJoint( aPos, aLayers, aNet );
jt.Link( aWhere );
}
void NODE::unlinkJoint( const VECTOR2I& aPos, const LAYER_RANGE& aLayers,
int aNet, ITEM* aWhere )
{
// fixme: remove dangling joints
JOINT& jt = touchJoint( aPos, aLayers, aNet );
jt.Unlink( aWhere );
}
void NODE::Dump( bool aLong )
{
#if 0
boost::unordered_set<SEGMENT*> all_segs;
SHAPE_INDEX_LIST<ITEM*>::iterator i;
for( i = m_items.begin(); i != m_items.end(); i++ )
{
if( (*i)->GetKind() == ITEM::SEGMENT_T )
all_segs.insert( static_cast<SEGMENT*>( *i ) );
}
if( !isRoot() )
{
for( i = m_root->m_items.begin(); i != m_root->m_items.end(); i++ )
{
if( (*i)->GetKind() == ITEM::SEGMENT_T && !overrides( *i ) )
all_segs.insert( static_cast<SEGMENT*>(*i) );
}
}
JOINT_MAP::iterator j;
if( aLong )
for( j = m_joints.begin(); j != m_joints.end(); ++j )
{
wxLogTrace( "PNS", "joint : %s, links : %d\n",
j->second.GetPos().Format().c_str(), j->second.LinkCount() );
JOINT::LINKED_ITEMS::const_iterator k;
for( k = j->second.GetLinkList().begin(); k != j->second.GetLinkList().end(); ++k )
{
const ITEM* m_item = *k;
switch( m_item->GetKind() )
{
case ITEM::SEGMENT_T:
{
const SEGMENT* seg = static_cast<const SEGMENT*>( m_item );
wxLogTrace( "PNS", " -> seg %s %s\n", seg->GetSeg().A.Format().c_str(),
seg->GetSeg().B.Format().c_str() );
break;
}
default:
break;
}
}
}
int lines_count = 0;
while( !all_segs.empty() )
{
SEGMENT* s = *all_segs.begin();
LINE* l = AssembleLine( s );
LINE::LinkedSegments* seg_refs = l->GetLinkedSegments();
if( aLong )
wxLogTrace( "PNS", "Line: %s, net %d ", l->GetLine().Format().c_str(), l->GetNet() );
for( std::vector<SEGMENT*>::iterator j = seg_refs->begin(); j != seg_refs->end(); ++j )
{
wxLogTrace( "PNS", "%s ", (*j)->GetSeg().A.Format().c_str() );
if( j + 1 == seg_refs->end() )
wxLogTrace( "PNS", "%s\n", (*j)->GetSeg().B.Format().c_str() );
all_segs.erase( *j );
}
lines_count++;
}
wxLogTrace( "PNS", "Local joints: %d, lines : %d \n", m_joints.size(), lines_count );
#endif
}
void NODE::GetUpdatedItems( ITEM_VECTOR& aRemoved, ITEM_VECTOR& aAdded )
{
aRemoved.reserve( m_override.size() );
aAdded.reserve( m_index->Size() );
if( isRoot() )
return;
for( ITEM* item : m_override )
aRemoved.push_back( item );
for( INDEX::ITEM_SET::iterator i = m_index->begin(); i != m_index->end(); ++i )
aAdded.push_back( *i );
}
void NODE::releaseChildren()
{
// copy the kids as the NODE destructor erases the item from the parent node.
std::set<NODE*> kids = m_children;
for( NODE* node : kids )
{
node->releaseChildren();
delete node;
}
}
void NODE::releaseGarbage()
{
if( !isRoot() )
return;
for( ITEM* item : m_garbageItems )
{
if( !item->BelongsTo( this ) )
delete item;
}
m_garbageItems.clear();
}
void NODE::Commit( NODE* aNode )
{
if( aNode->isRoot() )
return;
for( ITEM* item : aNode->m_override )
Remove( item );
for( INDEX::ITEM_SET::iterator i = aNode->m_index->begin();
i != aNode->m_index->end(); ++i )
{
(*i)->SetRank( -1 );
(*i)->Unmark();
Add( std::unique_ptr<ITEM>( *i ) );
}
releaseChildren();
releaseGarbage();
}
void NODE::KillChildren()
{
assert( isRoot() );
releaseChildren();
}
void NODE::AllItemsInNet( int aNet, std::set<ITEM*>& aItems )
{
INDEX::NET_ITEMS_LIST* l_cur = m_index->GetItemsForNet( aNet );
if( l_cur )
{
for( ITEM*item : *l_cur )
aItems.insert( item );
}
if( !isRoot() )
{
INDEX::NET_ITEMS_LIST* l_root = m_root->m_index->GetItemsForNet( aNet );
if( l_root )
for( INDEX::NET_ITEMS_LIST::iterator i = l_root->begin(); i!= l_root->end(); ++i )
if( !Overrides( *i ) )
aItems.insert( *i );
}
}
void NODE::ClearRanks( int aMarkerMask )
{
for( INDEX::ITEM_SET::iterator i = m_index->begin(); i != m_index->end(); ++i )
{
(*i)->SetRank( -1 );
(*i)->Mark( (*i)->Marker() & (~aMarkerMask) );
}
}
int NODE::FindByMarker( int aMarker, ITEM_SET& aItems )
{
for( INDEX::ITEM_SET::iterator i = m_index->begin(); i != m_index->end(); ++i )
{
if( (*i)->Marker() & aMarker )
aItems.Add( *i );
}
return 0;
}
int NODE::RemoveByMarker( int aMarker )
{
std::list<ITEM*> garbage;
for( INDEX::ITEM_SET::iterator i = m_index->begin(); i != m_index->end(); ++i )
{
if( (*i)->Marker() & aMarker )
{
garbage.push_back( *i );
}
}
for( std::list<ITEM*>::const_iterator i = garbage.begin(), end = garbage.end(); i != end; ++i )
{
Remove( *i );
}
return 0;
}
SEGMENT* NODE::findRedundantSegment( const VECTOR2I& A, const VECTOR2I& B, const LAYER_RANGE& lr,
int aNet )
{
JOINT* jtStart = FindJoint( A, lr.Start(), aNet );
if( !jtStart )
return nullptr;
for( ITEM* item : jtStart->LinkList() )
{
if( item->OfKind( ITEM::SEGMENT_T ) )
{
SEGMENT* seg2 = (SEGMENT*)item;
const VECTOR2I a2( seg2->Seg().A );
const VECTOR2I b2( seg2->Seg().B );
if( seg2->Layers().Start() == lr.Start() &&
((A == a2 && B == b2) || (A == b2 && B == a2)) )
return seg2;
}
}
return nullptr;
}
SEGMENT* NODE::findRedundantSegment( SEGMENT* aSeg )
{
return findRedundantSegment( aSeg->Seg().A, aSeg->Seg().B, aSeg->Layers(), aSeg->Net() );
}
ITEM *NODE::FindItemByParent( const BOARD_CONNECTED_ITEM* aParent )
{
INDEX::NET_ITEMS_LIST* l_cur = m_index->GetItemsForNet( aParent->GetNetCode() );
for( ITEM*item : *l_cur )
if( item->Parent() == aParent )
return item;
return NULL;
}
}