Skinner.cpp

Go to the documentation of this file.

/**
 * MOAB, a Mesh-Oriented datABase, is a software component for creating,
 * storing and accessing finite element mesh data.
 *
 * Copyright 2004 Sandia Corporation.  Under the terms of Contract
 * DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government
 * retains certain rights in this software.
 *
 * This library is free software; you can redistribute it and/or
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
 * version 2.1 of the License, or (at your option) any later version.
 *
 */
 
#ifdef __MFC_VER
#pragma warning( disable : 4786 )
#endif
 
#ifdef WIN32               /* windows */
#define _USE_MATH_DEFINES  // For M_PI
#endif
#include "moab/Skinner.hpp"
#include "moab/Range.hpp"
#include "moab/CN.hpp"
#include <vector>
#include <set>
#include <algorithm>
#include <cmath>
#include <cassert>
#include <iostream>
#include "moab/Util.hpp"
#include "Internals.hpp"
#include "MBTagConventions.hpp"
#include "moab/Core.hpp"
#include "AEntityFactory.hpp"
#include "moab/ScdInterface.hpp"
 
#ifdef M_PI
#define SKINNER_PI M_PI
#else
#define SKINNER_PI 3.1415926535897932384626
#endif
 
namespace moab
{
 
Skinner::~Skinner()
{
    // delete the adjacency tag
}
 
ErrorCode Skinner::initialize()
{
    // go through and mark all the target dimension entities
    // that already exist as not deleteable
    // also get the connectivity tags for each type
    // also populate adjacency information
    EntityType type;
    DimensionPair target_ent_types = CN::TypeDimensionMap[mTargetDim];
 
    void* null_ptr = NULL;
 
    ErrorCode result = thisMB->tag_get_handle( "skinner adj", sizeof( void* ), MB_TYPE_OPAQUE, mAdjTag,
                                               MB_TAG_DENSE | MB_TAG_CREAT, &null_ptr );MB_CHK_ERR( result );
 
    if( mDeletableMBTag == 0 )
    {
        result =
            thisMB->tag_get_handle( "skinner deletable", 1, MB_TYPE_BIT, mDeletableMBTag, MB_TAG_BIT | MB_TAG_CREAT );MB_CHK_ERR( result );
    }
 
    Range entities;
 
    // go through each type at this dimension
    for( type = target_ent_types.first; type <= target_ent_types.second; ++type )
    {
        // get the entities of this type in the MB
        thisMB->get_entities_by_type( 0, type, entities );
 
        // go through each entity of this type in the MB
        // and set its deletable tag to NO
        Range::iterator iter, end_iter;
        end_iter = entities.end();
        for( iter = entities.begin(); iter != end_iter; ++iter )
        {
            unsigned char bit = 0x1;
            result            = thisMB->tag_set_data( mDeletableMBTag, &( *iter ), 1, &bit );
            assert( MB_SUCCESS == result );
            // add adjacency information too
            if( TYPE_FROM_HANDLE( *iter ) != MBVERTEX ) add_adjacency( *iter );
        }
    }
 
    return MB_SUCCESS;
}
 
ErrorCode Skinner::deinitialize()
{
    ErrorCode result;
    if( 0 != mDeletableMBTag )
    {
        result          = thisMB->tag_delete( mDeletableMBTag );
        mDeletableMBTag = 0;MB_CHK_ERR( result );
    }
 
    // remove the adjacency tag
    std::vector< std::vector< EntityHandle >* > adj_arr;
    std::vector< std::vector< EntityHandle >* >::iterator i;
    if( 0 != mAdjTag )
    {
        for( EntityType t = MBVERTEX; t != MBMAXTYPE; ++t )
        {
            Range entities;
            result = thisMB->get_entities_by_type_and_tag( 0, t, &mAdjTag, 0, 1, entities );MB_CHK_ERR( result );
            adj_arr.resize( entities.size() );
            result = thisMB->tag_get_data( mAdjTag, entities, &adj_arr[0] );MB_CHK_ERR( result );
            for( i = adj_arr.begin(); i != adj_arr.end(); ++i )
                delete *i;
        }
 
        result  = thisMB->tag_delete( mAdjTag );
        mAdjTag = 0;MB_CHK_ERR( result );
    }
 
    return MB_SUCCESS;
}
 
ErrorCode Skinner::add_adjacency( EntityHandle entity )
{
    std::vector< EntityHandle >* adj = NULL;
    const EntityHandle* nodes;
    int num_nodes;
    ErrorCode result = thisMB->get_connectivity( entity, nodes, num_nodes, true );MB_CHK_ERR( result );
    const EntityHandle* iter = std::min_element( nodes, nodes + num_nodes );
 
    if( iter == nodes + num_nodes ) return MB_SUCCESS;
 
    // add this entity to the node
    if( thisMB->tag_get_data( mAdjTag, iter, 1, &adj ) == MB_SUCCESS && adj != NULL )
    {
        adj->push_back( entity );
    }
    // create a new vector and add it
    else
    {
        adj = new std::vector< EntityHandle >;
        adj->push_back( entity );
        result = thisMB->tag_set_data( mAdjTag, iter, 1, &adj );MB_CHK_ERR( result );
    }
 
    return MB_SUCCESS;
}
 
void Skinner::add_adjacency( EntityHandle entity, const EntityHandle* nodes, const int num_nodes )
{
    std::vector< EntityHandle >* adj = NULL;
    const EntityHandle* iter         = std::min_element( nodes, nodes + num_nodes );
 
    if( iter == nodes + num_nodes ) return;
 
    // should not be setting adjacency lists in ho-nodes
    assert( TYPE_FROM_HANDLE( entity ) == MBPOLYGON ||
            num_nodes == CN::VerticesPerEntity( TYPE_FROM_HANDLE( entity ) ) );
 
    // add this entity to the node
    if( thisMB->tag_get_data( mAdjTag, iter, 1, &adj ) == MB_SUCCESS && adj != NULL )
    {
        adj->push_back( entity );
    }
    // create a new vector and add it
    else
    {
        adj = new std::vector< EntityHandle >;
        adj->push_back( entity );
        thisMB->tag_set_data( mAdjTag, iter, 1, &adj );
    }
}
 
ErrorCode Skinner::find_geometric_skin( const EntityHandle meshset, Range& forward_target_entities )
{
    // attempts to find whole model skin, using geom topo sets first then
    // normal find_skin function
    bool debug = true;
 
    // look for geom topo sets
    Tag geom_tag;
    ErrorCode result =
        thisMB->tag_get_handle( GEOM_DIMENSION_TAG_NAME, 1, MB_TYPE_INTEGER, geom_tag, MB_TAG_SPARSE | MB_TAG_CREAT );
 
    if( MB_SUCCESS != result ) return result;
 
    // get face sets (dimension = 2)
    Range face_sets;
    int two             = 2;
    const void* two_ptr = &two;
    result = thisMB->get_entities_by_type_and_tag( meshset, MBENTITYSET, &geom_tag, &two_ptr, 1, face_sets );
 
    Range::iterator it;
    if( MB_SUCCESS != result )
        return result;
    else if( face_sets.empty() )
        return MB_ENTITY_NOT_FOUND;
 
    // ok, we have face sets; use those to determine skin
    Range skin_sets;
    if( debug ) std::cout << "Found " << face_sets.size() << " face sets total..." << std::endl;
 
    for( it = face_sets.begin(); it != face_sets.end(); ++it )
    {
        int num_parents;
        result = thisMB->num_parent_meshsets( *it, &num_parents );
        if( MB_SUCCESS != result )
            return result;
        else if( num_parents == 1 )
            skin_sets.insert( *it );
    }
 
    if( debug ) std::cout << "Found " << skin_sets.size() << " 1-parent face sets..." << std::endl;
 
    if( skin_sets.empty() ) return MB_FAILURE;
 
    // ok, we have the shell; gather up the elements, putting them all in forward for now
    for( it = skin_sets.begin(); it != skin_sets.end(); ++it )
    {
        result = thisMB->get_entities_by_handle( *it, forward_target_entities, true );
        if( MB_SUCCESS != result ) return result;
    }
 
    return result;
}
 
ErrorCode Skinner::find_skin( const EntityHandle meshset,
                              const Range& source_entities,
                              bool get_vertices,
                              Range& output_handles,
                              Range* output_reverse_handles,
                              bool create_vert_elem_adjs,
                              bool create_skin_elements,
                              bool look_for_scd )
{
    if( source_entities.empty() ) return MB_SUCCESS;
 
    if( look_for_scd )
    {
        ErrorCode rval = find_skin_scd( source_entities, get_vertices, output_handles, create_skin_elements );
        // if it returns success, it's all scd, and we don't need to do anything more
        if( MB_SUCCESS == rval ) return rval;
    }
 
    Core* this_core = dynamic_cast< Core* >( thisMB );
    if( this_core && create_vert_elem_adjs && !this_core->a_entity_factory()->vert_elem_adjacencies() )
        this_core->a_entity_factory()->create_vert_elem_adjacencies();
 
    return find_skin_vertices( meshset, source_entities, get_vertices ? &output_handles : 0,
                               get_vertices ? 0 : &output_handles, output_reverse_handles, create_skin_elements );
}
 
ErrorCode Skinner::find_skin_scd( const Range& source_entities,
                                  bool get_vertices,
                                  Range& output_handles,
                                  bool create_skin_elements )
{
    // get the scd interface and check if it's been initialized
    ScdInterface* scdi = NULL;
    ErrorCode rval     = thisMB->query_interface( scdi );
    if( !scdi ) return MB_FAILURE;
 
    // ok, there's scd mesh; see if the entities passed in are all in a scd box
    // a box needs to be wholly included in entities for this to work
    std::vector< ScdBox* > boxes, myboxes;
    Range myrange;
    rval = scdi->find_boxes( boxes );
    if( MB_SUCCESS != rval ) return rval;
    for( std::vector< ScdBox* >::iterator bit = boxes.begin(); bit != boxes.end(); ++bit )
    {
        Range belems( ( *bit )->start_element(), ( *bit )->start_element() + ( *bit )->num_elements() - 1 );
        if( source_entities.contains( belems ) )
        {
            myboxes.push_back( *bit );
            myrange.merge( belems );
        }
    }
    if( myboxes.empty() || myrange.size() != source_entities.size() ) return MB_FAILURE;
 
    // ok, we're all structured; get the skin for each box
    for( std::vector< ScdBox* >::iterator bit = boxes.begin(); bit != boxes.end(); ++bit )
    {
        rval = skin_box( *bit, get_vertices, output_handles, create_skin_elements );
        if( MB_SUCCESS != rval ) return rval;
    }
 
    return MB_SUCCESS;
}
 
ErrorCode Skinner::skin_box( ScdBox* box, bool get_vertices, Range& output_handles, bool create_skin_elements )
{
    HomCoord bmin = box->box_min(), bmax = box->box_max();
 
    // don't support 1d boxes
    if( bmin.j() == bmax.j() && bmin.k() == bmax.k() ) return MB_FAILURE;
 
    int dim = ( bmin.k() == bmax.k() ? 1 : 2 );
 
    ErrorCode rval;
    EntityHandle ent;
 
    // i=min
    for( int k = bmin.k(); k < bmax.k(); k++ )
    {
        for( int j = bmin.j(); j < bmax.j(); j++ )
        {
            ent  = 0;
            rval = box->get_adj_edge_or_face( dim, bmin.i(), j, k, 0, ent, create_skin_elements );
            if( MB_SUCCESS != rval ) return rval;
            if( ent ) output_handles.insert( ent );
        }
    }
    // i=max
    for( int k = bmin.k(); k < bmax.k(); k++ )
    {
        for( int j = bmin.j(); j < bmax.j(); j++ )
        {
            ent  = 0;
            rval = box->get_adj_edge_or_face( dim, bmax.i(), j, k, 0, ent, create_skin_elements );
            if( MB_SUCCESS != rval ) return rval;
            if( ent ) output_handles.insert( ent );
        }
    }
    // j=min
    for( int k = bmin.k(); k < bmax.k(); k++ )
    {
        for( int i = bmin.i(); i < bmax.i(); i++ )
        {
            ent  = 0;
            rval = box->get_adj_edge_or_face( dim, i, bmin.j(), k, 1, ent, create_skin_elements );
            if( MB_SUCCESS != rval ) return rval;
            if( ent ) output_handles.insert( ent );
        }
    }
    // j=max
    for( int k = bmin.k(); k < bmax.k(); k++ )
    {
        for( int i = bmin.i(); i < bmax.i(); i++ )
        {
            ent  = 0;
            rval = box->get_adj_edge_or_face( dim, i, bmax.j(), k, 1, ent, create_skin_elements );
            if( MB_SUCCESS != rval ) return rval;
            if( ent ) output_handles.insert( ent );
        }
    }
    // k=min
    for( int j = bmin.j(); j < bmax.j(); j++ )
    {
        for( int i = bmin.i(); i < bmax.i(); i++ )
        {
            ent  = 0;
            rval = box->get_adj_edge_or_face( dim, i, j, bmin.k(), 2, ent, create_skin_elements );
            if( MB_SUCCESS != rval ) return rval;
            if( ent ) output_handles.insert( ent );
        }
    }
    // k=max
    for( int j = bmin.j(); j < bmax.j(); j++ )
    {
        for( int i = bmin.i(); i < bmax.i(); i++ )
        {
            ent  = 0;
            rval = box->get_adj_edge_or_face( dim, i, j, bmax.k(), 2, ent, create_skin_elements );
            if( MB_SUCCESS != rval ) return rval;
            if( ent ) output_handles.insert( ent );
        }
    }
 
    if( get_vertices )
    {
        Range verts;
        rval = thisMB->get_adjacencies( output_handles, 0, true, verts, Interface::UNION );
        if( MB_SUCCESS != rval ) return rval;
        output_handles.merge( verts );
    }
 
    return MB_SUCCESS;
}
 
ErrorCode Skinner::find_skin_noadj(const Range &source_entities,
                                 Range &forward_target_entities,
                                 Range &reverse_target_entities/*,
                                 bool create_vert_elem_adjs*/)
{
    if( source_entities.empty() ) return MB_FAILURE;
 
    // get our working dimensions
    EntityType type      = thisMB->type_from_handle( *( source_entities.begin() ) );
    const int source_dim = CN::Dimension( type );
    mTargetDim           = source_dim - 1;
 
    // make sure we can handle the working dimensions
    if( mTargetDim < 0 || source_dim > 3 ) return MB_FAILURE;
 
    initialize();
 
    Range::const_iterator iter, end_iter;
    end_iter = source_entities.end();
    const EntityHandle* conn;
    EntityHandle match;
 
    direction direct;
    ErrorCode result;
    // assume we'll never have more than 32 vertices on a facet (checked
    // with assert later)
    EntityHandle sub_conn[32];
    std::vector< EntityHandle > tmp_conn_vec;
    int num_nodes, num_sub_nodes, num_sides;
    int sub_indices[32];  // Also, assume that no polygon has more than 32 nodes
    // we could increase that, but we will not display it right in visit moab h5m , anyway
    EntityType sub_type;
 
    // for each source entity
    for( iter = source_entities.begin(); iter != end_iter; ++iter )
    {
        // get the connectivity of this entity
        int actual_num_nodes_polygon = 0;
        result                       = thisMB->get_connectivity( *iter, conn, num_nodes, false, &tmp_conn_vec );
        if( MB_SUCCESS != result ) return result;
 
        type = thisMB->type_from_handle( *iter );
        Range::iterator seek_iter;
 
        // treat separately polygons (also, polyhedra will need special handling)
        if( MBPOLYGON == type )
        {
            // treat padded polygons, if existing; count backwards, see how many of the last nodes
            // are repeated assume connectivity is fine, otherwise we could be in trouble
            actual_num_nodes_polygon = num_nodes;
            while( actual_num_nodes_polygon >= 3 &&
                   conn[actual_num_nodes_polygon - 1] == conn[actual_num_nodes_polygon - 2] )
                actual_num_nodes_polygon--;
            num_sides     = actual_num_nodes_polygon;
            sub_type      = MBEDGE;
            num_sub_nodes = 2;
        }
        else  // get connectivity of each n-1 dimension entity
            num_sides = CN::NumSubEntities( type, mTargetDim );
        for( int i = 0; i < num_sides; i++ )
        {
            if( MBPOLYGON == type )
            {
                sub_conn[0] = conn[i];
                sub_conn[1] = conn[i + 1];
                if( i + 1 == actual_num_nodes_polygon ) sub_conn[1] = conn[0];
            }
            else
            {
                CN::SubEntityNodeIndices( type, num_nodes, mTargetDim, i, sub_type, num_sub_nodes, sub_indices );
                assert( (size_t)num_sub_nodes <= sizeof( sub_indices ) / sizeof( sub_indices[0] ) );
                for( int j = 0; j < num_sub_nodes; j++ )
                    sub_conn[j] = conn[sub_indices[j]];
            }
 
            // see if we can match this connectivity with
            // an existing entity
            find_match( sub_type, sub_conn, num_sub_nodes, match, direct );
 
            // if there is no match, create a new entity
            if( match == 0 )
            {
                EntityHandle tmphndl = 0;
                int indices[MAX_SUB_ENTITY_VERTICES];
                EntityType new_type;
                int num_new_nodes;
                if( MBPOLYGON == type )
                {
                    new_type      = MBEDGE;
                    num_new_nodes = 2;
                }
                else
                {
                    CN::SubEntityNodeIndices( type, num_nodes, mTargetDim, i, new_type, num_new_nodes, indices );
                    for( int j = 0; j < num_new_nodes; j++ )
                        sub_conn[j] = conn[indices[j]];
                }
                result = thisMB->create_element( new_type, sub_conn, num_new_nodes, tmphndl );
                assert( MB_SUCCESS == result );
                add_adjacency( tmphndl, sub_conn, CN::VerticesPerEntity( new_type ) );
                forward_target_entities.insert( tmphndl );
            }
            // if there is a match, delete the matching entity
            // if we can.
            else
            {
                if( ( seek_iter = forward_target_entities.find( match ) ) != forward_target_entities.end() )
                {
                    forward_target_entities.erase( seek_iter );
                    remove_adjacency( match );
                    if( /*!use_adjs &&*/ entity_deletable( match ) )
                    {
                        result = thisMB->delete_entities( &match, 1 );
                        assert( MB_SUCCESS == result );
                    }
                }
                else if( ( seek_iter = reverse_target_entities.find( match ) ) != reverse_target_entities.end() )
                {
                    reverse_target_entities.erase( seek_iter );
                    remove_adjacency( match );
                    if( /*!use_adjs &&*/ entity_deletable( match ) )
                    {
                        result = thisMB->delete_entities( &match, 1 );
                        assert( MB_SUCCESS == result );
                    }
                }
                else
                {
                    if( direct == FORWARD )
                    {
                        forward_target_entities.insert( match );
                    }
                    else
                    {
                        reverse_target_entities.insert( match );
                    }
                }
            }
        }
    }
 
    deinitialize();
 
    return MB_SUCCESS;
}
 
void Skinner::find_match( EntityType type,
                          const EntityHandle* conn,
                          const int num_nodes,
                          EntityHandle& match,
                          Skinner::direction& direct )
{
    match = 0;
 
    if( type == MBVERTEX )
    {
        match  = *conn;
        direct = FORWARD;
        return;
    }
 
    const EntityHandle* iter = std::min_element( conn, conn + num_nodes );
 
    std::vector< EntityHandle >* adj = NULL;
 
    ErrorCode result = thisMB->tag_get_data( mAdjTag, iter, 1, &adj );
    if( result == MB_FAILURE || adj == NULL )
    {
        return;
    }
 
    std::vector< EntityHandle >::iterator jter, end_jter;
    end_jter = adj->end();
 
    const EntityHandle* tmp;
    int num_verts;
 
    for( jter = adj->begin(); jter != end_jter; ++jter )
    {
        EntityType tmp_type;
        tmp_type = thisMB->type_from_handle( *jter );
 
        if( type != tmp_type ) continue;
 
        result = thisMB->get_connectivity( *jter, tmp, num_verts, false );
        assert( MB_SUCCESS == result && num_verts >= CN::VerticesPerEntity( type ) );
        // FIXME: connectivity_match appears to work only for linear elements,
        //        so ignore higher-order nodes.
        if( connectivity_match( conn, tmp, CN::VerticesPerEntity( type ), direct ) )
        {
            match = *jter;
            break;
        }
    }
}
 
bool Skinner::connectivity_match( const EntityHandle* conn1,
                                  const EntityHandle* conn2,
                                  const int num_verts,
                                  Skinner::direction& direct )
{
    const EntityHandle* iter = std::find( conn2, conn2 + num_verts, conn1[0] );
    if( iter == conn2 + num_verts ) return false;
 
    bool they_match = true;
 
    int i;
    unsigned int j = iter - conn2;
 
    // first compare forward
    for( i = 1; i < num_verts; ++i )
    {
        if( conn1[i] != conn2[( j + i ) % num_verts] )
        {
            they_match = false;
            break;
        }
    }
 
    if( they_match == true )
    {
        // need to check for reversed edges here
        direct = ( num_verts == 2 && j ) ? REVERSE : FORWARD;
        return true;
    }
 
    they_match = true;
 
    // then compare reverse
    j += num_verts;
    for( i = 1; i < num_verts; )
    {
        if( conn1[i] != conn2[( j - i ) % num_verts] )
        {
            they_match = false;
            break;
        }
        ++i;
    }
    if( they_match )
    {
        direct = REVERSE;
    }
    return they_match;
}
 
ErrorCode Skinner::remove_adjacency( EntityHandle entity )
{
    std::vector< EntityHandle > nodes, *adj = NULL;
    ErrorCode result = thisMB->get_connectivity( &entity, 1, nodes );MB_CHK_ERR( result );
    std::vector< EntityHandle >::iterator iter = std::min_element( nodes.begin(), nodes.end() );
 
    if( iter == nodes.end() ) return MB_FAILURE;
 
    // remove this entity from the node
    if( thisMB->tag_get_data( mAdjTag, &( *iter ), 1, &adj ) == MB_SUCCESS && adj != NULL )
    {
        iter = std::find( adj->begin(), adj->end(), entity );
        if( iter != adj->end() ) adj->erase( iter );
    }
 
    return result;
}
 
bool Skinner::entity_deletable( EntityHandle entity )
{
    unsigned char deletable = 0;
    ErrorCode result        = thisMB->tag_get_data( mDeletableMBTag, &entity, 1, &deletable );
    assert( MB_SUCCESS == result );
    if( MB_SUCCESS == result && deletable == 1 ) return false;
    return true;
}
 
ErrorCode Skinner::classify_2d_boundary( const Range& boundary,
                                         const Range& bar_elements,
                                         EntityHandle boundary_edges,
                                         EntityHandle inferred_edges,
                                         EntityHandle non_manifold_edges,
                                         EntityHandle other_edges,
                                         int& number_boundary_nodes )
{
    Range bedges, iedges, nmedges, oedges;
    ErrorCode result =
        classify_2d_boundary( boundary, bar_elements, bedges, iedges, nmedges, oedges, number_boundary_nodes );MB_CHK_ERR( result );
 
    // now set the input meshsets to the output ranges
    result = thisMB->clear_meshset( &boundary_edges, 1 );MB_CHK_ERR( result );
    result = thisMB->add_entities( boundary_edges, bedges );MB_CHK_ERR( result );
 
    result = thisMB->clear_meshset( &inferred_edges, 1 );MB_CHK_ERR( result );
    result = thisMB->add_entities( inferred_edges, iedges );MB_CHK_ERR( result );
 
    result = thisMB->clear_meshset( &non_manifold_edges, 1 );MB_CHK_ERR( result );
    result = thisMB->add_entities( non_manifold_edges, nmedges );MB_CHK_ERR( result );
 
    result = thisMB->clear_meshset( &other_edges, 1 );MB_CHK_ERR( result );
    result = thisMB->add_entities( other_edges, oedges );MB_CHK_ERR( result );
 
    return MB_SUCCESS;
}
 
ErrorCode Skinner::classify_2d_boundary( const Range& boundary,
                                         const Range& bar_elements,
                                         Range& boundary_edges,
                                         Range& inferred_edges,
                                         Range& non_manifold_edges,
                                         Range& other_edges,
                                         int& number_boundary_nodes )
{
 
    // clear out the edge lists
 
    boundary_edges.clear();
    inferred_edges.clear();
    non_manifold_edges.clear();
    other_edges.clear();
 
    number_boundary_nodes = 0;
 
    // make sure we have something to work with
    if( boundary.empty() )
    {
        return MB_FAILURE;
    }
 
    // get our working dimensions
    EntityType type      = thisMB->type_from_handle( *( boundary.begin() ) );
    const int source_dim = CN::Dimension( type );
 
    // make sure we can handle the working dimensions
    if( source_dim != 2 )
    {
        return MB_FAILURE;
    }
    mTargetDim = source_dim - 1;
 
    // initialize
    initialize();
 
    // additional initialization for this routine
    // define a tag for MBEDGE which counts the occurrences of the edge below
    // default should be 0 for existing edges, if any
 
    Tag count_tag;
    int default_count = 0;
    ErrorCode result =
        thisMB->tag_get_handle( 0, 1, MB_TYPE_INTEGER, count_tag, MB_TAG_DENSE | MB_TAG_CREAT, &default_count );MB_CHK_ERR( result );
 
    Range::const_iterator iter, end_iter;
    end_iter = boundary.end();
 
    std::vector< EntityHandle > conn;
    EntityHandle sub_conn[2];
    EntityHandle match;
 
    Range edge_list;
    Range boundary_nodes;
    Skinner::direction direct;
 
    EntityType sub_type;
    int num_edge, num_sub_ent_vert;
    const short* edge_verts;
 
    // now, process each entity in the boundary
 
    for( iter = boundary.begin(); iter != end_iter; ++iter )
    {
        // get the connectivity of this entity
        conn.clear();
        result = thisMB->get_connectivity( &( *iter ), 1, conn, false );
        assert( MB_SUCCESS == result );
 
        // add node handles to boundary_node range
        std::copy( conn.begin(), conn.begin() + CN::VerticesPerEntity( type ), range_inserter( boundary_nodes ) );
 
        type = thisMB->type_from_handle( *iter );
 
        // get connectivity of each n-1 dimension entity (edge in this case)
        const struct CN::ConnMap* conn_map = &( CN::mConnectivityMap[type][0] );
        num_edge                           = CN::NumSubEntities( type, 1 );
        for( int i = 0; i < num_edge; i++ )
        {
            edge_verts = CN::SubEntityVertexIndices( type, 1, i, sub_type, num_sub_ent_vert );
            assert( sub_type == MBEDGE && num_sub_ent_vert == 2 );
            sub_conn[0]       = conn[edge_verts[0]];
            sub_conn[1]       = conn[edge_verts[1]];
            int num_sub_nodes = conn_map->num_corners_per_sub_element[i];
 
            // see if we can match this connectivity with
            // an existing entity
            find_match( MBEDGE, sub_conn, num_sub_nodes, match, direct );
 
            // if there is no match, create a new entity
            if( match == 0 )
            {
                EntityHandle tmphndl = 0;
                int indices[MAX_SUB_ENTITY_VERTICES];
                EntityType new_type;
                int num_new_nodes;
                CN::SubEntityNodeIndices( type, conn.size(), 1, i, new_type, num_new_nodes, indices );
                for( int j = 0; j < num_new_nodes; j++ )
                    sub_conn[j] = conn[indices[j]];
 
                result = thisMB->create_element( new_type, sub_conn, num_new_nodes, tmphndl );
                assert( MB_SUCCESS == result );
                add_adjacency( tmphndl, sub_conn, num_sub_nodes );
                // target_entities.insert(tmphndl);
                edge_list.insert( tmphndl );
                int count;
                result = thisMB->tag_get_data( count_tag, &tmphndl, 1, &count );
                assert( MB_SUCCESS == result );
                count++;
                result = thisMB->tag_set_data( count_tag, &tmphndl, 1, &count );
                assert( MB_SUCCESS == result );
            }
            else
            {
                // We found a match, we must increment the count on the match
                int count;
                result = thisMB->tag_get_data( count_tag, &match, 1, &count );
                assert( MB_SUCCESS == result );
                count++;
                result = thisMB->tag_set_data( count_tag, &match, 1, &count );
                assert( MB_SUCCESS == result );
 
                // if the entity is not deletable, it was pre-existing in
                // the database.  We therefore may need to add it to the
                // edge_list.  Since it will not hurt the range, we add
                // whether it was added before or not
                if( !entity_deletable( match ) )
                {
                    edge_list.insert( match );
                }
            }
        }
    }
 
    // Any bar elements in the model should be classified separately
    // If the element is in the skin edge_list, then it should be put in
    // the non-manifold edge list.  Edges not in the edge_list are stand-alone
    // bars, and we make them simply boundary elements
 
    if( !bar_elements.empty() )
    {
        Range::iterator bar_iter;
        for( iter = bar_elements.begin(); iter != bar_elements.end(); ++iter )
        {
            EntityHandle handle = *iter;
            bar_iter            = edge_list.find( handle );
            if( bar_iter != edge_list.end() )
            {
                // it is in the list, erase it and put in non-manifold list
                edge_list.erase( bar_iter );
                non_manifold_edges.insert( handle );
            }
            else
            {
                // not in the edge list, make it a boundary edge
                boundary_edges.insert( handle );
            }
        }
    }
 
    // now all edges should be classified.  Go through the edge_list,
    // and put all in the appropriate lists
 
    Range::iterator edge_iter, edge_end_iter;
    edge_end_iter = edge_list.end();
    int count;
    for( edge_iter = edge_list.begin(); edge_iter != edge_end_iter; ++edge_iter )
    {
        // check the count_tag
        result = thisMB->tag_get_data( count_tag, &( *edge_iter ), 1, &count );
        assert( MB_SUCCESS == result );
        if( count == 1 )
        {
            boundary_edges.insert( *edge_iter );
        }
        else if( count == 2 )
        {
            other_edges.insert( *edge_iter );
        }
        else
        {
            non_manifold_edges.insert( *edge_iter );
        }
    }
 
    // find the inferred edges from the other_edge_list
 
    double min_angle_degrees = 20.0;
    find_inferred_edges( const_cast< Range& >( boundary ), other_edges, inferred_edges, min_angle_degrees );
 
    // we now want to remove the inferred_edges from the other_edges
 
    Range temp_range;
 
    std::set_difference( other_edges.begin(), other_edges.end(), inferred_edges.begin(), inferred_edges.end(),
                         range_inserter( temp_range ), std::less< EntityHandle >() );
 
    other_edges = temp_range;
 
    // get rid of count tag and deinitialize
 
    result = thisMB->tag_delete( count_tag );
    assert( MB_SUCCESS == result );
    deinitialize();
 
    // set the node count
    number_boundary_nodes = boundary_nodes.size();
 
    return MB_SUCCESS;
}
 
void Skinner::find_inferred_edges( Range& skin_boundary,
                                   Range& candidate_edges,
                                   Range& inferred_edges,
                                   double reference_angle_degrees )
{
 
    // mark all the entities in the skin boundary
    Tag mark_tag;
    ErrorCode result = thisMB->tag_get_handle( 0, 1, MB_TYPE_BIT, mark_tag, MB_TAG_CREAT );
    assert( MB_SUCCESS == result );
    unsigned char bit = true;
    result            = thisMB->tag_clear_data( mark_tag, skin_boundary, &bit );
    assert( MB_SUCCESS == result );
 
    // find the cosine of the reference angle
 
    double reference_cosine = cos( reference_angle_degrees * SKINNER_PI / 180.0 );
 
    // check all candidate edges for an angle greater than the minimum
 
    Range::iterator iter, end_iter = candidate_edges.end();
    std::vector< EntityHandle > adjacencies;
    std::vector< EntityHandle >::iterator adj_iter;
    EntityHandle face[2];
 
    for( iter = candidate_edges.begin(); iter != end_iter; ++iter )
    {
 
        // get the 2D elements connected to this edge
        adjacencies.clear();
        result = thisMB->get_adjacencies( &( *iter ), 1, 2, false, adjacencies );
        if( MB_SUCCESS != result ) continue;
 
        // there should be exactly two, that is why the edge is classified as nonBoundary
        // and manifold
 
        int faces_found = 0;
        for( adj_iter = adjacencies.begin(); adj_iter != adjacencies.end() && faces_found < 2; ++adj_iter )
        {
            // we need to find two of these which are in the skin
            unsigned char is_marked = 0;
            result                  = thisMB->tag_get_data( mark_tag, &( *adj_iter ), 1, &is_marked );
            assert( MB_SUCCESS == result );
            if( is_marked )
            {
                face[faces_found] = *adj_iter;
                faces_found++;
            }
        }
 
        //    assert(faces_found == 2 || faces_found == 0);
        if( 2 != faces_found ) continue;
 
        // see if the two entities have a sufficient angle
 
        if( has_larger_angle( face[0], face[1], reference_cosine ) )
        {
            inferred_edges.insert( *iter );
        }
    }
 
    result = thisMB->tag_delete( mark_tag );
    assert( MB_SUCCESS == result );
}
 
bool Skinner::has_larger_angle( EntityHandle& entity1, EntityHandle& entity2, double reference_angle_cosine )
{
    // compare normals to get angle.  We assume that the surface quads
    // which we test here will be approximately planar
 
    double norm[2][3];
    Util::normal( thisMB, entity1, norm[0][0], norm[0][1], norm[0][2] );
    Util::normal( thisMB, entity2, norm[1][0], norm[1][1], norm[1][2] );
 
    double cosine = norm[0][0] * norm[1][0] + norm[0][1] * norm[1][1] + norm[0][2] * norm[1][2];
 
    if( cosine < reference_angle_cosine )
    {
        return true;
    }
 
    return false;
}
 
// get skin entities of prescribed dimension
ErrorCode Skinner::find_skin( const EntityHandle this_set,
                              const Range& entities,
                              int dim,
                              Range& skin_entities,
                              bool create_vert_elem_adjs,
                              bool create_skin_elements )
{
    Range tmp_skin;
    ErrorCode result =
        find_skin( this_set, entities, ( dim == 0 ), tmp_skin, 0, create_vert_elem_adjs, create_skin_elements );
    if( MB_SUCCESS != result || tmp_skin.empty() ) return result;
 
    if( tmp_skin.all_of_dimension( dim ) )
    {
        if( skin_entities.empty() )
            skin_entities.swap( tmp_skin );
        else
            skin_entities.merge( tmp_skin );
    }
    else
    {
        result = thisMB->get_adjacencies( tmp_skin, dim, create_skin_elements, skin_entities, Interface::UNION );MB_CHK_ERR( result );
        if( this_set ) result = thisMB->add_entities( this_set, skin_entities );
    }
 
    return result;
}
 
ErrorCode Skinner::find_skin_vertices( const EntityHandle this_set,
                                       const Range& entities,
                                       Range* skin_verts,
                                       Range* skin_elems,
                                       Range* skin_rev_elems,
                                       bool create_skin_elems,
                                       bool corners_only )
{
    ErrorCode rval;
    if( entities.empty() ) return MB_SUCCESS;
 
    const int dim = CN::Dimension( TYPE_FROM_HANDLE( entities.front() ) );
    if( dim < 1 || dim > 3 || !entities.all_of_dimension( dim ) ) return MB_TYPE_OUT_OF_RANGE;
 
    // are we skinning all entities
    size_t count = entities.size();
    int num_total;
    rval = thisMB->get_number_entities_by_dimension( this_set, dim, num_total );
    if( MB_SUCCESS != rval ) return rval;
    bool all = ( count == (size_t)num_total );
 
    // Create a bit tag for fast intersection with input entities range.
    // If we're skinning all the entities in the mesh, we really don't
    // need the tag.  To save memory, just create it with a default value
    // of one and don't set it.  That way MOAB will return 1 for all
    // entities.
    Tag tag;
    char bit = all ? 1 : 0;
    rval     = thisMB->tag_get_handle( NULL, 1, MB_TYPE_BIT, tag, MB_TAG_CREAT, &bit );
    if( MB_SUCCESS != rval ) return rval;
 
    // tag all entities in input range
    if( !all )
    {
        std::vector< unsigned char > vect( count, 1 );
        rval = thisMB->tag_set_data( tag, entities, &vect[0] );
        if( MB_SUCCESS != rval )
        {
            thisMB->tag_delete( tag );
            return rval;
        }
    }
 
    switch( dim )
    {
        case 1:
            if( skin_verts )
                rval = find_skin_vertices_1D( tag, entities, *skin_verts );
            else if( skin_elems )
                rval = find_skin_vertices_1D( tag, entities, *skin_elems );
            else
                rval = MB_SUCCESS;
            break;
        case 2:
            rval = find_skin_vertices_2D( this_set, tag, entities, skin_verts, skin_elems, skin_rev_elems,
                                          create_skin_elems, corners_only );
            break;
        case 3:
            rval = find_skin_vertices_3D( this_set, tag, entities, skin_verts, skin_elems, skin_rev_elems,
                                          create_skin_elems, corners_only );
            break;
        default:
            rval = MB_TYPE_OUT_OF_RANGE;
            break;
    }
 
    thisMB->tag_delete( tag );
    return rval;
}
 
ErrorCode Skinner::find_skin_vertices_1D( Tag tag, const Range& edges, Range& skin_verts )
{
    // This rather simple algorithm is provided for completeness
    // (not sure how often one really wants the 'skin' of a chain
    // or tangle of edges.)
    //
    // A vertex is on the skin of the edges if it is contained in exactly
    // one of the edges *in the input range*.
    //
    // This function expects the caller to have tagged all edges in the
    // input range with a value of one for the passed bit tag, and all
    // other edges with a value of zero.  This allows us to do a faster
    // intersection with the input range and the edges adjacent to a vertex.
 
    ErrorCode rval;
    Range::iterator hint = skin_verts.begin();
 
    // All input entities must be edges.
    if( !edges.all_of_dimension( 1 ) ) return MB_TYPE_OUT_OF_RANGE;
 
    // get all the vertices
    Range verts;
    rval = thisMB->get_connectivity( edges, verts, true );
    if( MB_SUCCESS != rval ) return rval;
 
    // Test how many edges each input vertex is adjacent to.
    std::vector< char > tag_vals;
    std::vector< EntityHandle > adj;
    int n;
    for( Range::const_iterator it = verts.begin(); it != verts.end(); ++it )
    {
        // get edges adjacent to vertex
        adj.clear();
        rval = thisMB->get_adjacencies( &*it, 1, 1, false, adj );
        if( MB_SUCCESS != rval ) return rval;
        if( adj.empty() ) continue;
 
        // Intersect adjacent edges with the input list of edges
        tag_vals.resize( adj.size() );
        rval = thisMB->tag_get_data( tag, &adj[0], adj.size(), &tag_vals[0] );
        if( MB_SUCCESS != rval ) return rval;
#ifdef MOAB_OLD_STD_COUNT
        n = 0;
        std::count( tag_vals.begin(), tag_vals.end(), '\001' );
#else
        n = std::count( tag_vals.begin(), tag_vals.end(), '\001' );
#endif
        // If adjacent to only one input edge, then vertex is on skin
        if( n == 1 )
        {
            hint = skin_verts.insert( hint, *it );
        }
    }
 
    return MB_SUCCESS;
}
 
// A Container for storing a list of element sides adjacent
// to a vertex.  The template parameter is the number of
// corners for the side.
template < unsigned CORNERS >
class AdjSides
{
  public:
    /**
     * This struct is used to for a reduced representation of element
     * "sides" adjacent to a give vertex.  As such, it
     * a) does not store the initial vertex that all sides are adjacent to
     * b) orders the remaining vertices in a specific way for fast comparison.
     *
     * For edge elements, only the opposite vertex is stored.
     * For triangle elements, only the other two vertices are stored,
     *   and they are stored with the smaller of those two handles first.
     * For quad elements, only the other three vertices are stored.
     *  They are stored such that the vertex opposite the implicit (not
     *  stored) vertex is always in slot 1.  The other two vertices
     *  (in slots 0 and 2) are ordered such that the handle of the one in
     *  slot 0 is smaller than the handle in slot 2.
     *
     * For each side, the adj_elem field is used to store the element that
     * it is a side of as long as the element is considered to be on the skin.
     * The adj_elem field is cleared (set to zero) to indicate that this
     * side is no longer considered to be on the skin (and is the side of
     * more than one element.)
     */
    struct Side
    {
        EntityHandle handles[CORNERS - 1];  //!< side vertices, except for implicit one
        EntityHandle adj_elem;              //!< element that this is a side of, or zero
        bool skin() const
        {
            return 0 != adj_elem;
        }
 
        /** construct from connectivity of side
         *\param array The connectivity of the element side.
         *\param idx   The index of the implicit vertex (contained
         *             in all sides in the list.)
         *\param adj   The element that this is a side of.
         */
        Side( const EntityHandle* array, int idx, EntityHandle adj, unsigned short ) : adj_elem( adj )
        {
            switch( CORNERS )
            {
                case 3:
                    handles[1] = array[( idx + 2 ) % CORNERS];
                    // fall through
                case 2:
                    if( 3 == CORNERS ) handles[1] = array[( idx + 2 ) % CORNERS];
                    if( 2 <= CORNERS ) handles[0] = array[( idx + 1 ) % CORNERS];
                    break;
                default:
                    assert( false );
                    break;
            }
            if( CORNERS == 3 && handles[1] > handles[0] ) std::swap( handles[0], handles[1] );
        }
 
        /** construct from connectivity of parent element
         *\param array The connectivity of the parent element
         *\param idx   The index of the implicit vertex (contained
         *             in all sides in the list.)  This is an index
         *             into 'indices', not 'array'.
         *\param adj   The element that this is a side of.
         *\param indices  The indices into 'array' at which the vertices
         *             representing the side occur.
         */
        Side( const EntityHandle* array, int idx, EntityHandle adj, unsigned short, const short* indices )
            : adj_elem( adj )
        {
            switch( CORNERS )
            {
                case 3:
                    handles[1] = array[indices[( idx + 2 ) % CORNERS]];
                    // fall through
                case 2:
                    if( 3 == CORNERS ) handles[1] = array[indices[( idx + 2 ) % CORNERS]];
                    if( 2 <= CORNERS ) handles[0] = array[indices[( idx + 1 ) % CORNERS]];
                    break;
                default:
                    assert( false );
                    break;
            }
            if( CORNERS == 3 && handles[1] > handles[0] ) std::swap( handles[0], handles[1] );
        }
 
        // Compare two side instances.  Relies in the ordering of
        // vertex handles as described above.
        bool operator==( const Side& other ) const
        {
            switch( CORNERS )
            {
                case 3:
                    return handles[0] == other.handles[0] && handles[1] == other.handles[1];
                case 2:
                    return handles[0] == other.handles[0];
                default:
                    assert( false );
                    return false;
            }
        }
    };
 
  private:
    std::vector< Side > data;  //!< List of sides
    size_t skin_count;         //!< Cached count of sides that are skin
 
  public:
    typedef typename std::vector< Side >::iterator iterator;
    typedef typename std::vector< Side >::const_iterator const_iterator;
    const_iterator begin() const
    {
        return data.begin();
    }
    const_iterator end() const
    {
        return data.end();
    }
 
    void clear()
    {
        data.clear();
        skin_count = 0;
    }
    bool empty() const
    {
        return data.empty();
    }
 
    AdjSides() : skin_count( 0 ) {}
 
    size_t num_skin() const
    {
        return skin_count;
    }
 
    /** \brief insert side, specifying side connectivity
     *
     * Either insert a new side, or if the side is already in the
     * list, mark it as not on the skin.
     *
     *\param handles The connectivity of the element side.
     *\param skip_idx The index of the implicit vertex (contained
     *             in all sides in the list.)
     *\param adj_elem The element that this is a side of.
     *\param elem_side Which side of adj_elem are we storing
     *             (CN side number.)
     */
    void insert( const EntityHandle* handles, int skip_idx, EntityHandle adj_elem, unsigned short elem_side )
    {
        Side side( handles, skip_idx, adj_elem, elem_side );
        iterator p = std::find( data.begin(), data.end(), side );
        if( p == data.end() )
        {
            data.push_back( side );
            ++skin_count;  // not in list yet, so skin side (so far)
        }
        else if( p->adj_elem )
        {
            p->adj_elem = 0;  // mark as not on skin
            --skin_count;     // decrement cached count of skin elements
        }
    }
 
    /** \brief insert side, specifying list of indices into parent element
     * connectivity.
     *
     * Either insert a new side, or if the side is already in the
     * list, mark it as not on the skin.
     *
     *\param handles The connectivity of the parent element
     *\param skip_idx The index of the implicit vertex (contained
     *             in all sides in the list.)  This is an index
     *             into 'indices', not 'handles'.
     *\param adj_elem The element that this is a side of (parent handle).
     *\param indices  The indices into 'handles' at which the vertices
     *             representing the side occur.
     *\param elem_side Which side of adj_elem are we storing
     *             (CN side number.)
     */
    void insert( const EntityHandle* handles,
                 int skip_idx,
                 EntityHandle adj_elem,
                 unsigned short elem_side,
                 const short* indices )
    {
        Side side( handles, skip_idx, adj_elem, elem_side, indices );
        iterator p = std::find( data.begin(), data.end(), side );
        if( p == data.end() )
        {
            data.push_back( side );
            ++skin_count;  // not in list yet, so skin side (so far)
        }
        else if( p->adj_elem )
        {
            p->adj_elem = 0;  // mark as not on skin
            --skin_count;     // decrement cached count of skin elements
        }
    }
 
    /**\brief Search list for a given side, and if found, mark as not skin.
     *
     *\param other   Connectivity of side
     *\param skip_index Index in 'other' at which implicit vertex occurs.
     *\param elem_out If return value is true, the element that the side is a
     *                side of.  If return value is false, not set.
     *\return true if found and marked not-skin, false if not found.
     *
     * Given the connectivity of some existing element, check if it occurs
     * in the list.  If it does, clear the "is skin" state of the side so
     * that we know that we don't need to later create the side element.
     */
    bool find_and_unmark( const EntityHandle* other, int skip_index, EntityHandle& elem_out )
    {
        Side s( other, skip_index, 0, 0 );
        iterator p = std::find( data.begin(), data.end(), s );
        if( p == data.end() || !p->adj_elem )
            return false;
        else
        {
            elem_out    = p->adj_elem;
            p->adj_elem = 0;  // clear "is skin" state for side
            --skin_count;     // decrement cached count of skin sides
            return true;
        }
    }
};
 
/** construct from connectivity of side
 *\param array The connectivity of the element side.
 *\param idx   The index of the implicit vertex (contained
 *             in all sides in the list.)
 *\param adj   The element that this is a side of.
 */
template <>
AdjSides< 4 >::Side::Side( const EntityHandle* array, int idx, EntityHandle adj, unsigned short ) : adj_elem( adj )
{
    const unsigned int CORNERS = 4;
    handles[2]                 = array[( idx + 3 ) % CORNERS];
    handles[1]                 = array[( idx + 2 ) % CORNERS];
    handles[0]                 = array[( idx + 1 ) % CORNERS];
    if( handles[2] > handles[0] ) std::swap( handles[0], handles[2] );
}
 
/** construct from connectivity of parent element
 *\param array The connectivity of the parent element
 *\param idx   The index of the implicit vertex (contained
 *             in all sides in the list.)  This is an index
 *             into 'indices', not 'array'.
 *\param adj   The element that this is a side of.
 *\param indices  The indices into 'array' at which the vertices
 *             representing the side occur.
 */
template <>
AdjSides< 4 >::Side::Side( const EntityHandle* array, int idx, EntityHandle adj, unsigned short, const short* indices )
    : adj_elem( adj )
{
    const unsigned int CORNERS = 4;
    handles[2]                 = array[indices[( idx + 3 ) % CORNERS]];
    handles[1]                 = array[indices[( idx + 2 ) % CORNERS]];
    handles[0]                 = array[indices[( idx + 1 ) % CORNERS]];
    if( handles[2] > handles[0] ) std::swap( handles[0], handles[2] );
}
 
// Compare two side instances.  Relies in the ordering of
// vertex handles as described above.
template <>
bool AdjSides< 4 >::Side::operator==( const Side& other ) const
{
    return handles[0] == other.handles[0] && handles[1] == other.handles[1] && handles[2] == other.handles[2];
}
 
// Utility function used by find_skin_vertices_2D and
// find_skin_vertices_3D to create elements representing
// the skin side of a higher-dimension element if one
// does not already exist.
//
// Some arguments may seem redundant, but they are used
// to create the correct order of element when the input
// element contains higher-order nodes.
//
// This function always creates elements that have a "forward"
// orientation with respect to the parent element (have
// nodes ordered the same as CN returns for the "side").
//
// elem - The higher-dimension element for which to create
//        a lower-dim element representing the side.
// side_type - The EntityType of the desired side.
// side_conn - The connectivity of the new side.
ErrorCode Skinner::create_side( const EntityHandle this_set,
                                EntityHandle elem,
                                EntityType side_type,
                                const EntityHandle* side_conn,
                                EntityHandle& side_elem )
{
    const int max_side = 9;
    const EntityHandle* conn;
    int len, side_len, side, sense, offset, indices[max_side];
    ErrorCode rval;
    EntityType type   = TYPE_FROM_HANDLE( elem ), tmp_type;
    const int ncorner = CN::VerticesPerEntity( side_type );
    const int d       = CN::Dimension( side_type );
    std::vector< EntityHandle > storage;
 
    // Get the connectivity of the parent element
    rval = thisMB->get_connectivity( elem, conn, len, false, &storage );
    if( MB_SUCCESS != rval ) return rval;
 
    // treat separately MBPOLYGON; we want to create the edge in the
    // forward sense always ; so figure out the sense first, then get out
    if( MBPOLYGON == type && 1 == d && MBEDGE == side_type )
    {
        // first find the first vertex in the conn list
        int i = 0;
        for( i = 0; i < len; i++ )
        {
            if( conn[i] == side_conn[0] ) break;
        }
        if( len == i ) return MB_FAILURE;  // not found, big error
        // now, what if the polygon is padded?
        // the previous index is fine always. but the next one could be trouble :(
        int prevIndex = ( i + len - 1 ) % len;
        int nextIndex = ( i + 1 ) % len;
        // if the next index actually point to the same node, as current, it means it is padded
        if( conn[nextIndex] == conn[i] )
        {
            // it really means we are at the end of proper nodes, the last nodes are repeated, so it
            // should be the first node
            nextIndex = 0;  // this is the first node!
        }
        EntityHandle conn2[2] = { side_conn[0], side_conn[1] };
        if( conn[prevIndex] == side_conn[1] )
        {
            // reverse, so the edge will be forward
            conn2[0] = side_conn[1];
            conn2[1] = side_conn[0];
        }
        else if( conn[nextIndex] != side_conn[1] )
            return MB_FAILURE;  // it is not adjacent to the polygon
 
        rval = thisMB->create_element( MBEDGE, conn2, 2, side_elem );MB_CHK_ERR( rval );
        if( this_set )
        {
            rval = thisMB->add_entities( this_set, &side_elem, 1 );MB_CHK_ERR( rval );
        }
        return MB_SUCCESS;
    }
    // Find which side we are creating and get indices of all nodes
    // (including higher-order, if any.)
    CN::SideNumber( type, conn, side_conn, ncorner, d, side, sense, offset );
    CN::SubEntityNodeIndices( type, len, d, side, tmp_type, side_len, indices );
    assert( side_len <= max_side );
    assert( side_type == tmp_type );
 
    // NOTE: re-create conn array even when no higher-order nodes
    //      because we want it to always be forward with respect
    //      to the side ordering.
    EntityHandle side_conn_full[max_side];
    for( int i = 0; i < side_len; ++i )
        side_conn_full[i] = conn[indices[i]];
 
    rval = thisMB->create_element( side_type, side_conn_full, side_len, side_elem );MB_CHK_ERR( rval );
    if( this_set )
    {
        rval = thisMB->add_entities( this_set, &side_elem, 1 );MB_CHK_ERR( rval );
    }
    return MB_SUCCESS;
    ;
}
 
// Test if an edge is reversed with respect CN's ordering
// for the "side" of a face.
bool Skinner::edge_reversed( EntityHandle face, const EntityHandle* edge_ends )
{
    const EntityHandle* conn;
    int len, idx;
    ErrorCode rval = thisMB->get_connectivity( face, conn, len, true );
    if( MB_SUCCESS != rval )
    {
        assert( false );
        return false;
    }
    idx = std::find( conn, conn + len, edge_ends[0] ) - conn;
    if( idx == len )
    {
        assert( false );
        return false;
    }
    return ( edge_ends[1] == conn[( idx + len - 1 ) % len] );
}
 
// Test if a 2D element representing the side or face of a
// volume element is reversed with respect to the CN node
// ordering for the corresponding region element side.
bool Skinner::face_reversed( EntityHandle region, const EntityHandle* face_corners, EntityType face_type )
{
    const EntityHandle* conn;
    int len, side, sense, offset;
    ErrorCode rval = thisMB->get_connectivity( region, conn, len, true );
    if( MB_SUCCESS != rval )
    {
        assert( false );
        return false;
    }
    short r = CN::SideNumber( TYPE_FROM_HANDLE( region ), conn, face_corners, CN::VerticesPerEntity( face_type ),
                              CN::Dimension( face_type ), side, sense, offset );
    assert( 0 == r );
    return ( !r && sense == -1 );
}
 
ErrorCode Skinner::find_skin_vertices_2D( const EntityHandle this_set,
                                          Tag tag,
                                          const Range& faces,
                                          Range* skin_verts,
                                          Range* skin_edges,
                                          Range* reversed_edges,
                                          bool create_edges,
                                          bool corners_only )
{
    // This function iterates over all the vertices contained in the
    // input face list.  For each such vertex, it then iterates over
    // all of the sides of the face elements adjacent to the vertex.
    // If an adjacent side is the side of only one of the input
    // faces, then that side is on the skin.
    //
    // This algorithm will visit each skin vertex exactly once.  It
    // will visit each skin side once for each vertex in the side.
    //
    // This function expects the caller to have created the passed bit
    // tag and set it to one only for the faces in the passed range.  This
    // tag is used to do a fast intersection of the faces adjacent to a
    // vertex with the faces in the input range (discard any for which the
    // tag is not set to one.)
 
    ErrorCode rval;
    std::vector< EntityHandle >::iterator i, j;
    Range::iterator hint;
    if( skin_verts ) hint = skin_verts->begin();
    std::vector< EntityHandle > storage;
    const EntityHandle* conn;
    int len;
    bool find_edges                      = skin_edges || create_edges;
    bool printed_nonconformal_ho_warning = false;
    EntityHandle face;
 
    if( !faces.all_of_dimension( 2 ) ) return MB_TYPE_OUT_OF_RANGE;
 
    // get all the vertices
    Range verts;
    rval = thisMB->get_connectivity( faces, verts, true );
    if( MB_SUCCESS != rval ) return rval;
 
    std::vector< char > tag_vals;
    std::vector< EntityHandle > adj;
    AdjSides< 2 > adj_edges;
    for( Range::const_iterator it = verts.begin(); it != verts.end(); ++it )
    {
        bool higher_order = false;
 
        // get all adjacent faces
        adj.clear();
        rval = thisMB->get_adjacencies( &*it, 1, 2, false, adj );
        if( MB_SUCCESS != rval ) return rval;
        if( adj.empty() ) continue;
 
        // remove those not in the input list (intersect with input list)
        i = j = adj.begin();
        tag_vals.resize( adj.size() );
        rval = thisMB->tag_get_data( tag, &adj[0], adj.size(), &tag_vals[0] );
        if( MB_SUCCESS != rval ) return rval;
        // remove non-tagged entries
        i = j = adj.begin();
        for( ; i != adj.end(); ++i )
            if( tag_vals[i - adj.begin()] ) *( j++ ) = *i;
        adj.erase( j, adj.end() );
 
        // For each adjacent face, check the edges adjacent to the current vertex
        adj_edges.clear();  // other vertex for adjacent edges
        for( i = adj.begin(); i != adj.end(); ++i )
        {
            rval = thisMB->get_connectivity( *i, conn, len, false, &storage );
            if( MB_SUCCESS != rval ) return rval;
 
            // For a single face element adjacent to this vertex, there
            // will be exactly two sides (edges) adjacent to the vertex.
            // Find the other vertex for each of the two edges.
 
            EntityHandle prev, next;  // vertices of two adjacent edge-sides
            const int idx = std::find( conn, conn + len, *it ) - conn;
            assert( idx != len );
 
            if( TYPE_FROM_HANDLE( *i ) == MBTRI && len > 3 )
            {
                len          = 3;
                higher_order = true;
                if( idx > 2 )
                {  // skip higher-order nodes for now
                    if( !printed_nonconformal_ho_warning )
                    {
                        printed_nonconformal_ho_warning = true;
                        std::cerr << "Non-conformal higher-order mesh detected in skinner: "
                                  << "vertex " << ID_FROM_HANDLE( *it ) << " is a corner in "
                                  << "some elements and a higher-order node in others" << std::endl;
                    }
                    continue;
                }
            }
            else if( TYPE_FROM_HANDLE( *i ) == MBQUAD && len > 4 )
            {
                len          = 4;
                higher_order = true;
                if( idx > 3 )
                {  // skip higher-order nodes for now
                    if( !printed_nonconformal_ho_warning )
                    {
                        printed_nonconformal_ho_warning = true;
                        std::cerr << "Non-conformal higher-order mesh detected in skinner: "
                                  << "vertex " << ID_FROM_HANDLE( *it ) << " is a corner in "
                                  << "some elements and a higher-order node in others" << std::endl;
                    }
                    continue;
                }
            }
 
            // so it must be a MBPOLYGON
            const int prev_idx = ( idx + len - 1 ) % len;  // this should be fine, always, even for padded case
            prev               = conn[prev_idx];
            next               = conn[( idx + 1 ) % len];
            if( next == conn[idx] )  // it must be the padded case, so roll to the beginning
                next = conn[0];
 
            // Insert sides (edges) in our list of candidate skin sides
            adj_edges.insert( &prev, 1, *i, prev_idx );
            adj_edges.insert( &next, 1, *i, idx );
        }
 
        // If vertex is not on skin, advance to next vertex.
        // adj_edges handled checking for duplicates on insertion.
        // If every candidate skin edge occurred more than once (was
        // not in fact on the skin), then we're done with this vertex.
        if( 0 == adj_edges.num_skin() ) continue;
 
        // If user requested Range of *vertices* on the skin...
        if( skin_verts )
        {
            // Put skin vertex in output list
            hint = skin_verts->insert( hint, *it );
 
            // Add mid edge nodes to vertex list
            if( !corners_only && higher_order )
            {
                for( AdjSides< 2 >::const_iterator p = adj_edges.begin(); p != adj_edges.end(); ++p )
                {
                    if( p->skin() )
                    {
                        face            = p->adj_elem;
                        EntityType type = TYPE_FROM_HANDLE( face );
 
                        rval = thisMB->get_connectivity( face, conn, len, false );
                        if( MB_SUCCESS != rval ) return rval;
                        if( !CN::HasMidEdgeNodes( type, len ) ) continue;
 
                        EntityHandle ec[2] = { *it, p->handles[0] };
                        int side, sense, offset;
                        CN::SideNumber( type, conn, ec, 2, 1, side, sense, offset );
                        offset = CN::HONodeIndex( type, len, 1, side );
                        assert( offset >= 0 && offset < len );
                        skin_verts->insert( conn[offset] );
                    }
                }
            }
        }
 
        // If user requested Range of *edges* on the skin...
        if( find_edges )
        {
            // Search list of existing adjacent edges for any that are on the skin
            adj.clear();
            rval = thisMB->get_adjacencies( &*it, 1, 1, false, adj );
            if( MB_SUCCESS != rval ) return rval;
            for( i = adj.begin(); i != adj.end(); ++i )
            {
                rval = thisMB->get_connectivity( *i, conn, len, true );
                if( MB_SUCCESS != rval ) return rval;
 
                // bool equality expression within find_and_unmark call
                // will be evaluate to the index of *it in the conn array.
                //
                // Note that the order of the terms in the if statement is important.
                // We want to unmark any existing skin edges even if we aren't
                // returning them.  Otherwise we'll end up creating duplicates
                // if create_edges is true and skin_edges is not.
                if( adj_edges.find_and_unmark( conn, ( conn[1] == *it ), face ) && skin_edges )
                {
                    if( reversed_edges && edge_reversed( face, conn ) )
                        reversed_edges->insert( *i );
                    else
                        skin_edges->insert( *i );
                }
            }
        }
 
        // If the user requested that we create new edges for sides
        // on the skin for which there is no existing edge, and there
        // are still skin sides for which no corresponding edge was found...
        if( create_edges && adj_edges.num_skin() )
        {
            // Create any skin edges that don't exist
            for( AdjSides< 2 >::const_iterator p = adj_edges.begin(); p != adj_edges.end(); ++p )
            {
                if( p->skin() )
                {
                    EntityHandle edge, ec[] = { *it, p->handles[0] };
                    rval = create_side( this_set, p->adj_elem, MBEDGE, ec, edge );
                    if( MB_SUCCESS != rval ) return rval;
                    if( skin_edges ) skin_edges->insert( edge );
                }
            }
        }
 
    }  // end for each vertex
 
    return MB_SUCCESS;
}
 
ErrorCode Skinner::find_skin_vertices_3D( const EntityHandle this_set,
                                          Tag tag,
                                          const Range& entities,
                                          Range* skin_verts,
                                          Range* skin_faces,
                                          Range* reversed_faces,
                                          bool create_faces,
                                          bool corners_only )
{
    // This function iterates over all the vertices contained in the
    // input vol elem list.  For each such vertex, it then iterates over
    // all of the sides of the vol elements adjacent to the vertex.
    // If an adjacent side is the side of only one of the input
    // elements, then that side is on the skin.
    //
    // This algorithm will visit each skin vertex exactly once.  It
    // will visit each skin side once for each vertex in the side.
    //
    // This function expects the caller to have created the passed bit
    // tag and set it to one only for the elements in the passed range.  This
    // tag is used to do a fast intersection of the elements adjacent to a
    // vertex with the elements in the input range (discard any for which the
    // tag is not set to one.)
    //
    // For each vertex, iterate over each adjacent element.  Construct
    // lists of the sides of each adjacent element that contain the vertex.
    //
    // A list of three-vertex sides is kept for all triangular faces,
    // included three-vertex faces of type MBPOLYGON.  Putting polygons
    // in the same list ensures that we find polyhedron and non-polyhedron
    // elements that are adjacent.
    //
    // A list of four-vertex sides is kept for all quadrilateral faces,
    // including four-vertex faces of type MBPOLYGON.
    //
    // Sides with more than four vertices must have an explicit MBPOLYGON
    // element representing them because MBPOLYHEDRON connectivity is a
    // list of faces rather than vertices.  So the third list (vertices>=5),
    // need contain only the handle of the face rather than the vertex handles.
 
    ErrorCode rval;
    std::vector< EntityHandle >::iterator i, j;
    Range::iterator hint;
    if( skin_verts ) hint = skin_verts->begin();
    std::vector< EntityHandle > storage, storage2;  // temp storage for conn lists
    const EntityHandle *conn, *conn2;
    int len, len2;
    bool find_faces = skin_faces || create_faces;
    int clen, side, sense, offset, indices[9];
    EntityType face_type;
    EntityHandle elem;
    bool printed_nonconformal_ho_warning = false;
 
    if( !entities.all_of_dimension( 3 ) ) return MB_TYPE_OUT_OF_RANGE;
 
    // get all the vertices
    Range verts;
    rval = thisMB->get_connectivity( entities, verts, true );
    if( MB_SUCCESS != rval ) return rval;
    // if there are polyhedra in the input list, need to make another
    // call to get vertices from faces
    if( !verts.all_of_dimension( 0 ) )
    {
        Range::iterator it = verts.upper_bound( MBVERTEX );
        Range pfaces;
        pfaces.merge( it, verts.end() );
        verts.erase( it, verts.end() );
        rval = thisMB->get_connectivity( pfaces, verts, true );
        if( MB_SUCCESS != rval ) return rval;
        assert( verts.all_of_dimension( 0 ) );
    }
 
    AdjSides< 4 > adj_quads;  // 4-node sides adjacent to a vertex
    AdjSides< 3 > adj_tris;   // 3-node sides adjacent to a vertex
    AdjSides< 2 > adj_poly;   // n-node sides (n>5) adjacent to vertex
                              // (must have an explicit polygon, so store
                              // polygon handle rather than vertices.)
    std::vector< char > tag_vals;
    std::vector< EntityHandle > adj;
    for( Range::const_iterator it = verts.begin(); it != verts.end(); ++it )
    {
        bool higher_order = false;
 
        // get all adjacent elements
        adj.clear();
        rval = thisMB->get_adjacencies( &*it, 1, 3, false, adj );
        if( MB_SUCCESS != rval ) return rval;
        if( adj.empty() ) continue;
 
        // remove those not tagged (intersect with input range)
        i = j = adj.begin();
        tag_vals.resize( adj.size() );
        rval = thisMB->tag_get_data( tag, &adj[0], adj.size(), &tag_vals[0] );
        if( MB_SUCCESS != rval ) return rval;
        for( ; i != adj.end(); ++i )
            if( tag_vals[i - adj.begin()] ) *( j++ ) = *i;
        adj.erase( j, adj.end() );
 
        // Build lists of sides of 3D element adjacent to the current vertex
        adj_quads.clear();  // store three other vertices for each adjacent quad face
        adj_tris.clear();   // store two other vertices for each adjacent tri face
        adj_poly.clear();   // store handle of each adjacent polygonal face
        int idx;
        for( i = adj.begin(); i != adj.end(); ++i )
        {
            const EntityType type = TYPE_FROM_HANDLE( *i );
 
            // Special case for POLYHEDRA
            if( type == MBPOLYHEDRON )
            {
                rval = thisMB->get_connectivity( *i, conn, len );
                if( MB_SUCCESS != rval ) return rval;
                for( int k = 0; k < len; ++k )
                {
                    rval = thisMB->get_connectivity( conn[k], conn2, len2, true, &storage2 );
                    if( MB_SUCCESS != rval ) return rval;
                    idx = std::find( conn2, conn2 + len2, *it ) - conn2;
                    if( idx == len2 )  // vertex not in this face
                        continue;
 
                    // Treat 3- and 4-vertex faces specially, so that
                    // if the mesh contains both elements and polyhedra,
                    // we don't miss one type adjacent to the other.
                    switch( len2 )
                    {
                        case 3:
                            adj_tris.insert( conn2, idx, *i, k );
                            break;
                        case 4:
                            adj_quads.insert( conn2, idx, *i, k );
                            break;
                        default:
                            adj_poly.insert( conn + k, 1, *i, k );
                            break;
                    }
                }
            }
            else
            {
                rval = thisMB->get_connectivity( *i, conn, len, false, &storage );
                if( MB_SUCCESS != rval ) return rval;
 
                idx = std::find( conn, conn + len, *it ) - conn;
                assert( idx != len );
 
                if( len > CN::VerticesPerEntity( type ) )
                {
                    higher_order = true;
                    // skip higher-order nodes for now
                    if( idx >= CN::VerticesPerEntity( type ) )
                    {
                        if( !printed_nonconformal_ho_warning )
                        {
                            printed_nonconformal_ho_warning = true;
                            std::cerr << "Non-conformal higher-order mesh detected in skinner: "
                                      << "vertex " << ID_FROM_HANDLE( *it ) << " is a corner in "
                                      << "some elements and a higher-order node in others" << std::endl;
                        }
                        continue;
                    }
                }
 
                // For each side of the element...
                const int num_faces = CN::NumSubEntities( type, 2 );
                for( int f = 0; f < num_faces; ++f )
                {
                    int num_vtx;
                    const short* face_indices = CN::SubEntityVertexIndices( type, 2, f, face_type, num_vtx );
                    const short face_idx = std::find( face_indices, face_indices + num_vtx, (short)idx ) - face_indices;
                    // skip sides that do not contain vertex from outer loop
                    if( face_idx == num_vtx ) continue;  // current vertex not in this face
 
                    assert( num_vtx <= 4 );  // polyhedra handled above
                    switch( face_type )
                    {
                        case MBTRI:
                            adj_tris.insert( conn, face_idx, *i, f, face_indices );
                            break;
                        case MBQUAD:
                            adj_quads.insert( conn, face_idx, *i, f, face_indices );
                            break;
                        default:
                            return MB_TYPE_OUT_OF_RANGE;
                    }
                }
            }
        }  // end for (adj[3])
 
        // If vertex is not on skin, advance to next vertex
        if( 0 == ( adj_tris.num_skin() + adj_quads.num_skin() + adj_poly.num_skin() ) ) continue;
 
        // If user requested that skin *vertices* be passed back...
        if( skin_verts )
        {
            // Put skin vertex in output list
            hint = skin_verts->insert( hint, *it );
 
            // Add mid-edge and mid-face nodes to vertex list
            if( !corners_only && higher_order )
            {
                for( AdjSides< 3 >::const_iterator t = adj_tris.begin(); t != adj_tris.end(); ++t )
                {
                    if( t->skin() )
                    {
                        elem            = t->adj_elem;
                        EntityType type = TYPE_FROM_HANDLE( elem );
 
                        rval = thisMB->get_connectivity( elem, conn, len, false );
                        if( MB_SUCCESS != rval ) return rval;
                        if( !CN::HasMidNodes( type, len ) ) continue;
 
                        EntityHandle ec[3] = { *it, t->handles[0], t->handles[1] };
                        CN::SideNumber( type, conn, ec, 3, 2, side, sense, offset );
                        CN::SubEntityNodeIndices( type, len, 2, side, face_type, clen, indices );
                        assert( MBTRI == face_type );
                        for( int k = 3; k < clen; ++k )
                            skin_verts->insert( conn[indices[k]] );
                    }
                }
                for( AdjSides< 4 >::const_iterator q = adj_quads.begin(); q != adj_quads.end(); ++q )
                {
                    if( q->skin() )
                    {
                        elem            = q->adj_elem;
                        EntityType type = TYPE_FROM_HANDLE( elem );
 
                        rval = thisMB->get_connectivity( elem, conn, len, false );
                        if( MB_SUCCESS != rval ) return rval;
                        if( !CN::HasMidNodes( type, len ) ) continue;
 
                        EntityHandle ec[4] = { *it, q->handles[0], q->handles[1], q->handles[2] };
                        CN::SideNumber( type, conn, ec, 4, 2, side, sense, offset );
                        CN::SubEntityNodeIndices( type, len, 2, side, face_type, clen, indices );
                        assert( MBQUAD == face_type );
                        for( int k = 4; k < clen; ++k )
                            skin_verts->insert( conn[indices[k]] );
                    }
                }
            }
        }
 
        // If user requested that we pass back the list of 2D elements
        // representing the skin of the mesh...
        if( find_faces )
        {
            // Search list of adjacent faces for any that are on the skin
            adj.clear();
            rval = thisMB->get_adjacencies( &*it, 1, 2, false, adj );
            if( MB_SUCCESS != rval ) return rval;
 
            for( i = adj.begin(); i != adj.end(); ++i )
            {
                rval = thisMB->get_connectivity( *i, conn, len, true );
                if( MB_SUCCESS != rval ) return rval;
                const int idx2 = std::find( conn, conn + len, *it ) - conn;
                if( idx2 >= len )
                {
                    assert( printed_nonconformal_ho_warning );
                    continue;
                }
 
                // Note that the order of the terms in the if statements below
                // is important.  We want to unmark any existing skin faces even
                // if we aren't returning them.  Otherwise we'll end up creating
                // duplicates if create_faces is true.
                if( 3 == len )
                {
                    if( adj_tris.find_and_unmark( conn, idx2, elem ) && skin_faces )
                    {
                        if( reversed_faces && face_reversed( elem, conn, MBTRI ) )
                            reversed_faces->insert( *i );
                        else
                            skin_faces->insert( *i );
                    }
                }
                else if( 4 == len )
                {
                    if( adj_quads.find_and_unmark( conn, idx2, elem ) && skin_faces )
                    {
                        if( reversed_faces && face_reversed( elem, conn, MBQUAD ) )
                            reversed_faces->insert( *i );
                        else
                            skin_faces->insert( *i );
                    }
                }
                else
                {
                    if( adj_poly.find_and_unmark( &*i, 1, elem ) && skin_faces ) skin_faces->insert( *i );
                }
            }
        }
 
        // If user does not want use to create new faces representing
        // sides for which there is currently no explicit element,
        // skip the remaining code and advance the outer loop to the
        // next vertex.
        if( !create_faces ) continue;
 
        // Polyhedra always have explicitly defined faces, so
        // there is no way we could need to create such a face.
        assert( 0 == adj_poly.num_skin() );
 
        // Create any skin tris that don't exist
        if( adj_tris.num_skin() )
        {
            for( AdjSides< 3 >::const_iterator t = adj_tris.begin(); t != adj_tris.end(); ++t )
            {
                if( t->skin() )
                {
                    EntityHandle tri, c[3] = { *it, t->handles[0], t->handles[1] };
                    rval = create_side( this_set, t->adj_elem, MBTRI, c, tri );
                    if( MB_SUCCESS != rval ) return rval;
                    if( skin_faces ) skin_faces->insert( tri );
                }
            }
        }
 
        // Create any skin quads that don't exist
        if( adj_quads.num_skin() )
        {
            for( AdjSides< 4 >::const_iterator q = adj_quads.begin(); q != adj_quads.end(); ++q )
            {
                if( q->skin() )
                {
                    EntityHandle quad, c[4] = { *it, q->handles[0], q->handles[1], q->handles[2] };
                    rval = create_side( this_set, q->adj_elem, MBQUAD, c, quad );
                    if( MB_SUCCESS != rval ) return rval;
                    if( skin_faces ) skin_faces->insert( quad );
                }
            }
        }
    }  // end for each vertex
 
    return MB_SUCCESS;
}
 
}  // namespace moab