/* -*- mode: C++; indent-tabs-mode: nil; -*-
* This file is a part of LEMON, a generic C++ optimization library.
* Copyright (C) 2003-2008
* Egervary Jeno Kombinatorikus Optimalizalasi Kutatocsoport
* (Egervary Research Group on Combinatorial Optimization, EGRES).
* Permission to use, modify and distribute this software is granted
* provided that this copyright notice appears in all copies. For
* precise terms see the accompanying LICENSE file.
* This software is provided "AS IS" with no warranty of any kind,
* express or implied, and with no claim as to its suitability for any
#ifndef LEMON_UNION_FIND_H
#define LEMON_UNION_FIND_H
//!\brief Union-Find data structures.
/// \brief A \e Union-Find data structure implementation
/// The class implements the \e Union-Find data structure.
/// The union operation uses rank heuristic, while
/// the find operation uses path compression.
/// This is a very simple but efficient implementation, providing
/// only four methods: join (union), find, insert and size.
/// For more features see the \ref UnionFindEnum class.
/// It is primarily used in Kruskal algorithm for finding minimal
/// cost spanning tree in a graph.
/// \pre You need to add all the elements by the \ref insert()
template <typename _ItemIntMap>
typedef _ItemIntMap ItemIntMap;
typedef typename ItemIntMap::Key Item;
// If the items vector stores negative value for an item then
// that item is root item and it has -items[it] component size.
// Else the items[it] contains the index of the parent.
bool rep(int idx) const {
int repIndex(int idx) const {
const_cast<int&>(items[idx]) = k;
/// Constructor of the UnionFind class. You should give an item to
/// integer map which will be used from the data structure. If you
/// modify directly this map that may cause segmentation fault,
/// invalid data structure, or infinite loop when you use again
UnionFind(ItemIntMap& m) : index(m) {}
/// \brief Returns the index of the element's component.
/// The method returns the index of the element's component.
/// This is an integer between zero and the number of inserted elements.
int find(const Item& a) {
return repIndex(index[a]);
/// \brief Clears the union-find data structure
/// Erase each item from the data structure.
/// \brief Inserts a new element into the structure.
/// This method inserts a new element into the data structure.
/// The method returns the index of the new component.
int insert(const Item& a) {
/// \brief Joining the components of element \e a and element \e b.
/// This is the \e union operation of the Union-Find structure.
/// Joins the component of element \e a and component of
/// element \e b. If \e a and \e b are in the same component then
/// it returns false otherwise it returns true.
bool join(const Item& a, const Item& b) {
int ka = repIndex(index[a]);
int kb = repIndex(index[b]);
if (items[ka] < items[kb]) {
/// \brief Returns the size of the component of element \e a.
/// Returns the size of the component of element \e a.
int size(const Item& a) {
int k = repIndex(index[a]);
/// \brief A \e Union-Find data structure implementation which
/// is able to enumerate the components.
/// The class implements a \e Union-Find data structure
/// which is able to enumerate the components and the items in
/// a component. If you don't need this feature then perhaps it's
/// better to use the \ref UnionFind class which is more efficient.
/// The union operation uses rank heuristic, while
/// the find operation uses path compression.
/// \pre You need to add all the elements by the \ref insert()
template <typename _ItemIntMap>
typedef _ItemIntMap ItemIntMap;
typedef typename ItemIntMap::Key Item;
// If the parent stores negative value for an item then that item
// is root item and it has ~(items[it].parent) component id. Else
// the items[it].parent contains the index of the parent.
// The \c next and \c prev provides the double-linked
// cyclic list of one component's items.
std::vector<ItemT> items;
std::vector<ClassT> classes;
int firstClass, firstFreeClass;
if (firstFreeClass == -1) {
int cdx = classes.size();
classes.push_back(ClassT());
int cdx = firstFreeClass;
firstFreeClass = classes[firstFreeClass].next;
if (firstFreeItem == -1) {
items.push_back(ItemT());
firstFreeItem = items[firstFreeItem].next;
bool rep(int idx) const {
return items[idx].parent < 0;
int repIndex(int idx) const {
int next = items[idx].parent;
const_cast<int&>(items[idx].parent) = k;
int classIndex(int idx) const {
return ~(items[repIndex(idx)].parent);
void singletonItem(int idx) {
void laceItem(int idx, int rdx) {
items[idx].next = items[rdx].next;
items[items[rdx].next].prev = idx;
void unlaceItem(int idx) {
items[items[idx].prev].next = items[idx].next;
items[items[idx].next].prev = items[idx].prev;
items[idx].next = firstFreeItem;
void spliceItems(int ak, int bk) {
items[items[ak].prev].next = bk;
items[items[bk].prev].next = ak;
int tmp = items[ak].prev;
items[ak].prev = items[bk].prev;
void laceClass(int cls) {
classes[firstClass].prev = cls;
classes[cls].next = firstClass;
void unlaceClass(int cls) {
if (classes[cls].prev != -1) {
classes[classes[cls].prev].next = classes[cls].next;
firstClass = classes[cls].next;
if (classes[cls].next != -1) {
classes[classes[cls].next].prev = classes[cls].prev;
classes[cls].next = firstFreeClass;
UnionFindEnum(ItemIntMap& _index)
: index(_index), items(), firstFreeItem(-1),
firstClass(-1), firstFreeClass(-1) {}
/// \brief Inserts the given element into a new component.
/// This method creates a new component consisting only of the
int insert(const Item& item) {
items[idx].parent = ~cdx;
classes[cdx].firstItem = idx;
/// \brief Inserts the given element into the component of the others.
/// This methods inserts the element \e a into the component of the
void insert(const Item& item, int cls) {
int rdx = classes[cls].firstItem;
++classes[~(items[rdx].parent)].size;
/// \brief Clears the union-find data structure
/// Erase each item from the data structure.
/// \brief Finds the component of the given element.
/// The method returns the component id of the given element.
int find(const Item &item) const {
return ~(items[repIndex(index[item])].parent);
/// \brief Joining the component of element \e a and element \e b.
/// This is the \e union operation of the Union-Find structure.
/// Joins the component of element \e a and component of
/// element \e b. If \e a and \e b are in the same component then
/// returns -1 else returns the remaining class.
int join(const Item& a, const Item& b) {
int ak = repIndex(index[a]);
int bk = repIndex(index[b]);
int acx = ~(items[ak].parent);
int bcx = ~(items[bk].parent);
if (classes[acx].size > classes[bcx].size) {
classes[acx].size += classes[bcx].size;
classes[bcx].size += classes[acx].size;
/// \brief Returns the size of the class.
/// Returns the size of the class.
int size(int cls) const {
return classes[cls].size;
/// \brief Splits up the component.
/// Splitting the component into singleton components (component
int fdx = classes[cls].firstItem;
int idx = items[fdx].next;
int next = items[idx].next;
items[idx].parent = ~cdx;
classes[cdx].firstItem = idx;
classes[~(items[idx].parent)].size = 1;
/// \brief Removes the given element from the structure.
/// Removes the element from its component and if the component becomes
/// empty then removes that component from the component list.
/// \warning It is an error to remove an element which is not in
/// \warning This running time of this operation is proportional to the
/// number of the items in this class.
void erase(const Item& item) {
int fdx = items[idx].next;
int cdx = classIndex(idx);
items[idx].next = firstFreeItem;
classes[cdx].firstItem = fdx;
items[fdx].parent = ~cdx;
/// \brief Gives back a representant item of the component.
/// Gives back a representant item of the component.
Item item(int cls) const {
return items[classes[cls].firstItem].item;
/// \brief Removes the component of the given element from the structure.
/// Removes the component of the given element from the structure.
/// \warning It is an error to give an element which is not in the
void eraseClass(int cls) {
int fdx = classes[cls].firstItem;
items[items[fdx].prev].next = firstFreeItem;
/// \brief LEMON style iterator for the representant items.
/// ClassIt is a lemon style iterator for the components. It iterates
/// on the ids of the classes.
/// \brief Constructor of the iterator
/// Constructor of the iterator
ClassIt(const UnionFindEnum& ufe) : unionFind(&ufe) {
cdx = unionFind->firstClass;
/// \brief Constructor to get invalid iterator
/// Constructor to get invalid iterator
ClassIt(Invalid) : unionFind(0), cdx(-1) {}
/// \brief Increment operator
/// It steps to the next representant item.
cdx = unionFind->classes[cdx].next;
/// \brief Conversion operator
/// It converts the iterator to the current representant item.
/// \brief Equality operator
bool operator==(const ClassIt& i) {
/// \brief Inequality operator
bool operator!=(const ClassIt& i) {
const UnionFindEnum* unionFind;
/// \brief LEMON style iterator for the items of a component.
/// ClassIt is a lemon style iterator for the components. It iterates
/// on the items of a class. By example if you want to iterate on
/// each items of each classes then you may write the next code.
/// for (ClassIt cit(ufe); cit != INVALID; ++cit) {
/// std::cout << "Class: ";
/// for (ItemIt iit(ufe, cit); iit != INVALID; ++iit) {
/// std::cout << toString(iit) << ' ' << std::endl;
/// std::cout << std::endl;
/// \brief Constructor of the iterator
/// Constructor of the iterator. The iterator iterates
/// on the class of the \c item.
ItemIt(const UnionFindEnum& ufe, int cls) : unionFind(&ufe) {
fdx = idx = unionFind->classes[cls].firstItem;
/// \brief Constructor to get invalid iterator
/// Constructor to get invalid iterator
ItemIt(Invalid) : unionFind(0), idx(-1) {}
/// \brief Increment operator
/// It steps to the next item in the class.
idx = unionFind->items[idx].next;
if (idx == fdx) idx = -1;
/// \brief Conversion operator
/// It converts the iterator to the current item.
operator const Item&() const {
return unionFind->items[idx].item;
/// \brief Equality operator
bool operator==(const ItemIt& i) {
/// \brief Inequality operator
bool operator!=(const ItemIt& i) {
const UnionFindEnum* unionFind;
/// \brief A \e Extend-Find data structure implementation which
/// is able to enumerate the components.
/// The class implements an \e Extend-Find data structure which is
/// able to enumerate the components and the items in a
/// component. The data structure is a simplification of the
/// Union-Find structure, and it does not allow to merge two components.
/// \pre You need to add all the elements by the \ref insert()
template <typename _ItemIntMap>
typedef _ItemIntMap ItemIntMap;
typedef typename ItemIntMap::Key Item;
std::vector<ItemT> items;
std::vector<ClassT> classes;
int firstClass, firstFreeClass;
if (firstFreeClass != -1) {
int cdx = firstFreeClass;
firstFreeClass = classes[cdx].next;
classes.push_back(ClassT());
return classes.size() - 1;
if (firstFreeItem != -1) {
firstFreeItem = items[idx].next;
items.push_back(ItemT());
ExtendFindEnum(ItemIntMap& _index)
: index(_index), items(), firstFreeItem(-1),
classes(), firstClass(-1), firstFreeClass(-1) {}
/// \brief Inserts the given element into a new component.
/// This method creates a new component consisting only of the
int insert(const Item& item) {
classes[cdx].next = firstClass;
classes[firstClass].prev = cdx;
classes[cdx].firstItem = idx;
/// \brief Inserts the given element into the given component.
/// This methods inserts the element \e item a into the \e cls class.
void insert(const Item& item, int cls) {
int rdx = classes[cls].firstItem;
items[idx].next = items[rdx].next;
items[items[rdx].next].prev = idx;
/// \brief Clears the union-find data structure
/// Erase each item from the data structure.
firstClass = firstFreeClass = firstFreeItem = -1;
/// \brief Gives back the class of the \e item.
/// Gives back the class of the \e item.
int find(const Item &item) const {
return items[index[item]].cls;
/// \brief Gives back a representant item of the component.
/// Gives back a representant item of the component.
Item item(int cls) const {
return items[classes[cls].firstItem].item;
/// \brief Removes the given element from the structure.
/// Removes the element from its component and if the component becomes
/// empty then removes that component from the component list.
/// \warning It is an error to remove an element which is not in
void erase(const Item &item) {
int cdx = items[idx].cls;
if (idx == items[idx].next) {
if (classes[cdx].prev != -1) {
classes[classes[cdx].prev].next = classes[cdx].next;
firstClass = classes[cdx].next;
if (classes[cdx].next != -1) {
classes[classes[cdx].next].prev = classes[cdx].prev;
classes[cdx].next = firstFreeClass;
classes[cdx].firstItem = items[idx].next;
items[items[idx].next].prev = items[idx].prev;
items[items[idx].prev].next = items[idx].next;
items[idx].next = firstFreeItem;
/// \brief Removes the component of the given element from the structure.
/// Removes the component of the given element from the structure.
/// \warning It is an error to give an element which is not in the
void eraseClass(int cdx) {
int idx = classes[cdx].firstItem;
items[items[idx].prev].next = firstFreeItem;
if (classes[cdx].prev != -1) {
classes[classes[cdx].prev].next = classes[cdx].next;
firstClass = classes[cdx].next;
if (classes[cdx].next != -1) {
classes[classes[cdx].next].prev = classes[cdx].prev;
classes[cdx].next = firstFreeClass;
/// \brief LEMON style iterator for the classes.
/// ClassIt is a lemon style iterator for the components. It iterates
/// on the ids of classes.
/// \brief Constructor of the iterator
/// Constructor of the iterator
ClassIt(const ExtendFindEnum& ufe) : extendFind(&ufe) {
cdx = extendFind->firstClass;
/// \brief Constructor to get invalid iterator
/// Constructor to get invalid iterator
ClassIt(Invalid) : extendFind(0), cdx(-1) {}
/// \brief Increment operator
/// It steps to the next representant item.
cdx = extendFind->classes[cdx].next;
/// \brief Conversion operator
/// It converts the iterator to the current class id.
/// \brief Equality operator
bool operator==(const ClassIt& i) {
/// \brief Inequality operator
bool operator!=(const ClassIt& i) {
const ExtendFindEnum* extendFind;
/// \brief LEMON style iterator for the items of a component.
/// ClassIt is a lemon style iterator for the components. It iterates
/// on the items of a class. By example if you want to iterate on
/// each items of each classes then you may write the next code.
/// for (ClassIt cit(ufe); cit != INVALID; ++cit) {
/// std::cout << "Class: ";
/// for (ItemIt iit(ufe, cit); iit != INVALID; ++iit) {
/// std::cout << toString(iit) << ' ' << std::endl;
/// std::cout << std::endl;
/// \brief Constructor of the iterator
/// Constructor of the iterator. The iterator iterates
/// on the class of the \c item.
ItemIt(const ExtendFindEnum& ufe, int cls) : extendFind(&ufe) {
fdx = idx = extendFind->classes[cls].firstItem;
/// \brief Constructor to get invalid iterator
/// Constructor to get invalid iterator
ItemIt(Invalid) : extendFind(0), idx(-1) {}
/// \brief Increment operator
/// It steps to the next item in the class.
idx = extendFind->items[idx].next;
if (fdx == idx) idx = -1;
/// \brief Conversion operator
/// It converts the iterator to the current item.
operator const Item&() const {
return extendFind->items[idx].item;
/// \brief Equality operator
bool operator==(const ItemIt& i) {
/// \brief Inequality operator
bool operator!=(const ItemIt& i) {
const ExtendFindEnum* extendFind;
/// \brief A \e Union-Find data structure implementation which
/// is able to store a priority for each item and retrieve the minimum of
/// A \e Union-Find data structure implementation which is able to
/// store a priority for each item and retrieve the minimum of each
/// class. In addition, it supports the joining and splitting the
/// components. If you don't need this feature then you makes
/// better to use the \ref UnionFind class which is more efficient.
/// The union-find data strcuture based on a (2, 16)-tree with a
/// tournament minimum selection on the internal nodes. The insert
/// operation takes O(1), the find, set, decrease and increase takes
/// O(log(n)), where n is the number of nodes in the current
/// component. The complexity of join and split is O(log(n)*k),
/// where n is the sum of the number of the nodes and k is the
/// number of joined components or the number of the components
/// \pre You need to add all the elements by the \ref insert()
template <typename _Value, typename _ItemIntMap,
typename _Comp = std::less<_Value> >
typedef typename _ItemIntMap::Key Item;
typedef _ItemIntMap ItemIntMap;
static const int cmax = 16;
std::vector<ClassNode> classes;
if (first_free_class < 0) {
classes.push_back(ClassNode());
int id = first_free_class;
first_free_class = classes[id].next;
void deleteClass(int id) {
classes[id].next = first_free_class;
std::vector<ItemNode> nodes;
if (first_free_node < 0) {
nodes.push_back(ItemNode());
int id = first_free_node;
first_free_node = nodes[id].next;
void deleteNode(int id) {
nodes[id].next = first_free_node;
int findClass(int id) const {
int leftNode(int id) const {
int kd = ~(classes[id].parent);
for (int i = 0; i < classes[id].depth; ++i) {
int nextNode(int id) const {
while (id >= 0 && nodes[id].next == -1) {
nodes[id].prio = nodes[jd].prio;
nodes[id].item = nodes[jd].item;
if (comp(nodes[jd].prio, nodes[id].prio)) {
nodes[id].prio = nodes[jd].prio;
nodes[id].item = nodes[jd].item;
void push(int id, int jd) {
nodes[id].left = nodes[id].right = jd;
nodes[jd].next = nodes[jd].prev = -1;
void pushAfter(int id, int jd) {
int kd = nodes[id].parent;
if (nodes[id].next != -1) {
nodes[nodes[id].next].prev = jd;
nodes[jd].next = nodes[id].next;
void pushRight(int id, int jd) {
nodes[jd].prev = nodes[id].right;
nodes[nodes[id].right].next = jd;
int jd = nodes[id].right;
nodes[nodes[jd].prev].next = -1;
nodes[id].right = nodes[jd].prev;
void splice(int id, int jd) {
nodes[id].size += nodes[jd].size;
nodes[nodes[id].right].next = nodes[jd].left;
nodes[nodes[jd].left].prev = nodes[id].right;
nodes[id].right = nodes[jd].right;
void split(int id, int jd) {
int kd = nodes[id].parent;
nodes[kd].right = nodes[id].prev;
nodes[nodes[id].prev].next = -1;
void pushLeft(int id, int jd) {
nodes[jd].next = nodes[id].left;
nodes[nodes[id].left].prev = jd;
nodes[nodes[jd].next].prev = -1;
nodes[id].left = nodes[jd].next;
void repairLeft(int id) {
int jd = ~(classes[id].parent);
while (nodes[jd].left != -1) {
if (nodes[jd].size == 1) {
if (nodes[jd].parent < 0) {
classes[id].parent = ~kd;
int pd = nodes[jd].parent;
if (nodes[nodes[jd].next].size < cmax) {
pushLeft(nodes[jd].next, nodes[jd].left);
if (nodes[jd].item == nodes[pd].item) {
nodes[nodes[jd].next].prio = nodes[jd].prio;
nodes[nodes[jd].next].item = nodes[jd].item;
int ld = nodes[nodes[jd].next].left;
if (less(ld, nodes[jd].left) ||
nodes[ld].item == nodes[pd].item) {
nodes[jd].item = nodes[ld].item;
nodes[jd].prio = nodes[ld].prio;
if (nodes[nodes[jd].next].item == nodes[ld].item) {
void repairRight(int id) {
int jd = ~(classes[id].parent);
while (nodes[jd].right != -1) {
int kd = nodes[jd].right;
if (nodes[jd].size == 1) {
if (nodes[jd].parent < 0) {
classes[id].parent = ~kd;
int pd = nodes[jd].parent;
if (nodes[nodes[jd].prev].size < cmax) {
pushRight(nodes[jd].prev, nodes[jd].right);
if (nodes[jd].item == nodes[pd].item) {
nodes[nodes[jd].prev].prio = nodes[jd].prio;
nodes[nodes[jd].prev].item = nodes[jd].item;
int ld = nodes[nodes[jd].prev].right;
popRight(nodes[jd].prev);
if (less(ld, nodes[jd].right) ||
nodes[ld].item == nodes[pd].item) {
nodes[jd].item = nodes[ld].item;
nodes[jd].prio = nodes[ld].prio;
if (nodes[nodes[jd].prev].item == nodes[ld].item) {
bool less(int id, int jd) const {
return comp(nodes[id].prio, nodes[jd].prio);
bool equal(int id, int jd) const {
return !less(id, jd) && !less(jd, id);
/// \brief Returns true when the given class is alive.
/// Returns true when the given class is alive, ie. the class is
/// not nested into other class.
bool alive(int cls) const {
return classes[cls].parent < 0;
/// \brief Returns true when the given class is trivial.
/// Returns true when the given class is trivial, ie. the class
/// contains just one item directly.
bool trivial(int cls) const {
return classes[cls].left == -1;
/// \brief Constructs the union-find.
/// Constructs the union-find.
/// \brief _index The index map of the union-find. The data
/// structure uses internally for store references.
HeapUnionFind(ItemIntMap& _index)
: index(_index), first_class(-1),
first_free_class(-1), first_free_node(-1) {}
/// \brief Insert a new node into a new component.
/// Insert a new node into a new component.
/// \param item The item of the new node.
/// \param prio The priority of the new node.
/// \return The class id of the one-item-heap.
int insert(const Item& item, const Value& prio) {
int class_id = newClass();
classes[class_id].parent = ~id;
classes[class_id].depth = 0;
classes[class_id].left = -1;
classes[class_id].right = -1;
classes[first_class].prev = class_id;
classes[class_id].next = first_class;
classes[class_id].prev = -1;
nodes[id].parent = ~class_id;
/// \brief The class of the item.
/// \return The alive class id of the item, which is not nested into
/// The time complexity is O(log(n)).
int find(const Item& item) const {
return findClass(index[item]);
/// \brief Joins the classes.
/// The current function joins the given classes. The parameter is
/// an STL range which should be contains valid class ids. The
/// time complexity is O(log(n)*k) where n is the overall number
/// of the joined nodes and k is the number of classes.
/// \return The class of the joined classes.
/// \pre The range should contain at least two class ids.
template <typename Iterator>
int join(Iterator begin, Iterator end) {
for (Iterator it = begin; it != end; ++it) {
int class_id = newClass();
classes[first_class].prev = class_id;
classes[class_id].next = first_class;
classes[class_id].prev = -1;
classes[class_id].depth = classes[cs[0]].depth;
classes[class_id].parent = classes[cs[0]].parent;
nodes[~(classes[class_id].parent)].parent = ~class_id;
classes[class_id].left = l;
classes[class_id].right = l;
if (classes[l].next != -1) {
classes[classes[l].next].prev = classes[l].prev;
classes[classes[l].prev].next = classes[l].next;
classes[l].depth = leftNode(l);
classes[l].parent = class_id;
for (int ci = 1; ci < int(cs.size()); ++ci) {
if (classes[l].depth > classes[r].depth) {
int id = ~(classes[l].parent);
for (int i = classes[r].depth + 1; i < classes[l].depth; ++i) {
while (id >= 0 && nodes[id].size == cmax) {
int right_id = nodes[id].right;
if (nodes[id].item == nodes[right_id].item) {
pushRight(new_id, ~(classes[r].parent));
if (less(~classes[r].parent, right_id)) {
nodes[new_id].item = nodes[~classes[r].parent].item;
nodes[new_id].prio = nodes[~classes[r].parent].prio;
nodes[new_id].item = nodes[right_id].item;
nodes[new_id].prio = nodes[right_id].prio;
classes[r].parent = ~new_id;
int new_parent = newNode();
nodes[new_parent].next = -1;
nodes[new_parent].prev = -1;
nodes[new_parent].parent = ~l;
push(new_parent, ~(classes[l].parent));
pushRight(new_parent, ~(classes[r].parent));
classes[l].parent = ~new_parent;
pushRight(id, ~(classes[r].parent));
while (id >= 0 && less(~(classes[r].parent), id)) {
nodes[id].prio = nodes[~(classes[r].parent)].prio;
nodes[id].item = nodes[~(classes[r].parent)].item;
} else if (classes[r].depth > classes[l].depth) {
int id = ~(classes[r].parent);
for (int i = classes[l].depth + 1; i < classes[r].depth; ++i) {
while (id >= 0 && nodes[id].size == cmax) {
int left_id = nodes[id].left;
if (nodes[id].prio == nodes[left_id].prio) {
pushLeft(new_id, ~(classes[l].parent));
if (less(~classes[l].parent, left_id)) {
nodes[new_id].item = nodes[~classes[l].parent].item;
nodes[new_id].prio = nodes[~classes[l].parent].prio;
nodes[new_id].item = nodes[left_id].item;
nodes[new_id].prio = nodes[left_id].prio;
classes[l].parent = ~new_id;
int new_parent = newNode();
nodes[new_parent].next = -1;
nodes[new_parent].prev = -1;
nodes[new_parent].parent = ~l;
push(new_parent, ~(classes[r].parent));
pushLeft(new_parent, ~(classes[l].parent));
classes[r].parent = ~new_parent;
pushLeft(id, ~(classes[l].parent));
while (id >= 0 && less(~(classes[l].parent), id)) {
nodes[id].prio = nodes[~(classes[l].parent)].prio;
nodes[id].item = nodes[~(classes[l].parent)].item;
nodes[~(classes[r].parent)].parent = ~l;
classes[l].parent = classes[r].parent;
classes[l].depth = classes[r].depth;
if (classes[l].depth != 0 &&
nodes[~(classes[l].parent)].size +
nodes[~(classes[r].parent)].size <= cmax) {
splice(~(classes[l].parent), ~(classes[r].parent));
deleteNode(~(classes[r].parent));
if (less(~(classes[r].parent), ~(classes[l].parent))) {
nodes[~(classes[l].parent)].prio =
nodes[~(classes[r].parent)].prio;
nodes[~(classes[l].parent)].item =
nodes[~(classes[r].parent)].item;
int new_parent = newNode();
nodes[new_parent].next = nodes[new_parent].prev = -1;
push(new_parent, ~(classes[l].parent));
pushRight(new_parent, ~(classes[r].parent));
classes[l].parent = ~new_parent;
nodes[new_parent].parent = ~l;
if (classes[r].next != -1) {
classes[classes[r].next].prev = classes[r].prev;
classes[classes[r].prev].next = classes[r].next;
classes[r].prev = classes[l].right;
classes[classes[l].right].next = r;
/// \brief Split the class to subclasses.
/// The current function splits the given class. The join, which
/// made the current class, stored a reference to the
/// subclasses. The \c splitClass() member restores the classes
/// and creates the heaps. The parameter is an STL output iterator
/// which will be filled with the subclass ids. The time
/// complexity is O(log(n)*k) where n is the overall number of
/// nodes in the splitted classes and k is the number of the
template <typename Iterator>
void split(int cls, Iterator out) {
{ // splitting union-find
int l = classes[id].left;
classes[l].parent = classes[id].parent;
classes[l].depth = classes[id].depth;
nodes[~(classes[l].parent)].parent = ~l;
classes[classes[id].right].next = first_class;
classes[first_class].prev = classes[id].right;
first_class = classes[id].left;
if (classes[id].next != -1) {
classes[classes[id].next].prev = classes[id].prev;
classes[classes[id].prev].next = classes[id].next;
for (int i = 1; i < int(cs.size()); ++i) {
int l = classes[cs[i]].depth;
while (nodes[nodes[l].parent].left == l) {
while (nodes[l].parent >= 0) {
int new_node = newNode();
nodes[new_node].prev = -1;
nodes[new_node].next = -1;
classes[cs[i]].parent = ~r;
classes[cs[i]].depth = classes[~(nodes[l].parent)].depth;
nodes[r].parent = ~cs[i];
repairRight(~(nodes[l].parent));
/// \brief Gives back the priority of the current item.
/// \return Gives back the priority of the current item.
const Value& operator[](const Item& item) const {
return nodes[index[item]].prio;
/// \brief Sets the priority of the current item.
/// Sets the priority of the current item.
void set(const Item& item, const Value& prio) {
if (comp(prio, nodes[index[item]].prio)) {
} else if (!comp(prio, nodes[index[item]].prio)) {
/// \brief Increase the priority of the current item.
/// Increase the priority of the current item.
void increase(const Item& item, const Value& prio) {
int kd = nodes[id].parent;
while (kd >= 0 && nodes[kd].item == item) {
/// \brief Increase the priority of the current item.
/// Increase the priority of the current item.
void decrease(const Item& item, const Value& prio) {
int kd = nodes[id].parent;
while (kd >= 0 && less(id, kd)) {
/// \brief Gives back the minimum priority of the class.
/// \return Gives back the minimum priority of the class.
const Value& classPrio(int cls) const {
return nodes[~(classes[cls].parent)].prio;
/// \brief Gives back the minimum priority item of the class.
/// \return Gives back the minimum priority item of the class.
const Item& classTop(int cls) const {
return nodes[~(classes[cls].parent)].item;
/// \brief Gives back a representant item of the class.
/// The representant is indpendent from the priorities of the
/// \return Gives back a representant item of the class.
const Item& classRep(int id) const {
int parent = classes[id].parent;
return nodes[parent >= 0 ? classes[id].depth : leftNode(id)].item;
/// \brief LEMON style iterator for the items of a class.
/// ClassIt is a lemon style iterator for the components. It iterates
/// on the items of a class. By example if you want to iterate on
/// each items of each classes then you may write the next code.
/// for (ClassIt cit(huf); cit != INVALID; ++cit) {
/// std::cout << "Class: ";
/// for (ItemIt iit(huf, cit); iit != INVALID; ++iit) {
/// std::cout << toString(iit) << ' ' << std::endl;
/// std::cout << std::endl;
const HeapUnionFind* _huf;
/// \brief Default constructor
ItemIt(const HeapUnionFind& huf, int cls) : _huf(&huf) {
int parent = _huf->classes[id].parent;
_id = _huf->classes[id].depth;
if (_huf->classes[id].next != -1) {
_lid = _huf->classes[_huf->classes[id].next].depth;
_id = _huf->leftNode(id);
/// \brief Increment operator
/// It steps to the next item in the class.
_id = _huf->nextNode(_id);
/// \brief Conversion operator
/// It converts the iterator to the current item.
operator const Item&() const {
return _huf->nodes[_id].item;
/// \brief Equality operator
bool operator==(const ItemIt& i) {
/// \brief Inequality operator
bool operator!=(const ItemIt& i) {
/// \brief Equality operator
bool operator==(Invalid) {
/// \brief Inequality operator
bool operator!=(Invalid) {
/// \brief Class iterator
const HeapUnionFind* _huf;
ClassIt(const HeapUnionFind& huf)
: _huf(&huf), _id(huf.first_class) {}
ClassIt(const HeapUnionFind& huf, int cls)
: _huf(&huf), _id(huf.classes[cls].left) {}
ClassIt(Invalid) : _huf(0), _id(-1) {}
const ClassIt& operator++() {
_id = _huf->classes[_id].next;
/// \brief Equality operator
bool operator==(const ClassIt& i) {
/// \brief Inequality operator
bool operator!=(const ClassIt& i) {
#endif //LEMON_UNION_FIND_H