Java多线程系列--【JUC集合05】- ConcurrentSkipListMap

参考：http://www.cnblogs.com/skywang12345/p/3498556.html

概要

本章对Java.util.concurrent包中的ConcurrentSkipListMap类进行详细的介绍。内容包括：
ConcurrentSkipListMap介绍
ConcurrentSkipListMap原理和数据结构
ConcurrentSkipListMap函数列表
ConcurrentSkipListMap源码分析(JDK1.7.0_40版本)
ConcurrentSkipListMap示例

转载请注明出处：http://www.cnblogs.com/skywang12345/p/3498556.html

 

ConcurrentSkipListMap介绍

ConcurrentSkipListMap是线程安全的有序的哈希表，适用于高并发的场景。
ConcurrentSkipListMap和TreeMap，它们虽然都是有序的哈希表。但是，第一，它们的线程安全机制不同，TreeMap是非线程安全的，而ConcurrentSkipListMap是线程安全的。第二，ConcurrentSkipListMap是通过跳表实现的，而TreeMap是通过红黑树实现的。
关于跳表(Skip List)，它是平衡树的一种替代的数据结构，但是和红黑树不相同的是，跳表对于树的平衡的实现是基于一种随机化的算法的，这样也就是说跳表的插入和删除的工作是比较简单的。

 

ConcurrentSkipListMap原理和数据结构

ConcurrentSkipListMap的数据结构，如下图所示：

说明：

先以数据“7,14,21,32,37,71,85”序列为例，来对跳表进行简单说明。

跳表分为许多层(level)，每一层都可以看作是数据的索引，这些索引的意义就是加快跳表查找数据速度。每一层的数据都是有序的，上一层数据是下一层数据的子集，并且第一层(level 1)包含了全部的数据；层次越高，跳跃性越大，包含的数据越少。
跳表包含一个表头，它查找数据时，是从上往下，从左往右进行查找。现在“需要找出值为32的节点”为例，来对比说明跳表和普遍的链表。

情况1：链表中查找“32”节点
路径如下图1-02所示：

需要4步(红色部分表示路径)。

 

情况2：跳表中查找“32”节点
路径如下图1-03所示：

忽略索引垂直线路上路径的情况下，只需要2步(红色部分表示路径)。

下面说说Java中ConcurrentSkipListMap的数据结构。
(01) ConcurrentSkipListMap继承于AbstractMap类，也就意味着它是一个哈希表。
(02) Index是ConcurrentSkipListMap的内部类，它与“跳表中的索引相对应”。HeadIndex继承于Index，ConcurrentSkipListMap中含有一个HeadIndex的对象head，head是“跳表的表头”。
(03) Index是跳表中的索引，它包含“右索引的指针(right)”，“下索引的指针(down)”和“哈希表节点node”。node是Node的对象，Node也是ConcurrentSkipListMap中的内部类。

 

ConcurrentSkipListMap函数列表

// 构造一个新的空映射，该映射按照键的自然顺序进行排序。
ConcurrentSkipListMap()
// 构造一个新的空映射，该映射按照指定的比较器进行排序。
ConcurrentSkipListMap(Comparatorsuper K> comparator)
// 构造一个新映射，该映射所包含的映射关系与给定映射包含的映射关系相同，并按照键的自然顺序进行排序。
ConcurrentSkipListMap(Mapextends K,? extends V> m)
// 构造一个新映射，该映射所包含的映射关系与指定的有序映射包含的映射关系相同，使用的顺序也相同。
ConcurrentSkipListMap(SortedMapextends V> m)

// 返回与大于等于给定键的最小键关联的键-值映射关系；如果不存在这样的条目，则返回 null。
Map.Entry ceilingEntry(K key)
// 返回大于等于给定键的最小键；如果不存在这样的键，则返回 null。
K ceilingKey(K key)
// 从此映射中移除所有映射关系。
void clear()
// 返回此 ConcurrentSkipListMap 实例的浅表副本。
ConcurrentSkipListMap clone()
// 返回对此映射中的键进行排序的比较器；如果此映射使用键的自然顺序，则返回 null。
Comparatorsuper K> comparator()
// 如果此映射包含指定键的映射关系，则返回 true。
boolean containsKey(Object key)
// 如果此映射为指定值映射一个或多个键，则返回 true。
boolean containsValue(Object value)
// 返回此映射中所包含键的逆序 NavigableSet 视图。
NavigableSet descendingKeySet()
// 返回此映射中所包含映射关系的逆序视图。
ConcurrentNavigableMap descendingMap()
// 返回此映射中所包含的映射关系的 Set 视图。
Set> entrySet()
// 比较指定对象与此映射的相等性。
boolean equals(Object o)
// 返回与此映射中的最小键关联的键-值映射关系；如果该映射为空，则返回 null。
Map.Entry firstEntry()
// 返回此映射中当前第一个（最低）键。
K firstKey()
// 返回与小于等于给定键的最大键关联的键-值映射关系；如果不存在这样的键，则返回 null。
Map.Entry floorEntry(K key)
// 返回小于等于给定键的最大键；如果不存在这样的键，则返回 null。
K floorKey(K key)
// 返回指定键所映射到的值；如果此映射不包含该键的映射关系，则返回 null。
V get(Object key)
// 返回此映射的部分视图，其键值严格小于 toKey。
ConcurrentNavigableMap headMap(K toKey)
// 返回此映射的部分视图，其键小于（或等于，如果 inclusive 为 true）toKey。
ConcurrentNavigableMap headMap(K toKey, boolean inclusive)
// 返回与严格大于给定键的最小键关联的键-值映射关系；如果不存在这样的键，则返回 null。
Map.Entry higherEntry(K key)
// 返回严格大于给定键的最小键；如果不存在这样的键，则返回 null。
K higherKey(K key)
// 如果此映射未包含键-值映射关系，则返回 true。
boolean isEmpty()
// 返回此映射中所包含键的 NavigableSet 视图。
NavigableSet keySet()
// 返回与此映射中的最大键关联的键-值映射关系；如果该映射为空，则返回 null。
Map.Entry lastEntry()
// 返回映射中当前最后一个（最高）键。
K lastKey()
// 返回与严格小于给定键的最大键关联的键-值映射关系；如果不存在这样的键，则返回 null。
Map.Entry lowerEntry(K key)
// 返回严格小于给定键的最大键；如果不存在这样的键，则返回 null。
K lowerKey(K key)
// 返回此映射中所包含键的 NavigableSet 视图。
NavigableSet navigableKeySet()
// 移除并返回与此映射中的最小键关联的键-值映射关系；如果该映射为空，则返回 null。
Map.Entry pollFirstEntry()
// 移除并返回与此映射中的最大键关联的键-值映射关系；如果该映射为空，则返回 null。
Map.Entry pollLastEntry()
// 将指定值与此映射中的指定键关联。
V put(K key, V value)
// 如果指定键已经不再与某个值相关联，则将它与给定值关联。
V putIfAbsent(K key, V value)
// 从此映射中移除指定键的映射关系（如果存在）。
V remove(Object key)
// 只有目前将键的条目映射到给定值时，才移除该键的条目。
boolean remove(Object key, Object value)
// 只有目前将键的条目映射到某一值时，才替换该键的条目。
V replace(K key, V value)
// 只有目前将键的条目映射到给定值时，才替换该键的条目。
boolean replace(K key, V oldValue, V newValue)
// 返回此映射中的键-值映射关系数。
int size()
// 返回此映射的部分视图，其键的范围从 fromKey 到 toKey。
ConcurrentNavigableMap subMap(K fromKey, boolean fromInclusive, K toKey, boolean toInclusive)
// 返回此映射的部分视图，其键值的范围从 fromKey（包括）到 toKey（不包括）。
ConcurrentNavigableMap subMap(K fromKey, K toKey)
// 返回此映射的部分视图，其键大于等于 fromKey。
ConcurrentNavigableMap tailMap(K fromKey)
// 返回此映射的部分视图，其键大于（或等于，如果 inclusive 为 true）fromKey。
ConcurrentNavigableMap tailMap(K fromKey, boolean inclusive)
// 返回此映射中所包含值的 Collection 视图。
Collection values()

 

ConcurrentSkipListMap源码分析(JDK1.7.0_40版本)

ConcurrentSkipListMap.java的完整源码如下：

   1 /*
   2  * ORACLE PROPRIETARY/CONFIDENTIAL. Use is subject to license terms.
   3  *
   4  *
   5  *
   6  *
   7  *
   8  *
   9  *
  10  *
  11  *
  12  *
  13  *
  14  *
  15  *
  16  *
  17  *
  18  *
  19  *
  20  *
  21  *
  22  *
  23  */
  24 
  25 /*
  26  *
  27  *
  28  *
  29  *
  30  *
  31  * Written by Doug Lea with assistance from members of JCP JSR-166
  32  * Expert Group and released to the public domain, as explained at
  33  * http://creativecommons.org/publicdomain/zero/1.0/
  34  */
  35 
  36 package java.util.concurrent;
  37 import java.util.*;
  38 import java.util.concurrent.atomic.*;
  39 
  40 /**
  41  * A scalable concurrent {@link ConcurrentNavigableMap} implementation.
  42  * The map is sorted according to the {@linkplain Comparable natural
  43  * ordering} of its keys, or by a {@link Comparator} provided at map
  44  * creation time, depending on which constructor is used.
  45  *
  46  * 
This class implements a concurrent variant of   47  * href="http://en.wikipedia.org/wiki/Skip_list" target="_top">SkipLists
  48  * providing expected average log(n) time cost for the
  49  * containsKey, get, put and
  50  * remove operations and their variants.  Insertion, removal,
  51  * update, and access operations safely execute concurrently by
  52  * multiple threads.  Iterators are weakly consistent, returning
  53  * elements reflecting the state of the map at some point at or since
  54  * the creation of the iterator.  They do not throw {@link
  55  * ConcurrentModificationException}, and may proceed concurrently with
  56  * other operations. Ascending key ordered views and their iterators
  57  * are faster than descending ones.
  58  *
  59  * 
All Map.Entry pairs returned by methods in this class
  60  * and its views represent snapshots of mappings at the time they were
  61  * produced. They do not support the Entry.setValue
  62  * method. (Note however that it is possible to change mappings in the
  63  * associated map using put, putIfAbsent, or
  64  * replace, depending on exactly which effect you need.)
  65  *
  66  * 
Beware that, unlike in most collections, the size
  67  * method is not a constant-time operation. Because of the
  68  * asynchronous nature of these maps, determining the current number
  69  * of elements requires a traversal of the elements, and so may report
  70  * inaccurate results if this collection is modified during traversal.
  71  * Additionally, the bulk operations putAll, equals,
  72  * toArray, containsValue, and clear are
  73  * not guaranteed to be performed atomically. For example, an
  74  * iterator operating concurrently with a putAll operation
  75  * might view only some of the added elements.
  76  *
  77  * 
This class and its views and iterators implement all of the
  78  * optional methods of the {@link Map} and {@link Iterator}
  79  * interfaces. Like most other concurrent collections, this class does
  80  * not permit the use of null keys or values because some
  81  * null return values cannot be reliably distinguished from the absence of
  82  * elements.
  83  *
  84  * 
This class is a member of the
  85  * docRoot}/../technotes/guides/collections/index.html">
  86  * Java Collections Framework.
  87  *
  88  * @author Doug Lea
  89  * @param  the type of keys maintained by this map
  90  * @param  the type of mapped values
  91  * @since 1.6
  92  */
  93 public class ConcurrentSkipListMap extends AbstractMap
  94     implements ConcurrentNavigableMap,
  95                Cloneable,
  96                java.io.Serializable {
  97     /*
  98      * This class implements a tree-like two-dimensionally linked skip
  99      * list in which the index levels are represented in separate
 100      * nodes from the base nodes holding data.  There are two reasons
 101      * for taking this approach instead of the usual array-based
 102      * structure: 1) Array based implementations seem to encounter
 103      * more complexity and overhead 2) We can use cheaper algorithms
 104      * for the heavily-traversed index lists than can be used for the
 105      * base lists.  Here's a picture of some of the basics for a
 106      * possible list with 2 levels of index:
 107      *
 108      * Head nodes          Index nodes
 109      * +-+    right        +-+                      +-+
 110      * |2|---------------->| |--------------------->| |->null
 111      * +-+                 +-+                      +-+
 112      *  | down              |                        |
 113      *  v                   v                        v
 114      * +-+            +-+  +-+       +-+            +-+       +-+
 115      * |1|----------->| |->| |------>| |----------->| |------>| |->null
 116      * +-+            +-+  +-+       +-+            +-+       +-+
 117      *  v              |    |         |              |         |
 118      * Nodes  next     v    v         v              v         v
 119      * +-+  +-+  +-+  +-+  +-+  +-+  +-+  +-+  +-+  +-+  +-+  +-+
 120      * | |->|A|->|B|->|C|->|D|->|E|->|F|->|G|->|H|->|I|->|J|->|K|->null
 121      * +-+  +-+  +-+  +-+  +-+  +-+  +-+  +-+  +-+  +-+  +-+  +-+
 122      *
 123      * The base lists use a variant of the HM linked ordered set
 124      * algorithm. See Tim Harris, "A pragmatic implementation of
 125      * non-blocking linked lists"
 126      * http://www.cl.cam.ac.uk/~tlh20/publications.html and Maged
 127      * Michael "High Performance Dynamic Lock-Free Hash Tables and
 128      * List-Based Sets"
 129      * http://www.research.ibm.com/people/m/michael/pubs.htm.  The
 130      * basic idea in these lists is to mark the "next" pointers of
 131      * deleted nodes when deleting to avoid conflicts with concurrent
 132      * insertions, and when traversing to keep track of triples
 133      * (predecessor, node, successor) in order to detect when and how
 134      * to unlink these deleted nodes.
 135      *
 136      * Rather than using mark-bits to mark list deletions (which can
 137      * be slow and space-intensive using AtomicMarkedReference), nodes
 138      * use direct CAS'able next pointers.  On deletion, instead of
 139      * marking a pointer, they splice in another node that can be
 140      * thought of as standing for a marked pointer (indicating this by
 141      * using otherwise impossible field values).  Using plain nodes
 142      * acts roughly like "boxed" implementations of marked pointers,
 143      * but uses new nodes only when nodes are deleted, not for every
 144      * link.  This requires less space and supports faster
 145      * traversal. Even if marked references were better supported by
 146      * JVMs, traversal using this technique might still be faster
 147      * because any search need only read ahead one more node than
 148      * otherwise required (to check for trailing marker) rather than
 149      * unmasking mark bits or whatever on each read.
 150      *
 151      * This approach maintains the essential property needed in the HM
 152      * algorithm of changing the next-pointer of a deleted node so
 153      * that any other CAS of it will fail, but implements the idea by
 154      * changing the pointer to point to a different node, not by
 155      * marking it.  While it would be possible to further squeeze
 156      * space by defining marker nodes not to have key/value fields, it
 157      * isn't worth the extra type-testing overhead.  The deletion
 158      * markers are rarely encountered during traversal and are
 159      * normally quickly garbage collected. (Note that this technique
 160      * would not work well in systems without garbage collection.)
 161      *
 162      * In addition to using deletion markers, the lists also use
 163      * nullness of value fields to indicate deletion, in a style
 164      * similar to typical lazy-deletion schemes.  If a node's value is
 165      * null, then it is considered logically deleted and ignored even
 166      * though it is still reachable. This maintains proper control of
 167      * concurrent replace vs delete operations -- an attempted replace
 168      * must fail if a delete beat it by nulling field, and a delete
 169      * must return the last non-null value held in the field. (Note:
 170      * Null, rather than some special marker, is used for value fields
 171      * here because it just so happens to mesh with the Map API
 172      * requirement that method get returns null if there is no
 173      * mapping, which allows nodes to remain concurrently readable
 174      * even when deleted. Using any other marker value here would be
 175      * messy at best.)
 176      *
 177      * Here's the sequence of events for a deletion of node n with
 178      * predecessor b and successor f, initially:
 179      *
 180      *        +------+       +------+      +------+
 181      *   ...  |   b  |------>|   n  |----->|   f  | ...
 182      *        +------+       +------+      +------+
 183      *
 184      * 1. CAS n's value field from non-null to null.
 185      *    From this point on, no public operations encountering
 186      *    the node consider this mapping to exist. However, other
 187      *    ongoing insertions and deletions might still modify
 188      *    n's next pointer.
 189      *
 190      * 2. CAS n's next pointer to point to a new marker node.
 191      *    From this point on, no other nodes can be appended to n.
 192      *    which avoids deletion errors in CAS-based linked lists.
 193      *
 194      *        +------+       +------+      +------+       +------+
 195      *   ...  |   b  |------>|   n  |----->|marker|------>|   f  | ...
 196      *        +------+       +------+      +------+       +------+
 197      *
 198      * 3. CAS b's next pointer over both n and its marker.
 199      *    From this point on, no new traversals will encounter n,
 200      *    and it can eventually be GCed.
 201      *        +------+                                    +------+
 202      *   ...  |   b  |----------------------------------->|   f  | ...
 203      *        +------+                                    +------+
 204      *
 205      * A failure at step 1 leads to simple retry due to a lost race
 206      * with another operation. Steps 2-3 can fail because some other
 207      * thread noticed during a traversal a node with null value and
 208      * helped out by marking and/or unlinking.  This helping-out
 209      * ensures that no thread can become stuck waiting for progress of
 210      * the deleting thread.  The use of marker nodes slightly
 211      * complicates helping-out code because traversals must track
 212      * consistent reads of up to four nodes (b, n, marker, f), not
 213      * just (b, n, f), although the next field of a marker is
 214      * immutable, and once a next field is CAS'ed to point to a
 215      * marker, it never again changes, so this requires less care.
 216      *
 217      * Skip lists add indexing to this scheme, so that the base-level
 218      * traversals start close to the locations being found, inserted
 219      * or deleted -- usually base level traversals only traverse a few
 220      * nodes. This doesn't change the basic algorithm except for the
 221      * need to make sure base traversals start at predecessors (here,
 222      * b) that are not (structurally) deleted, otherwise retrying
 223      * after processing the deletion.
 224      *
 225      * Index levels are maintained as lists with volatile next fields,
 226      * using CAS to link and unlink.  Races are allowed in index-list
 227      * operations that can (rarely) fail to link in a new index node
 228      * or delete one. (We can't do this of course for data nodes.)
 229      * However, even when this happens, the index lists remain sorted,
 230      * so correctly serve as indices.  This can impact performance,
 231      * but since skip lists are probabilistic anyway, the net result
 232      * is that under contention, the effective "p" value may be lower
 233      * than its nominal value. And race windows are kept small enough
 234      * that in practice these failures are rare, even under a lot of
 235      * contention.
 236      *
 237      * The fact that retries (for both base and index lists) are
 238      * relatively cheap due to indexing allows some minor
 239      * simplifications of retry logic. Traversal restarts are
 240      * performed after most "helping-out" CASes. This isn't always
 241      * strictly necessary, but the implicit backoffs tend to help
 242      * reduce other downstream failed CAS's enough to outweigh restart
 243      * cost.  This worsens the worst case, but seems to improve even
 244      * highly contended cases.
 245      *
 246      * Unlike most skip-list implementations, index insertion and
 247      * deletion here require a separate traversal pass occuring after
 248      * the base-level action, to add or remove index nodes.  This adds
 249      * to single-threaded overhead, but improves contended
 250      * multithreaded performance by narrowing interference windows,
 251      * and allows deletion to ensure that all index nodes will be made
 252      * unreachable upon return from a public remove operation, thus
 253      * avoiding unwanted garbage retention. This is more important
 254      * here than in some other data structures because we cannot null
 255      * out node fields referencing user keys since they might still be
 256      * read by other ongoing traversals.
 257      *
 258      * Indexing uses skip list parameters that maintain good search
 259      * performance while using sparser-than-usual indices: The
 260      * hardwired parameters k=1, p=0.5 (see method randomLevel) mean
 261      * that about one-quarter of the nodes have indices. Of those that
 262      * do, half have one level, a quarter have two, and so on (see
 263      * Pugh's Skip List Cookbook, sec 3.4).  The expected total space
 264      * requirement for a map is slightly less than for the current
 265      * implementation of java.util.TreeMap.
 266      *
 267      * Changing the level of the index (i.e, the height of the
 268      * tree-like structure) also uses CAS. The head index has initial
 269      * level/height of one. Creation of an index with height greater
 270      * than the current level adds a level to the head index by
 271      * CAS'ing on a new top-most head. To maintain good performance
 272      * after a lot of removals, deletion methods heuristically try to
 273      * reduce the height if the topmost levels appear to be empty.
 274      * This may encounter races in which it possible (but rare) to
 275      * reduce and "lose" a level just as it is about to contain an
 276      * index (that will then never be encountered). This does no
 277      * structural harm, and in practice appears to be a better option
 278      * than allowing unrestrained growth of levels.
 279      *
 280      * The code for all this is more verbose than you'd like. Most
 281      * operations entail locating an element (or position to insert an
 282      * element). The code to do this can't be nicely factored out
 283      * because subsequent uses require a snapshot of predecessor
 284      * and/or successor and/or value fields which can't be returned
 285      * all at once, at least not without creating yet another object
 286      * to hold them -- creating such little objects is an especially
 287      * bad idea for basic internal search operations because it adds
 288      * to GC overhead.  (This is one of the few times I've wished Java
 289      * had macros.) Instead, some traversal code is interleaved within
 290      * insertion and removal operations.  The control logic to handle
 291      * all the retry conditions is sometimes twisty. Most search is
 292      * broken into 2 parts. findPredecessor() searches index nodes
 293      * only, returning a base-level predecessor of the key. findNode()
 294      * finishes out the base-level search. Even with this factoring,
 295      * there is a fair amount of near-duplication of code to handle
 296      * variants.
 297      *
 298      * For explanation of algorithms sharing at least a couple of
 299      * features with this one, see Mikhail Fomitchev's thesis
 300      * (http://www.cs.yorku.ca/~mikhail/), Keir Fraser's thesis
 301      * (http://www.cl.cam.ac.uk/users/kaf24/), and Hakan Sundell's
 302      * thesis (http://www.cs.chalmers.se/~phs/).
 303      *
 304      * Given the use of tree-like index nodes, you might wonder why
 305      * this doesn't use some kind of search tree instead, which would
 306      * support somewhat faster search operations. The reason is that
 307      * there are no known efficient lock-free insertion and deletion
 308      * algorithms for search trees. The immutability of the "down"
 309      * links of index nodes (as opposed to mutable "left" fields in
 310      * true trees) makes this tractable using only CAS operations.
 311      *
 312      * Notation guide for local variables
 313      * Node:         b, n, f    for  predecessor, node, successor
 314      * Index:        q, r, d    for index node, right, down.
 315      *               t          for another index node
 316      * Head:         h
 317      * Levels:       j
 318      * Keys:         k, key
 319      * Values:       v, value
 320      * Comparisons:  c
 321      */
 322 
 323     private static final long serialVersionUID = -8627078645895051609L;
 324 
 325     /**
 326      * Generates the initial random seed for the cheaper per-instance
 327      * random number generators used in randomLevel.
 328      */
 329     private static final Random seedGenerator = new Random();
 330 
 331     /**
 332      * Special value used to identify base-level header
 333      */
 334     private static final Object BASE_HEADER = new Object();
 335 
 336     /**
 337      * The topmost head index of the skiplist.
 338      */
 339     private transient volatile HeadIndex head;
 340 
 341     /**
 342      * The comparator used to maintain order in this map, or null
 343      * if using natural ordering.
 344      * @serial
 345      */
 346     private final Comparatorsuper K> comparator;
 347 
 348     /**
 349      * Seed for simple random number generator.  Not volatile since it
 350      * doesn't matter too much if different threads don't see updates.
 351      */
 352     private transient int randomSeed;
 353 
 354     /** Lazily initialized key set */
 355     private transient KeySet keySet;
 356     /** Lazily initialized entry set */
 357     private transient EntrySet entrySet;
 358     /** Lazily initialized values collection */
 359     private transient Values values;
 360     /** Lazily initialized descending key set */
 361     private transient ConcurrentNavigableMap descendingMap;
 362 
 363     /**
 364      * Initializes or resets state. Needed by constructors, clone,
 365      * clear, readObject. and ConcurrentSkipListSet.clone.
 366      * (Note that comparator must be separately initialized.)
 367      */
 368     final void initialize() {
 369         keySet = null;
 370         entrySet = null;
 371         values = null;
 372         descendingMap = null;
 373         randomSeed = seedGenerator.nextInt() | 0x0100; // ensure nonzero
 374         head = new HeadIndex(new Node(null, BASE_HEADER, null),
 375                                   null, null, 1);
 376     }
 377 
 378     /**
 379      * compareAndSet head node
 380      */
 381     private boolean casHead(HeadIndex cmp, HeadIndex val) {
 382         return UNSAFE.compareAndSwapObject(this, headOffset, cmp, val);
 383     }
 384 
 385     /* ---------------- Nodes -------------- */
 386 
 387     /**
 388      * Nodes hold keys and values, and are singly linked in sorted
 389      * order, possibly with some intervening marker nodes. The list is
 390      * headed by a dummy node accessible as head.node. The value field
 391      * is declared only as Object because it takes special non-V
 392      * values for marker and header nodes.
 393      */
 394     static final class Node {
 395         final K key;
 396         volatile Object value;
 397         volatile Node next;
 398 
 399         /**
 400          * Creates a new regular node.
 401          */
 402         Node(K key, Object value, Node next) {
 403             this.key = key;
 404             this.value = value;
 405             this.next = next;
 406         }
 407 
 408         /**
 409          * Creates a new marker node. A marker is distinguished by
 410          * having its value field point to itself.  Marker nodes also
 411          * have null keys, a fact that is exploited in a few places,
 412          * but this doesn't distinguish markers from the base-level
 413          * header node (head.node), which also has a null key.
 414          */
 415         Node(Node next) {
 416             this.key = null;
 417             this.value = this;
 418             this.next = next;
 419         }
 420 
 421         /**
 422          * compareAndSet value field
 423          */
 424         boolean casValue(Object cmp, Object val) {
 425             return UNSAFE.compareAndSwapObject(this, valueOffset, cmp, val);
 426         }
 427 
 428         /**
 429          * compareAndSet next field
 430          */
 431         boolean casNext(Node cmp, Node val) {
 432             return UNSAFE.compareAndSwapObject(this, nextOffset, cmp, val);
 433         }
 434 
 435         /**
 436          * Returns true if this node is a marker. This method isn't
 437          * actually called in any current code checking for markers
 438          * because callers will have already read value field and need
 439          * to use that read (not another done here) and so directly
 440          * test if value points to node.
 441          * @param n a possibly null reference to a node
 442          * @return true if this node is a marker node
 443          */
 444         boolean isMarker() {
 445             return value == this;
 446         }
 447 
 448         /**
 449          * Returns true if this node is the header of base-level list.
 450          * @return true if this node is header node
 451          */
 452         boolean isBaseHeader() {
 453             return value == BASE_HEADER;
 454         }
 455 
 456         /**
 457          * Tries to append a deletion marker to this node.
 458          * @param f the assumed current successor of this node
 459          * @return true if successful
 460          */
 461         boolean appendMarker(Node f) {
 462             return casNext(f, new Node(f));
 463         }
 464 
 465         /**
 466          * Helps out a deletion by appending marker or unlinking from
 467          * predecessor. This is called during traversals when value
 468          * field seen to be null.
 469          * @param b predecessor
 470          * @param f successor
 471          */
 472         void helpDelete(Node b, Node f) {
 473             /*
 474              * Rechecking links and then doing only one of the
 475              * help-out stages per call tends to minimize CAS
 476              * interference among helping threads.
 477              */
 478             if (f == next && this == b.next) {
 479                 if (f == null || f.value != f) // not already marked
 480                     appendMarker(f);
 481                 else
 482                     b.casNext(this, f.next);
 483             }
 484         }
 485 
 486         /**
 487          * Returns value if this node contains a valid key-value pair,
 488          * else null.
 489          * @return this node's value if it isn't a marker or header or
 490          * is deleted, else null.
 491          */
 492         V getValidValue() {
 493             Object v = value;
 494             if (v == this || v == BASE_HEADER)
 495                 return null;
 496             return (V)v;
 497         }
 498 
 499         /**
 500          * Creates and returns a new SimpleImmutableEntry holding current
 501          * mapping if this node holds a valid value, else null.
 502          * @return new entry or null
 503          */
 504         AbstractMap.SimpleImmutableEntry createSnapshot() {
 505             V v = getValidValue();
 506             if (v == null)
 507                 return null;
 508             return new AbstractMap.SimpleImmutableEntry(key, v);
 509         }
 510 
 511         // UNSAFE mechanics
 512 
 513         private static final sun.misc.Unsafe UNSAFE;
 514         private static final long valueOffset;
 515         private static final long nextOffset;
 516 
 517         static {
 518             try {
 519                 UNSAFE = sun.misc.Unsafe.getUnsafe();
 520                 Class k = Node.class;
 521                 valueOffset = UNSAFE.objectFieldOffset
 522                     (k.getDeclaredField("value"));
 523                 nextOffset = UNSAFE.objectFieldOffset
 524                     (k.getDeclaredField("next"));
 525             } catch (Exception e) {
 526                 throw new Error(e);
 527             }
 528         }
 529     }
 530 
 531     /* ---------------- Indexing -------------- */
 532 
 533     /**
 534      * Index nodes represent the levels of the skip list.  Note that
 535      * even though both Nodes and Indexes have forward-pointing
 536      * fields, they have different types and are handled in different
 537      * ways, that can't nicely be captured by placing field in a
 538      * shared abstract class.
 539      */
 540     static class Index {
 541         final Node node;
 542         final Index down;
 543         volatile Index right;
 544 
 545         /**
 546          * Creates index node with given values.
 547          */
 548         Index(Node node, Index down, Index right) {
 549             this.node = node;
 550             this.down = down;
 551             this.right = right;
 552         }
 553 
 554         /**
 555          * compareAndSet right field
 556          */
 557         final boolean casRight(Index cmp, Index val) {
 558             return UNSAFE.compareAndSwapObject(this, rightOffset, cmp, val);
 559         }
 560 
 561         /**
 562          * Returns true if the node this indexes has been deleted.
 563          * @return true if indexed node is known to be deleted
 564          */
 565         final boolean indexesDeletedNode() {
 566             return node.value == null;
 567         }
 568 
 569         /**
 570          * Tries to CAS newSucc as successor.  To minimize races with
 571          * unlink that may lose this index node, if the node being
 572          * indexed is known to be deleted, it doesn't try to link in.
 573          * @param succ the expected current successor
 574          * @param newSucc the new successor
 575          * @return true if successful
 576          */
 577         final boolean link(Index succ, Index newSucc) {
 578             Node n = node;
 579             newSucc.right = succ;
 580             return n.value != null && casRight(succ, newSucc);
 581         }
 582 
 583         /**
 584          * Tries to CAS right field to skip over apparent successor
 585          * succ.  Fails (forcing a retraversal by caller) if this node
 586          * is known to be deleted.
 587          * @param succ the expected current successor
 588          * @return true if successful
 589          */
 590         final boolean unlink(Index succ) {
 591             return !indexesDeletedNode() && casRight(succ, succ.right);
 592         }
 593 
 594         // Unsafe mechanics
 595         private static final sun.misc.Unsafe UNSAFE;
 596         private static final long rightOffset;
 597         static {
 598             try {
 599                 UNSAFE = sun.misc.Unsafe.getUnsafe();
 600                 Class k = Index.class;
 601                 rightOffset = UNSAFE.objectFieldOffset
 602                     (k.getDeclaredField("right"));
 603             } catch (Exception e) {
 604                 throw new Error(e);
 605             }
 606         }
 607     }
 608 
 609     /* ---------------- Head nodes -------------- */
 610 
 611     /**
 612      * Nodes heading each level keep track of their level.
 613      */
 614     static final class HeadIndex extends Index {
 615         final int level;
 616         HeadIndex(Node node, Index down, Index right, int level) {
 617             super(node, down, right);
 618             this.level = level;
 619         }
 620     }
 621 
 622     /* ---------------- Comparison utilities -------------- */
 623 
 624     /**
 625      * Represents a key with a comparator as a Comparable.
 626      *
 627      * Because most sorted collections seem to use natural ordering on
 628      * Comparables (Strings, Integers, etc), most internal methods are
 629      * geared to use them. This is generally faster than checking
 630      * per-comparison whether to use comparator or comparable because
 631      * it doesn't require a (Comparable) cast for each comparison.
 632      * (Optimizers can only sometimes remove such redundant checks
 633      * themselves.) When Comparators are used,
 634      * ComparableUsingComparators are created so that they act in the
 635      * same way as natural orderings. This penalizes use of
 636      * Comparators vs Comparables, which seems like the right
 637      * tradeoff.
 638      */
 639     static final class ComparableUsingComparator implements Comparable {
 640         final K actualKey;
 641         final Comparatorsuper K> cmp;
 642         ComparableUsingComparator(K key, Comparatorsuper K> cmp) {
 643             this.actualKey = key;
 644             this.cmp = cmp;
 645         }
 646         public int compareTo(K k2) {
 647             return cmp.compare(actualKey, k2);
 648         }
 649     }
 650 
 651     /**
 652      * If using comparator, return a ComparableUsingComparator, else
 653      * cast key as Comparable, which may cause ClassCastException,
 654      * which is propagated back to caller.
 655      */
 656     private Comparablesuper K> comparable(Object key)
 657             throws ClassCastException {
 658         if (key == null)
 659             throw new NullPointerException();
 660         if (comparator != null)
 661             return new ComparableUsingComparator((K)key, comparator);
 662         else
 663             return (Comparablesuper K>)key;
 664     }
 665 
 666     /**
 667      * Compares using comparator or natural ordering. Used when the
 668      * ComparableUsingComparator approach doesn't apply.
 669      */
 670     int compare(K k1, K k2) throws ClassCastException {
 671         Comparatorsuper K> cmp = comparator;
 672         if (cmp != null)
 673             return cmp.compare(k1, k2);
 674         else
 675             return ((Comparablesuper K>)k1).compareTo(k2);
 676     }
 677 
 678     /**
 679      * Returns true if given key greater than or equal to least and
 680      * strictly less than fence, bypassing either test if least or
 681      * fence are null. Needed mainly in submap operations.
 682      */
 683     boolean inHalfOpenRange(K key, K least, K fence) {
 684         if (key == null)
 685             throw new NullPointerException();
 686         return ((least == null || compare(key, least) >= 0) &&
 687                 (fence == null || compare(key, fence) <  0));
 688     }
 689 
 690     /**
 691      * Returns true if given key greater than or equal to least and less
 692      * or equal to fence. Needed mainly in submap operations.
 693      */
 694     boolean inOpenRange(K key, K least, K fence) {
 695         if (key == null)
 696             throw new NullPointerException();
 697         return ((least == null || compare(key, least) >= 0) &&
 698                 (fence == null || compare(key, fence) <= 0));
 699     }
 700 
 701     /* ---------------- Traversal -------------- */
 702 
 703     /**
 704      * Returns a base-level node with key strictly less than given key,
 705      * or the base-level header if there is no such node.  Also
 706      * unlinks indexes to deleted nodes found along the way.  Callers
 707      * rely on this side-effect of clearing indices to deleted nodes.
 708      * @param key the key
 709      * @return a predecessor of key
 710      */
 711     private Node findPredecessor(Comparablesuper K> key) {
 712         if (key == null)
 713             throw new NullPointerException(); // don't postpone errors
 714         for (;;) {
 715             Index q = head;
 716             Index r = q.right;
 717             for (;;) {
 718                 if (r != null) {
 719                     Node n = r.node;
 720                     K k = n.key;
 721                     if (n.value == null) {
 722                         if (!q.unlink(r))
 723                             break;           // restart
 724                         r = q.right;         // reread r
 725                         continue;
 726                     }
 727                     if (key.compareTo(k) > 0) {
 728                         q = r;
 729                         r = r.right;
 730                         continue;
 731                     }
 732                 }
 733                 Index d = q.down;
 734                 if (d != null) {
 735                     q = d;
 736                     r = d.right;
 737                 } else
 738                     return q.node;
 739             }
 740         }
 741     }
 742 
 743     /**
 744      * Returns node holding key or null if no such, clearing out any
 745      * deleted nodes seen along the way.  Repeatedly traverses at
 746      * base-level looking for key starting at predecessor returned
 747      * from findPredecessor, processing base-level deletions as
 748      * encountered. Some callers rely on this side-effect of clearing
 749      * deleted nodes.
 750      *
 751      * Restarts occur, at traversal step centered on node n, if:
 752      *
 753      *   (1) After reading n's next field, n is no longer assumed
 754      *       predecessor b's current successor, which means that
 755      *       we don't have a consistent 3-node snapshot and so cannot
 756      *       unlink any subsequent deleted nodes encountered.
 757      *
 758      *   (2) n's value field is null, indicating n is deleted, in
 759      *       which case we help out an ongoing structural deletion
 760      *       before retrying.  Even though there are cases where such
 761      *       unlinking doesn't require restart, they aren't sorted out
 762      *       here because doing so would not usually outweigh cost of
 763      *       restarting.
 764      *
 765      *   (3) n is a marker or n's predecessor's value field is null,
 766      *       indicating (among other possibilities) that
 767      *       findPredecessor returned a deleted node. We can't unlink
 768      *       the node because we don't know its predecessor, so rely
 769      *       on another call to findPredecessor to notice and return
 770      *       some earlier predecessor, which it will do. This check is
 771      *       only strictly needed at beginning of loop, (and the
 772      *       b.value check isn't strictly needed at all) but is done
 773      *       each iteration to help avoid contention with other
 774      *       threads by callers that will fail to be able to change
 775      *       links, and so will retry anyway.
 776      *
 777      * The traversal loops in doPut, doRemove, and findNear all
 778      * include the same three kinds of checks. And specialized
 779      * versions appear in findFirst, and findLast and their
 780      * variants. They can't easily share code because each uses the
 781      * reads of fields held in locals occurring in the orders they
 782      * were performed.
 783      *
 784      * @param key the key
 785      * @return node holding key, or null if no such
 786      */
 787     private Node findNode(Comparablesuper K> key) {
 788         for (;;) {
 789             Node b = findPredecessor(key);
 790             Node n = b.next;
 791             for (;;) {
 792                 if (n == null)
 793                     return null;
 794                 Node f = n.next;
 795                 if (n != b.next)                // inconsistent read
 796                     break;
 797                 Object v = n.value;
 798                 if (v == null) {                // n is deleted
 799                     n.helpDelete(b, f);
 800                     break;
 801                 }
 802                 if (v == n || b.value == null)  // b is deleted
 803                     break;
 804                 int c = key.compareTo(n.key);
 805                 if (c == 0)
 806                     return n;
 807                 if (c < 0)
 808                     return null;
 809                 b = n;
 810                 n = f;
 811             }
 812         }
 813     }
 814 
 815     /**
 816      * Gets value for key using findNode.
 817      * @param okey the key
 818      * @return the value, or null if absent
 819      */
 820     private V doGet(Object okey) {
 821         Comparablesuper K> key = comparable(okey);
 822         /*
 823          * Loop needed here and elsewhere in case value field goes
 824          * null just as it is about to be returned, in which case we
 825          * lost a race with a deletion, so must retry.
 826          */
 827         for (;;) {
 828             Node n = findNode(key);
 829             if (n == null)
 830                 return null;
 831             Object v = n.value;
 832             if (v != null)
 833                 return (V)v;
 834         }
 835     }
 836 
 837     /* ---------------- Insertion -------------- */
 838 
 839     /**
 840      * Main insertion method.  Adds element if not present, or
 841      * replaces value if present and onlyIfAbsent is false.
 842      * @param kkey the key
 843      * @param value  the value that must be associated with key
 844      * @param onlyIfAbsent if should not insert if already present
 845      * @return the old value, or null if newly inserted
 846      */
 847     private V doPut(K kkey, V value, boolean onlyIfAbsent) {
 848         Comparablesuper K> key = comparable(kkey);
 849         for (;;) {
 850             Node b = findPredecessor(key);
 851             Node n = b.next;
 852             for (;;) {
 853                 if (n != null) {
 854                     Node f = n.next;
 855                     if (n != b.next)               // inconsistent read
 856                         break;
 857                     Object v = n.value;
 858                     if (v == null) {               // n is deleted
 859                         n.helpDelete(b, f);
 860                         break;
 861                     }
 862                     if (v == n || b.value == null) // b is deleted
 863                         break;
 864                     int c = key.compareTo(n.key);
 865                     if (c > 0) {
 866                         b = n;
 867                         n = f;
 868                         continue;
 869                     }
 870                     if (c == 0) {
 871                         if (onlyIfAbsent || n.casValue(v, value))
 872                             return (V)v;
 873                         else
 874                             break; // restart if lost race to replace value
 875                     }
 876                     // else c < 0; fall through
 877                 }
 878 
 879                 Node z = new Node(kkey, value, n);
 880                 if (!b.casNext(n, z))
 881                     break;         // restart if lost race to append to b
 882                 int level = randomLevel();
 883                 if (level > 0)
 884                     insertIndex(z, level);
 885                 return null;
 886             }
 887         }
 888     }
 889 
 890     /**
 891      * Returns a random level for inserting a new node.
 892      * Hardwired to k=1, p=0.5, max 31 (see above and
 893      * Pugh's "Skip List Cookbook", sec 3.4).
 894      *
 895      * This uses the simplest of the generators described in George
 896      * Marsaglia's "Xorshift RNGs" paper.  This is not a high-quality
 897      * generator but is acceptable here.
 898      */
 899     private int randomLevel() {
 900         int x = randomSeed;
 901         x ^= x << 13;
 902         x ^= x >>> 17;
 903         randomSeed = x ^= x << 5;
 904         if ((x & 0x80000001) != 0) // test highest and lowest bits
 905             return 0;
 906         int level = 1;
 907         while (((x >>>= 1) & 1) != 0) ++level;
 908         return level;
 909     }
 910 
 911     /**
 912      * Creates and adds index nodes for the given node.
 913      * @param z the node
 914      * @param level the level of the index
 915      */
 916     private void insertIndex(Node z, int level) {
 917         HeadIndex h = head;
 918         int max = h.level;
 919 
 920         if (level <= max) {
 921             Index idx = null;
 922             for (int i = 1; i <= level; ++i)
 923                 idx = new Index(z, idx, null);
 924             addIndex(idx, h, level);
 925 
 926         } else { // Add a new level
 927             /*
 928              * To reduce interference by other threads checking for
 929              * empty levels in tryReduceLevel, new levels are added
 930              * with initialized right pointers. Which in turn requires
 931              * keeping levels in an array to access them while
 932              * creating new head index nodes from the opposite
 933              * direction.
 934              */
 935             level = max + 1;
 936             Index[] idxs = (Index[])new Index[level+1];
 937             Index idx = null;
 938             for (int i = 1; i <= level; ++i)
 939                 idxs[i] = idx = new Index(z, idx, null);
 940 
 941             HeadIndex oldh;
 942             int k;
 943             for (;;) {
 944                 oldh = head;
 945                 int oldLevel = oldh.level;
 946                 if (level <= oldLevel) { // lost race to add level
 947                     k = level;
 948                     break;
 949                 }
 950                 HeadIndex newh = oldh;
 951                 Node oldbase = oldh.node;
 952                 for (int j = oldLevel+1; j <= level; ++j)
 953                     newh = new HeadIndex(oldbase, newh, idxs[j], j);
 954                 if (casHead(oldh, newh)) {
 955                     k = oldLevel;
 956                     break;
 957                 }
 958             }
 959             addIndex(idxs[k], oldh, k);
 960         }
 961     }
 962 
 963     /**
 964      * Adds given index nodes from given level down to 1.
 965      * @param idx the topmost index node being inserted
 966      * @param h the value of head to use to insert. This must be
 967      * snapshotted by callers to provide correct insertion level
 968      * @param indexLevel the level of the index
 969      */
 970     private void addIndex(Index idx, HeadIndex h, int indexLevel) {
 971         // Track next level to insert in case of retries
 972         int insertionLevel = indexLevel;
 973         Comparablesuper K> key = comparable(idx.node.key);
 974         if (key == null) throw new NullPointerException();
 975 
 976         // Similar to findPredecessor, but adding index nodes along
 977         // path to key.
 978         for (;;) {
 979             int j = h.level;
 980             Index q = h;
 981             Index r = q.right;
 982             Index t = idx;
 983             for (;;) {
 984                 if (r != null) {
 985                     Node n = r.node;
 986                     // compare before deletion check avoids needing recheck
 987                     int c = key.compareTo(n.key);
 988                     if (n.value == null) {
 989                         if (!q.unlink(r))
 990                             break;
 991                         r = q.right;
 992                         continue;
 993                     }
 994                     if (c > 0) {
 995                         q = r;
 996                         r = r.right;
 997                         continue;
 998                     }
 999                 }
1000 
1001                 if (j == insertionLevel) {
1002                     // Don't insert index if node already deleted
1003                     if (t.indexesDeletedNode()) {
1004                         findNode(key); // cleans up
1005                         return;
1006                     }
1007                     if (!q.link(r, t))
1008                         break; // restart
1009                     if (--insertionLevel == 0) {
1010                         // need final deletion check before return
1011                         if (t.indexesDeletedNode())
1012                             findNode(key);
1013                         return;
1014                     }
1015                 }
1016 
1017                 if (--j >= insertionLevel && j < indexLevel)
1018                     t = t.down;
1019                 q = q.down;
1020                 r = q.right;
1021             }
1022         }
1023     }
1024 
1025     /* ---------------- Deletion -------------- */
1026 
1027     /**
1028      * Main deletion method. Locates node, nulls value, appends a
1029      * deletion marker, unlinks predecessor, removes associated index
1030      * nodes, and possibly reduces head index level.
1031      *
1032      * Index nodes are cleared out simply by calling findPredecessor.
1033      * which unlinks indexes to deleted nodes found along path to key,
1034      * which will include the indexes to this node.  This is done
1035      * unconditionally. We can't check beforehand whether there are
1036      * index nodes because it might be the case that some or all
1037      * indexes hadn't been inserted yet for this node during initial
1038      * search for it, and we'd like to ensure lack of garbage
1039      * retention, so must call to be sure.
1040      *
1041      * @param okey the key
1042      * @param value if non-null, the value that must be
1043      * associated with key
1044      * @return the node, or null if not found
1045      */
1046     final V doRemove(Object okey, Object value) {
1047         Comparablesuper K> key = comparable(okey);
1048         for (;;) {
1049             Node b = findPredecessor(key);
1050             Node n = b.next;
1051             for (;;) {
1052                 if (n == null)
1053                     return null;
1054                 Node f = n.next;
1055                 if (n != b.next)                    // inconsistent read
1056                     break;
1057                 Object v = n.value;
1058                 if (v == null) {                    // n is deleted
1059                     n.helpDelete(b, f);
1060                     break;
1061                 }
1062                 if (v == n || b.value == null)      // b is deleted
1063                     break;
1064                 int c = key.compareTo(n.key);
1065                 if (c < 0)
1066                     return null;
1067                 if (c > 0) {
1068                     b = n;
1069                     n = f;
1070                     continue;
1071                 }
1072                 if (value != null && !value.equals(v))
1073                     return null;
1074                 if (!n.casValue(v, null))
1075                     break;
1076                 if (!n.appendMarker(f) || !b.casNext(n, f))
1077                     findNode(key);                  // Retry via findNode
1078                 else {
1079                     findPredecessor(key);           // Clean index
1080                     if (head.right == null)
1081                         tryReduceLevel();
1082                 }
1083                 return (V)v;
1084             }
1085         }
1086     }
1087 
1088     /**
1089      * Possibly reduce head level if it has no nodes.  This method can
1090      * (rarely) make mistakes, in which case levels can disappear even
1091      * though they are about to contain index nodes. This impacts
1092      * performance, not correctness.  To minimize mistakes as well as
1093      * to reduce hysteresis, the level is reduced by one only if the
1094      * topmost three levels look empty. Also, if the removed level
1095      * looks non-empty after CAS, we try to change it back quick
1096      * before anyone notices our mistake! (This trick works pretty
1097      * well because this method will practically never make mistakes
1098      * unless current thread stalls immediately before first CAS, in
1099      * which case it is very unlikely to stall again immediately
1100      * afterwards, so will recover.)
1101      *
1102      * We put up with all this rather than just let levels grow
1103      * because otherwise, even a small map that has undergone a large
1104      * number of insertions and removals will have a lot of levels,
1105      * slowing down access more than would an occasional unwanted
1106      * reduction.
1107      */
1108     private void tryReduceLevel() {
1109         HeadIndex h = head;
1110         HeadIndex d;
1111         HeadIndex e;
1112         if (h.level > 3 &&
1113             (d = (HeadIndex)h.down) != null &&
1114             (e = (HeadIndex)d.down) != null &&
1115             e.right == null &&
1116             d.right == null &&
1117             h.right == null &&
1118             casHead(h, d) && // try to set
1119             h.right != null) // recheck
1120             casHead(d, h);   // try to backout
1121     }
1122 
1123     /* ---------------- Finding and removing first element -------------- */
1124 
1125     /**
1126      * Specialized variant of findNode to get first valid node.
1127      * @return first node or null if empty
1128      */
1129     Node findFirst() {
1130         for (;;) {
1131             Node b = head.node;
1132             Node n = b.next;
1133             if (n == null)
1134                 return null;
1135             if (n.value != null)
1136                 return n;
1137             n.helpDelete(b, n.next);
1138         }
1139     }
1140 
1141     /**
1142      * Removes first entry; returns its snapshot.
1143      * @return null if empty, else snapshot of first entry
1144      */
1145     Map.Entry doRemoveFirstEntry() {
1146         for (;;) {
1147             Node b = head.node;
1148             Node n = b.next;
1149             if (n == null)
1150                 return null;
1151             Node f = n.next;
1152             if (n != b.next)
1153                 continue;
1154             Object v = n.value;
1155             if (v == null) {
1156                 n.helpDelete(b, f);
1157                 continue;
1158             }
1159             if (!n.casValue(v, null))
1160                 continue;
1161             if (!n.appendMarker(f) || !b.casNext(n, f))
1162                 findFirst(); // retry
1163             clearIndexToFirst();
1164             return new AbstractMap.SimpleImmutableEntry(n.key, (V)v);
1165         }
1166     }
1167 
1168     /**
1169      * Clears out index nodes associated with deleted first entry.
1170      */
1171     private void clearIndexToFirst() {
1172         for (;;) {
1173             Index q = head;
1174             for (;;) {
1175                 Index r = q.right;
1176                 if (r != null && r.indexesDeletedNode() && !q.unlink(r))
1177                     break;
1178                 if ((q = q.down) == null) {
1179                     if (head.right == null)
1180                         tryReduceLevel();
1181                     return;
1182                 }
1183             }
1184         }
1185     }
1186 
1187 
1188     /* ---------------- Finding and removing last element -------------- */
1189 
1190     /**
1191      * Specialized version of find to get last valid node.
1192      * @return last node or null if empty
1193      */
1194     Node findLast() {
1195         /*
1196          * findPredecessor can't be used to traverse index level
1197          * because this doesn't use comparisons.  So traversals of
1198          * both levels are folded together.
1199          */
1200         Index q = head;
1201         for (;;) {
1202             Index d, r;
1203             if ((r = q.right) != null) {
1204                 if (r.indexesDeletedNode()) {
1205                     q.unlink(r);
1206                     q = head; // restart
1207                 }
1208                 else
1209                     q = r;
1210             } else if ((d = q.down) != null) {
1211                 q = d;
1212             } else {
1213                 Node b = q.node;
1214                 Node n = b.next;
1215                 for (;;) {
1216                     if (n == null)
1217                         return b.isBaseHeader() ? null : b;
1218                     Node f = n.next;            // inconsistent read
1219                     if (n != b.next)
1220                         break;
1221                     Object v = n.value;
1222                     if (v == null) {                 // n is deleted
1223                         n.helpDelete(b, f);
1224                         break;
1225                     }
1226                     if (v == n || b.value == null)   // b is deleted
1227                         break;
1228                     b = n;
1229                     n = f;
1230                 }
1231                 q = head; // restart
1232             }
1233         }
1234     }
1235 
1236     /**
1237      * Specialized variant of findPredecessor to get predecessor of last
1238      * valid node.  Needed when removing the last entry.  It is possible
1239      * that all successors of returned node will have been deleted upon
1240      * return, in which case this method can be retried.
1241      * @return likely predecessor of last node
1242      */
1243     private Node findPredecessorOfLast() {
1244         for (;;) {
1245             Index q = head;
1246             for (;;) {
1247                 Index d, r;
1248                 if ((r = q.right) != null) {
1249                     if (r.indexesDeletedNode()) {
1250                         q.unlink(r);
1251                         break;    // must restart
1252                     }
1253                     // proceed as far across as possible without overshooting
1254                     if (r.node.next != null) {
1255                         q = r;
1256                         continue;
1257                     }
1258                 }
1259                 if ((d = q.down) != null)
1260                     q = d;
1261                 else
1262                     return q.node;
1263             }
1264         }
1265     }
1266 
1267     /**
1268      * Removes last entry; returns its snapshot.
1269      * Specialized variant of doRemove.
1270      * @return null if empty, else snapshot of last entry
1271      */
1272     Map.Entry doRemoveLastEntry() {
1273         for (;;) {
1274             Node b = findPredecessorOfLast();
1275             Node n = b.next;
1276             if (n == null) {
1277                 if (b.isBaseHeader())               // empty
1278                     return null;
1279                 else
1280                     continue; // all b's successors are deleted; retry
1281             }
1282             for (;;) {
1283                 Node f = n.next;
1284                 if (n != b.next)                    // inconsistent read
1285                     break;
1286                 Object v = n.value;
1287                 if (v == null) {                    // n is deleted
1288                     n.helpDelete(b, f);
1289                     break;
1290                 }
1291                 if (v == n || b.value == null)      // b is deleted
1292                     break;
1293                 if (f != null) {
1294                     b = n;
1295                     n = f;
1296                     continue;
1297                 }
1298                 if (!n.casValue(v, null))
1299                     break;
1300                 K key = n.key;
1301                 Comparablesuper K> ck = comparable(key);
1302                 if (!n.appendMarker(f) || !b.casNext(n, f))
1303                     findNode(ck);                  // Retry via findNode
1304                 else {
1305                     findPredecessor(ck);           // Clean index
1306                     if (head.right == null)
1307                         tryReduceLevel();
1308                 }
1309                 return new AbstractMap.SimpleImmutableEntry(key, (V)v);
1310             }
1311         }
1312     }
1313 
1314     /* ---------------- Relational operations -------------- */
1315 
1316     // Control values OR'ed as arguments to findNear
1317 
1318     private static final int EQ = 1;
1319     private static final int LT = 2;
1320     private static final int GT = 0; // Actually checked as !LT
1321 
1322     /**
1323      * Utility for ceiling, floor, lower, higher methods.
1324      * @param kkey the key
1325      * @param rel the relation -- OR'ed combination of EQ, LT, GT
1326      * @return nearest node fitting relation, or null if no such
1327      */
1328     Node findNear(K kkey, int rel) {
1329         Comparablesuper K> key = comparable(kkey);
1330         for (;;) {
1331             Node b = findPredecessor(key);
1332             Node n = b.next;
1333             for (;;) {
1334                 if (n == null)
1335                     return ((rel & LT) == 0 || b.isBaseHeader()) ? null : b;
1336                 Node f = n.next;
1337                 if (n != b.next)                  // inconsistent read
1338                     break;
1339                 Object v = n.value;
1340                 if (v == null) {                  // n is deleted
1341                     n.helpDelete(b, f);
1342                     break;
1343                 }
1344                 if (v == n || b.value == null)    // b is deleted
1345                     break;
1346                 int c = key.compareTo(n.key);
1347                 if ((c == 0 && (rel & EQ) != 0) ||
1348                     (c <  0 && (rel & LT) == 0))
1349                     return n;
1350                 if ( c <= 0 && (rel & LT) != 0)
1351                     return b.isBaseHeader() ? null : b;
1352                 b = n;
1353                 n = f;
1354             }
1355         }
1356     }
1357 
1358     /**
1359      * Returns SimpleImmutableEntry for results of findNear.
1360      * @param key the key
1361      * @param rel the relation -- OR'ed combination of EQ, LT, GT
1362      * @return Entry fitting relation, or null if no such
1363      */
1364     AbstractMap.SimpleImmutableEntry getNear(K key, int rel) {
1365         for (;;) {
1366             Node n = findNear(key, rel);
1367             if (n == null)
1368                 return null;
1369             AbstractMap.SimpleImmutableEntry e = n.createSnapshot();
1370             if (e != null)
1371                 return e;
1372         }
1373     }
1374 
1375 
1376     /* ---------------- Constructors -------------- */
1377 
1378     /**
1379      * Constructs a new, empty map, sorted according to the
1380      * {@linkplain Comparable natural ordering} of the keys.
1381      */
1382     public ConcurrentSkipListMap() {
1383         this.comparator = null;
1384         initialize();
1385     }
1386 
1387     /**
1388      * Constructs a new, empty map, sorted according to the specified
1389      * comparator.
1390      *
1391      * @param comparator the comparator that will be used to order this map.
1392      *        If null, the {@linkplain Comparable natural
1393      *        ordering} of the keys will be used.
1394      */
1395     public ConcurrentSkipListMap(Comparatorsuper K> comparator) {
1396         this.comparator = comparator;
1397         initialize();
1398     }
1399 
1400     /**
1401      * Constructs a new map containing the same mappings as the given map,
1402      * sorted according to the {@linkplain Comparable natural ordering} of
1403      * the keys.
1404      *
1405      * @param  m the map whose mappings are to be placed in this map
1406      * @throws ClassCastException if the keys in m are not
1407      *         {@link Comparable}, or are not mutually comparable
1408      * @throws NullPointerException if the specified map or any of its keys
1409      *         or values are null
1410      */
1411     public ConcurrentSkipListMap(Mapextends K, ? extends V> m) {
1412         this.comparator = null;
1413         initialize();
1414         putAll(m);
1415     }
1416 
1417     /**
1418      * Constructs a new map containing the same mappings and using the
1419      * same ordering as the specified sorted map.
1420      *
1421      * @param m the sorted map whose mappings are to be placed in this
1422      *        map, and whose comparator is to be used to sort this map
1423      * @throws NullPointerException if the specified sorted map or any of
1424      *         its keys or values are null
1425      */
1426     public ConcurrentSkipListMap(SortedMapextends V> m) {
1427         this.comparator = m.comparator();
1428         initialize();
1429         buildFromSorted(m);
1430     }
1431 
1432     /**
1433      * Returns a shallow copy of this ConcurrentSkipListMap
1434      * instance. (The keys and values themselves are not cloned.)
1435      *
1436      * @return a shallow copy of this map
1437      */
1438     public ConcurrentSkipListMap clone() {
1439         ConcurrentSkipListMap clone = null;
1440         try {
1441             clone = (ConcurrentSkipListMap) super.clone();
1442         } catch (CloneNotSupportedException e) {
1443             throw new InternalError();
1444         }
1445 
1446         clone.initialize();
1447         clone.buildFromSorted(this);
1448         return clone;
1449     }
1450 
1451     /**
1452      * Streamlined bulk insertion to initialize from elements of
1453      * given sorted map.  Call only from constructor or clone
1454      * method.
1455      */
1456     private void buildFromSorted(SortedMapextends V> map) {
1457         if (map == null)
1458             throw new NullPointerException();
1459 
1460         HeadIndex h = head;
1461         Node basepred = h.node;
1462 
1463         // Track the current rightmost node at each level. Uses an
1464         // ArrayList to avoid committing to initial or maximum level.
1465         ArrayList> preds = new ArrayList>();
1466 
1467         // initialize
1468         for (int i = 0; i <= h.level; ++i)
1469             preds.add(null);
1470         Index q = h;
1471         for (int i = h.level; i > 0; --i) {
1472             preds.set(i, q);
1473             q = q.down;
1474         }
1475 
1476         Iteratorextends Map.Entryextends K, ? extends V>> it =
1477             map.entrySet().iterator();
1478         while (it.hasNext()) {
1479             Map.Entryextends K, ? extends V> e = it.next();
1480             int j = randomLevel();
1481             if (j > h.level) j = h.level + 1;
1482             K k = e.getKey();
1483             V v = e.getValue();
1484             if (k == null || v == null)
1485                 throw new NullPointerException();
1486             Node z = new Node(k, v, null);
1487             basepred.next = z;
1488             basepred = z;
1489             if (j > 0) {
1490                 Index idx = null;
1491                 for (int i = 1; i <= j; ++i) {
1492                     idx = new Index(z, idx, null);
1493                     if (i > h.level)
1494                         h = new HeadIndex(h.node, h, idx, i);
1495 
1496                     if (i < preds.size()) {
1497                         preds.get(i).right = idx;
1498                         preds.set(i, idx);
1499                     } else
1500                         preds.add(idx);
1501                 }
1502             }
1503         }
1504         head = h;
1505     }
1506 
1507     /* ---------------- Serialization -------------- */
1508 
1509     /**
1510      * Save the state of this map to a stream.
1511      *
1512      * @serialData The key (Object) and value (Object) for each
1513      * key-value mapping represented by the map, followed by
1514      * null. The key-value mappings are emitted in key-order
1515      * (as determined by the Comparator, or by the keys' natural
1516      * ordering if no Comparator).
1517      */
1518     private void writeObject(java.io.ObjectOutputStream s)
1519         throws java.io.IOException {
1520         // Write out the Comparator and any hidden stuff
1521         s.defaultWriteObject();
1522 
1523         // Write out keys and values (alternating)
1524         for (Node n = findFirst(); n != null; n = n.next) {
1525             V v = n.getValidValue();
1526             if (v != null) {
1527                 s.writeObject(n.key);
1528                 s.writeObject(v);
1529             }
1530         }
1531         s.writeObject(null);
1532     }
1533 
1534     /**
1535      * Reconstitute the map from a stream.
1536      */
1537     private void readObject(final java.io.ObjectInputStream s)
1538         throws java.io.IOException, ClassNotFoundException {
1539         // Read in the Comparator and any hidden stuff
1540         s.defaultReadObject();
1541         // Reset transients
1542         initialize();
1543 
1544         /*
1545          * This is nearly identical to buildFromSorted, but is
1546          * distinct because readObject calls can't be nicely adapted
1547          * as the kind of iterator needed by buildFromSorted. (They
1548          * can be, but doing so requires type cheats and/or creation
1549          * of adaptor classes.) It is simpler to just adapt the code.
1550          */
1551 
1552         HeadIndex h = head;
1553         Node basepred = h.node;
1554         ArrayList> preds = new ArrayList>();
1555         for (int i = 0; i <= h.level; ++i)
1556             preds.add(null);
1557         Index q = h;
1558         for (int i = h.level; i > 0; --i) {
1559             preds.set(i, q);
1560             q = q.down;
1561         }
1562 
1563         for (;;) {
1564             Object k = s.readObject();
1565             if (k == null)
1566                 break;
1567             Object v = s.readObject();
1568             if (v == null)
1569                 throw new NullPointerException();
1570             K key = (K) k;
1571             V val = (V) v;
1572             int j = randomLevel();
1573             if (j > h.level) j = h.level + 1;
1574             Node z = new Node(key, val, null);
1575             basepred.next = z;
1576             basepred = z;
1577             if (j > 0) {
1578                 Index idx = null;
1579                 for (int i = 1; i <= j; ++i) {
1580                     idx = new Index(z, idx, null);
1581                     if (i > h.level)
1582                         h = new HeadIndex(h.node, h, idx, i);
1583 
1584                     if (i < preds.size()) {
1585                         preds.get(i).right = idx;
1586                         preds.set(i, idx);
1587                     } else
1588                         preds.add(idx);
1589                 }
1590             }
1591         }
1592         head = h;
1593     }
1594 
1595     /* ------ Map API methods ------ */
1596 
1597     /**
1598      * Returns true if this map contains a mapping for the specified
1599      * key.
1600      *
1601      * @param key key whose presence in this map is to be tested
1602      * @return true if this map contains a mapping for the specified key
1603      * @throws ClassCastException if the specified key cannot be compared
1604      *         with the keys currently in the map
1605      * @throws NullPointerException if the specified key is null
1606      */
1607     public boolean containsKey(Object key) {
1608         return doGet(key) != null;
1609     }
1610 
1611     /**
1612      * Returns the value to which the specified key is mapped,
1613      * or {@code null} if this map contains no mapping for the key.
1614      *
1615      * 
More formally, if this map contains a mapping from a key
1616      * {@code k} to a value {@code v} such that {@code key} compares
1617      * equal to {@code k} according to the map's ordering, then this
1618      * method returns {@code v}; otherwise it returns {@code null}.
1619      * (There can be at most one such mapping.)
1620      *
1621      * @throws ClassCastException if the specified key cannot be compared
1622      *         with the keys currently in the map
1623      * @throws NullPointerException if the specified key is null
1624      */
1625     public V get(Object key) {
1626         return doGet(key);
1627     }
1628 
1629     /**
1630      * Associates the specified value with the specified key in this map.
1631      * If the map previously contained a mapping for the key, the old
1632      * value is replaced.
1633      *
1634      * @param key key with which the specified value is to be associated
1635      * @param value value to be associated with the specified key
1636      * @return the previous value associated with the specified key, or
1637      *         null if there was no mapping for the key
1638      * @throws ClassCastException if the specified key cannot be compared
1639      *         with the keys currently in the map
1640      * @throws NullPointerException if the specified key or value is null
1641      */
1642     public V put(K key, V value) {
1643         if (value == null)
1644             throw new NullPointerException();
1645         return doPut(key, value, false);
1646     }
1647 
1648     /**
1649      * Removes the mapping for the specified key from this map if present.
1650      *
1651      * @param  key key for which mapping should be removed
1652      * @return the previous value associated with the specified key, or
1653      *         null if there was no mapping for the key
1654      * @throws ClassCastException if the specified key cannot be compared
1655      *         with the keys currently in the map
1656      * @throws NullPointerException if the specified key is null
1657      */
1658     public V remove(Object key) {
1659         return doRemove(key, null);
1660     }
1661 
1662     /**
1663      * Returns true if this map maps one or more keys to the
1664      * specified value.  This operation requires time linear in the
1665      * map size. Additionally, it is possible for the map to change
1666      * during execution of this method, in which case the returned
1667      * result may be inaccurate.
1668      *
1669      * @param value value whose presence in this map is to be tested
1670      * @return true if a mapping to value exists;
1671      *         false otherwise
1672      * @throws NullPointerException if the specified value is null
1673      */
1674     public boolean containsValue(Object value) {
1675         if (value == null)
1676             throw new NullPointerException();
1677         for (Node n = findFirst(); n != null; n = n.next) {
1678             V v = n.getValidValue();
1679             if (v != null && value.equals(v))
1680                 return true;
1681         }
1682         return false;
1683     }
1684 
1685     /**
1686      * Returns the number of key-value mappings in this map.  If this map
1687      * contains more than Integer.MAX_VALUE elements, it
1688      * returns Integer.MAX_VALUE.
1689      *
1690      * 
Beware that, unlike in most collections, this method is
1691      * NOT a constant-time operation. Because of the
1692      * asynchronous nature of these maps, determining the current
1693      * number of elements requires traversing them all to count them.
1694      * Additionally, it is possible for the size to change during
1695      * execution of this method, in which case the returned result
1696      * will be inaccurate. Thus, this method is typically not very
1697      * useful in concurrent applications.
1698      *
1699      * @return the number of elements in this map
1700      */
1701     public int size() {
1702         long count = 0;
1703         for (Node n = findFirst(); n != null; n = n.next) {
1704             if (n.getValidValue() != null)
1705                 ++count;
1706         }
1707         return (count >= Integer.MAX_VALUE) ? Integer.MAX_VALUE : (int) count;
1708     }
1709 
1710     /**
1711      * Returns true if this map contains no key-value mappings.
1712      * @return true if this map contains no key-value mappings
1713      */
1714     public boolean isEmpty() {
1715         return findFirst() == null;
1716     }
1717 
1718     /**
1719      * Removes all of the mappings from this map.
1720      */
1721     public void clear() {
1722         initialize();
1723     }
1724 
1725     /* ---------------- View methods -------------- */
1726 
1727     /*
1728      * Note: Lazy initialization works for views because view classes
1729      * are stateless/immutable so it doesn't matter wrt correctness if
1730      * more than one is created (which will only rarely happen).  Even
1731      * so, the following idiom conservatively ensures that the method
1732      * returns the one it created if it does so, not one created by
1733      * another racing thread.
1734      */
1735 
1736     /**
1737      * Returns a {@link NavigableSet} view of the keys contained in this map.
1738      * The set's iterator returns the keys in ascending order.
1739      * The set is backed by the map, so changes to the map are
1740      * reflected in the set, and vice-versa.  The set supports element
1741      * removal, which removes the corresponding mapping from the map,
1742      * via the {@code Iterator.remove}, {@code Set.remove},
1743      * {@code removeAll}, {@code retainAll}, and {@code clear}
1744      * operations.  It does not support the {@code add} or {@code addAll}
1745      * operations.
1746      *
1747      * 
The view's {@code iterator} is a "weakly consistent" iterator
1748      * that will never throw {@link ConcurrentModificationException},
1749      * and guarantees to traverse elements as they existed upon
1750      * construction of the iterator, and may (but is not guaranteed to)
1751      * reflect any modifications subsequent to construction.
1752      *
1753      * 
This method is equivalent to method {@code navigableKeySet}.
1754      *
1755      * @return a navigable set view of the keys in this map
1756      */
1757     public NavigableSet keySet() {
1758         KeySet ks = keySet;
1759         return (ks != null) ? ks : (keySet = new KeySet(this));
1760     }
1761 
1762     public NavigableSet navigableKeySet() {
1763         KeySet ks = keySet;
1764         return (ks != null) ? ks : (keySet = new KeySet(this));
1765     }
1766 
1767     /**
1768      * Returns a {@link Collection} view of the values contained in this map.
1769      * The collection's iterator returns the values in ascending order
1770      * of the corresponding keys.
1771      * The collection is backed by the map, so changes to the map are
1772      * reflected in the collection, and vice-versa.  The collection
1773      * supports element removal, which removes the corresponding
1774      * mapping from the map, via the Iterator.remove,
1775      * Collection.remove, removeAll,
1776      * retainAll and clear operations.  It does not
1777      * support the add or addAll operations.
1778      *
1779      * 
The view's iterator is a "weakly consistent" iterator
1780      * that will never throw {@link ConcurrentModificationException},
1781      * and guarantees to traverse elements as they existed upon
1782      * construction of the iterator, and may (but is not guaranteed to)
1783      * reflect any modifications subsequent to construction.
1784      */
1785     public Collection values() {
1786         Values vs = values;
1787         return (vs != null) ? vs : (values = new Values(this));
1788     }
1789 
1790     /**
1791      * Returns a {@link Set} view of the mappings contained in this map.
1792      * The set's iterator returns the entries in ascending key order.
1793      * The set is backed by the map, so changes to the map are
1794      * reflected in the set, and vice-versa.  The set supports element
1795      * removal, which removes the corresponding mapping from the map,
1796      * via the Iterator.remove, Set.remove,
1797      * removeAll, retainAll and clear
1798      * operations.  It does not support the add or
1799      * addAll operations.
1800      *
1801      * 
The view's iterator is a "weakly consistent" iterator
1802      * that will never throw {@link ConcurrentModificationException},
1803      * and guarantees to traverse elements as they existed upon
1804      * construction of the iterator, and may (but is not guaranteed to)
1805      * reflect any modifications subsequent to construction.
1806      *
1807      * 
The Map.Entry elements returned by
1808      * iterator.next() do not support the
1809      * setValue operation.
1810      *
1811      * @return a set view of the mappings contained in this map,
1812      *         sorted in ascending key order
1813      */
1814     public Set> entrySet() {
1815         EntrySet es = entrySet;
1816         return (es != null) ? es : (entrySet = new EntrySet(this));
1817     }
1818 
1819     public ConcurrentNavigableMap descendingMap() {
1820         ConcurrentNavigableMap dm = descendingMap;
1821         return (dm != null) ? dm : (descendingMap = new SubMap
1822                                     (this, null, false, null, false, true));
1823     }
1824 
1825     public NavigableSet descendingKeySet() {
1826         return descendingMap().navigableKeySet();
1827     }
1828 
1829     /* ---------------- AbstractMap Overrides -------------- */
1830 
1831     /**
1832      * Compares the specified object with this map for equality.
1833      * Returns true if the given object is also a map and the
1834      * two maps represent the same mappings.  More formally, two maps
1835      * m1 and m2 represent the same mappings if
1836      * m1.entrySet().equals(m2.entrySet()).  This
1837      * operation may return misleading results if either map is
1838      * concurrently modified during execution of this method.
1839      *
1840      * @param o object to be compared for equality with this map
1841      * @return true if the specified object is equal to this map
1842      */
1843     public boolean equals(Object o) {
1844         if (o == this)
1845             return true;
1846         if (!(o instanceof Map))
1847             return false;
1848         Map m = (Map) o;
1849         try {
1850             for (Map.Entry e : this.entrySet())
1851                 if (! e.getValue().equals(m.get(e.getKey())))
1852                     return false;
1853             for (Map.Entry e : m.entrySet()) {
1854                 Object k = e.getKey();
1855                 Object v = e.getValue();
1856                 if (k == null || v == null || !v.equals(get(k)))
1857                     return false;
1858             }
1859             return true;
1860         } catch (ClassCastException unused) {
1861             return false;
1862         } catch (NullPointerException unused) {
1863             return false;
1864         }
1865     }
1866 
1867     /* ------ ConcurrentMap API methods ------ */
1868 
1869     /**
1870      * {@inheritDoc}
1871      *
1872      * @return the previous value associated with the specified key,
1873      *         or null if there was no mapping for the key
1874      * @throws ClassCastException if the specified key cannot be compared
1875      *         with the keys currently in the map
1876      * @throws NullPointerException if the specified key or value is null
1877      */
1878     public V putIfAbsent(K key, V value) {
1879         if (value == null)
1880             throw new NullPointerException();
1881         return doPut(key, value, true);
1882     }
1883 
1884     /**
1885      * {@inheritDoc}
1886      *
1887      * @throws ClassCastException if the specified key cannot be compared
1888      *         with the keys currently in the map
1889      * @throws NullPointerException if the specified key is null
1890      */
1891     public boolean remove(Object key, Object value) {
1892         if (key == null)
1893             throw new NullPointerException();
1894         if (value == null)
1895             return false;
1896         return doRemove(key, value) != null;
1897     }
1898 
1899     /**
1900      * {@inheritDoc}
1901      *
1902      * @throws ClassCastException if the specified key cannot be compared
1903      *         with the keys currently in the map
1904      * @throws NullPointerException if any of the arguments are null
1905      */
1906     public boolean replace(K key, V oldValue, V newValue) {
1907         if (oldValue == null || newValue == null)
1908             throw new NullPointerException();
1909         Comparablesuper K> k = comparable(key);
1910         for (;;) {
1911             Node n = findNode(k);
1912             if (n == null)
1913                 return false;
1914             Object v = n.value;
1915             if (v != null) {
1916                 if (!oldValue.equals(v))
1917                     return false;
1918                 if (n.casValue(v, newValue))
1919                     return true;
1920             }
1921         }
1922     }
1923 
1924     /**
1925      * {@inheritDoc}
1926      *
1927      * @return the previous value associated with the specified key,
1928      *         or null if there was no mapping for the key
1929      * @throws ClassCastException if the specified key cannot be compared
1930      *         with the keys currently in the map
1931      * @throws NullPointerException if the specified key or value is null
1932      */
1933     public V replace(K key, V value) {
1934         if (value == null)
1935             throw new NullPointerException();
1936         Comparablesuper K> k = comparable(key);
1937         for (;;) {
1938             Node n = findNode(k);
1939             if (n == null)
1940                 return null;
1941             Object v = n.value;
1942             if (v != null && n.casValue(v, value))
1943                 return (V)v;
1944         }
1945     }
1946 
1947     /* ------ SortedMap API methods ------ */
1948 
1949     public Comparatorsuper K> comparator() {
1950         return comparator;
1951     }
1952 
1953     /**
1954      * @throws NoSuchElementException {@inheritDoc}
1955      */
1956     public K firstKey() {
1957         Node n = findFirst();
1958         if (n == null)
1959             throw new NoSuchElementException();
1960         return n.key;
1961     }
1962 
1963     /**
1964      * @throws NoSuchElementException {@inheritDoc}
1965      */
1966     public K lastKey() {
1967         Node n = findLast();
1968         if (n == null)
1969             throw new NoSuchElementException();
1970         return n.key;
1971     }
1972 
1973     /**
1974      * @throws ClassCastException {@inheritDoc}
1975      * @throws NullPointerException if {@code fromKey} or {@code toKey} is null
1976      * @throws IllegalArgumentException {@inheritDoc}
1977      */
1978     public ConcurrentNavigableMap subMap(K fromKey,
1979                                               boolean fromInclusive,
1980                                               K toKey,
1981                                               boolean toInclusive) {
1982         if (fromKey == null || toKey == null)
1983             throw new NullPointerException();
1984         return new SubMap
1985             (this, fromKey, fromInclusive, toKey, toInclusive, false);
1986     }
1987 
1988     /**
1989      * @throws ClassCastException {@inheritDoc}
1990      * @throws NullPointerException if {@code toKey} is null
1991      * @throws IllegalArgumentException {@inheritDoc}
1992      */
1993     public ConcurrentNavigableMap headMap(K toKey,
1994                                                boolean inclusive) {
1995         if (toKey == null)
1996             throw new NullPointerException();
1997         return new SubMap
1998             (this, null, false, toKey, inclusive, false);
1999     }
2000 
2001     /**
2002      * @throws ClassCastException {@inheritDoc}
2003      * @throws NullPointerException if {@code fromKey} is null
2004      * @throws IllegalArgumentException {@inheritDoc}
2005      */
2006     public ConcurrentNavigableMap tailMap(K fromKey,
2007                                                boolean inclusive) {
2008         if (fromKey == null)
2009             throw new NullPointerException();
2010         return new SubMap
2011             (this, fromKey, inclusive, null, false, false);
2012     }
2013 
2014     /**
2015      * @throws ClassCastException {@inheritDoc}
2016      * @throws NullPointerException if {@code fromKey} or {@code toKey} is null
2017      * @throws IllegalArgumentException {@inheritDoc}
2018      */
2019     public ConcurrentNavigableMap subMap(K fromKey, K toKey) {
2020         return subMap(fromKey, true, toKey, false);
2021     }
2022 
2023     /**
2024      * @throws ClassCastException {@inheritDoc}
2025      * @throws NullPointerException if {@code toKey} is null
2026      * @throws IllegalArgumentException {@inheritDoc}
2027      */
2028     public ConcurrentNavigableMap headMap(K toKey) {
2029         return headMap(toKey, false);
2030     }
2031 
2032     /**
2033      * @throws ClassCastException {@inheritDoc}
2034      * @throws NullPointerException if {@code fromKey} is null
2035      * @throws IllegalArgumentException {@inheritDoc}
2036      */
2037     public ConcurrentNavigableMap tailMap(K fromKey) {
2038         return tailMap(fromKey, true);
2039     }
2040 
2041     /* ---------------- Relational operations -------------- */
2042 
2043     /**
2044      * Returns a key-value mapping associated with the greatest key
2045      * strictly less than the given key, or null if there is
2046      * no such key. The returned entry does not support the
2047      * Entry.setValue method.
2048      *
2049      * @throws ClassCastException {@inheritDoc}
2050      * @throws NullPointerException if the specified key is null
2051      */
2052     public Map.Entry lowerEntry(K key) {
2053         return getNear(key, LT);
2054     }
2055 
2056     /**
2057      * @throws ClassCastException {@inheritDoc}
2058      * @throws NullPointerException if the specified key is null
2059      */
2060     public K lowerKey(K key) {
2061         Node n = findNear(key, LT);
2062         return (n == null) ? null : n.key;
2063     }
2064 
2065     /**
2066      * Returns a key-value mapping associated with the greatest key
2067      * less than or equal to the given key, or null if there
2068      * is no such key. The returned entry does not support
2069      * the Entry.setValue method.
2070      *
2071      * @param key the key
2072      * @throws ClassCastException {@inheritDoc}
2073      * @throws NullPointerException if the specified key is null
2074      */
2075     public Map.Entry floorEntry(K key) {
2076         return getNear(key, LT|EQ);
2077     }
2078 
2079     /**
2080      * @param key the key
2081      * @throws ClassCastException {@inheritDoc}
2082      * @throws NullPointerException if the specified key is null
2083      */
2084     public K floorKey(K key) {
2085         Node n = findNear(key, LT|EQ);
2086         return (n == null) ? null : n.key;
2087     }
2088 
2089     /**
2090      * Returns a key-value mapping associated with the least key
2091      * greater than or equal to the given key, or null if
2092      * there is no such entry. The returned entry does not
2093      * support the Entry.setValue method.
2094      *
2095      * @throws ClassCastException {@inheritDoc}
2096      * @throws NullPointerException if the specified key is null
2097      */
2098     public Map.Entry ceilingEntry(K key) {
2099         return getNear(key, GT|EQ);
2100     }
2101 
2102     /**
2103      * @throws ClassCastException {@inheritDoc}
2104      * @throws NullPointerException if the specified key is null
2105      */
2106     public K ceilingKey(K key) {
2107         Node n = findNear(key, GT|EQ);
2108         return (n == null) ? null : n.key;
2109     }
2110 
2111     /**
2112      * Returns a key-value mapping associated with the least key
2113      * strictly greater than the given key, or null if there
2114      * is no such key. The returned entry does not support
2115      * the Entry.setValue method.
2116      *
2117      * @param key the key
2118      * @throws ClassCastException {@inheritDoc}
2119      * @throws NullPointerException if the specified key is null
2120      */
2121     public Map.Entry higherEntry(K key) {
2122         return getNear(key, GT);
2123     }
2124 
2125     /**
2126      * @param key the key
2127      * @throws ClassCastException {@inheritDoc}
2128      * @throws NullPointerException if the specified key is null
2129      */
2130     public K higherKey(K key) {
2131         Node n = findNear(key, GT);
2132         return (n == null) ? null : n.key;
2133     }
2134 
2135     /**
2136      * Returns a key-value mapping associated with the least
2137      * key in this map, or null if the map is empty.
2138      * The returned entry does not support
2139      * the Entry.setValue method.
2140      */
2141     public Map.Entry firstEntry() {
2142         for (;;) {
2143             Node n = findFirst();
2144             if (n == null)
2145                 return null;
2146             AbstractMap.SimpleImmutableEntry e = n.createSnapshot();
2147             if (e != null)
2148                 return e;
2149         }
2150     }
2151 
2152     /**
2153      * Returns a key-value mapping associated with the greatest
2154      * key in this map, or null if the map is empty.
2155      * The returned entry does not support
2156      * the Entry.setValue method.
2157      */
2158     public Map.Entry lastEntry() {
2159         for (;;) {
2160             Node n = findLast();
2161             if (n == null)
2162                 return null;
2163             AbstractMap.SimpleImmutableEntry e = n.createSnapshot();
2164             if (e != null)
2165                 return e;
2166         }
2167     }
2168 
2169     /**
2170      * Removes and returns a key-value mapping associated with
2171      * the least key in this map, or null if the map is empty.
2172      * The returned entry does not support
2173      * the Entry.setValue method.
2174      */
2175     public Map.Entry pollFirstEntry() {
2176         return doRemoveFirstEntry();
2177     }
2178 
2179     /**
2180      * Removes and returns a key-value mapping associated with
2181      * the greatest key in this map, or null if the map is empty.
2182      * The returned entry does not support
2183      * the Entry.setValue method.
2184      */
2185     public Map.Entry pollLastEntry() {
2186         return doRemoveLastEntry();
2187     }
2188 
2189 
2190     /* ---------------- Iterators -------------- */
2191 
2192     /**
2193      * Base of iterator classes:
2194      */
2195     abstract class Iter implements Iterator {
2196         /** the last node returned by next() */
2197         Node lastReturned;
2198         /** the next node to return from next(); */
2199         Node next;
2200         /** Cache of next value field to maintain weak consistency */
2201         V nextValue;
2202 
2203         /** Initializes ascending iterator for entire range. */
2204         Iter() {
2205             for (;;) {
2206                 next = findFirst();
2207                 if (next == null)
2208                     break;
2209                 Object x = next.value;
2210                 if (x != null && x != next) {
2211                     nextValue = (V) x;
2212                     break;
2213                 }
2214             }
2215         }
2216 
2217         public final boolean hasNext() {
2218             return next != null;
2219         }
2220 
2221         /** Advances next to higher entry. */
2222         final void advance() {
2223             if (next == null)
2224                 throw new NoSuchElementException();
2225             lastReturned = next;
2226             for (;;) {
2227                 next = next.next;
2228                 if (next == null)
2229                     break;
2230                 Object x = next.value;
2231                 if (x != null && x != next) {
2232                     nextValue = (V) x;
2233                     break;
2234                 }
2235             }
2236         }
2237 
2238         public void remove() {
2239             Node l = lastReturned;
2240             if (l == null)
2241                 throw new IllegalStateException();
2242             // It would not be worth all of the overhead to directly
2243             // unlink from here. Using remove is fast enough.
2244             ConcurrentSkipListMap.this.remove(l.key);
2245             lastReturned = null;
2246         }
2247 
2248     }
2249 
2250     final class ValueIterator extends Iter {
2251         public V next() {
2252             V v = nextValue;
2253             advance();
2254             return v;
2255         }
2256     }
2257 
2258     final class KeyIterator extends Iter {
2259         public K next() {
2260             Node n = next;
2261             advance();
2262             return n.key;
2263         }
2264     }
2265 
2266     final class EntryIterator extends Iter> {
2267         public Map.Entry next() {
2268             Node n = next;
2269             V v = nextValue;
2270             advance();
2271             return new AbstractMap.SimpleImmutableEntry(n.key, v);
2272         }
2273     }
2274 
2275     // Factory methods for iterators needed by ConcurrentSkipListSet etc
2276 
2277     Iterator keyIterator() {
2278         return new KeyIterator();
2279     }
2280 
2281     Iterator valueIterator() {
2282         return new ValueIterator();
2283     }
2284 
2285     Iterator> entryIterator() {
2286         return new EntryIterator();
2287     }
2288 
2289     /* ---------------- View Classes -------------- */
2290 
2291     /*
2292      * View classes are static, delegating to a ConcurrentNavigableMap
2293      * to allow use by SubMaps, which outweighs the ugliness of
2294      * needing type-tests for Iterator methods.
2295      */
2296 
2297     static final  List toList(Collection c) {
2298         // Using size() here would be a pessimization.
2299         List list = new ArrayList();
2300         for (E e : c)
2301             list.add(e);
2302         return list;
2303     }
2304 
2305     static final class KeySet
2306             extends AbstractSet implements NavigableSet {
2307         private final ConcurrentNavigableMap m;
2308         KeySet(ConcurrentNavigableMap map) { m = map; }
2309         public int size() { return m.size(); }
2310         public boolean isEmpty() { return m.isEmpty(); }
2311         public boolean contains(Object o) { return m.containsKey(o); }
2312         public boolean remove(Object o) { return m.remove(o) != null; }
2313         public void clear() { m.clear(); }
2314         public E lower(E e) { return m.lowerKey(e); }
2315         public E floor(E e) { return m.floorKey(e); }
2316         public E ceiling(E e) { return m.ceilingKey(e); }
2317         public E higher(E e) { return m.higherKey(e); }
2318         public Comparatorsuper E> comparator() { return m.comparator(); }
2319         public E first() { return m.firstKey(); }
2320         public E last() { return m.lastKey(); }
2321         public E pollFirst() {
2322             Map.Entry e = m.pollFirstEntry();
2323             return (e == null) ? null : e.getKey();
2324         }
2325         public E pollLast() {
2326             Map.Entry e = m.pollLastEntry();
2327             return (e == null) ? null : e.getKey();
2328         }
2329         public Iterator iterator() {
2330             if (m instanceof ConcurrentSkipListMap)
2331                 return ((ConcurrentSkipListMap)m).keyIterator();
2332             else
2333                 return ((ConcurrentSkipListMap.SubMap)m).keyIterator();
2334         }
2335         public boolean equals(Object o) {
2336             if (o == this)
2337                 return true;
2338             if (!(o instanceof Set))
2339                 return false;
2340             Collection c = (Collection) o;
2341             try {
2342                 return containsAll(c) && c.containsAll(this);
2343             } catch (ClassCastException unused)   {
2344                 return false;
2345             } catch (NullPointerException unused) {
2346                 return false;
2347             }
2348         }
2349         public Object[] toArray()     { return toList(this).toArray();  }
2350         public  T[] toArray(T[] a) { return toList(this).toArray(a); }
2351         public Iterator descendingIterator() {
2352             return descendingSet().iterator();
2353         }
2354         public NavigableSet subSet(E fromElement,
2355                                       boolean fromInclusive,
2356                                       E toElement,
2357                                       boolean toInclusive) {
2358             return new KeySet(m.subMap(fromElement, fromInclusive,
2359                                           toElement,   toInclusive));
2360         }
2361         public NavigableSet headSet(E toElement, boolean inclusive) {
2362             return new KeySet(m.headMap(toElement, inclusive));
2363         }
2364         public NavigableSet tailSet(E fromElement, boolean inclusive) {
2365             return new KeySet(m.tailMap(fromElement, inclusive));
2366         }
2367         public NavigableSet subSet(E fromElement, E toElement) {
2368             return subSet(fromElement, true, toElement, false);
2369         }
2370         public NavigableSet headSet(E toElement) {
2371             return headSet(toElement, false);
2372         }
2373         public NavigableSet tailSet(E fromElement) {
2374             return tailSet(fromElement, true);
2375         }
2376         public NavigableSet descendingSet() {
2377             return new KeySet(m.descendingMap());
2378         }
2379     }
2380 
2381     static final class Values extends AbstractCollection {
2382         private final ConcurrentNavigableMap m;
2383         Values(ConcurrentNavigableMap map) {
2384             m = map;
2385         }
2386         public Iterator iterator() {
2387             if (m instanceof ConcurrentSkipListMap)
2388                 return ((ConcurrentSkipListMap)m).valueIterator();
2389             else
2390                 return ((SubMap)m).valueIterator();
2391         }
2392         public boolean isEmpty() {
2393             return m.isEmpty();
2394         }
2395         public int size() {
2396             return m.size();
2397         }
2398         public boolean contains(Object o) {
2399             return m.containsValue(o);
2400         }
2401         public void clear() {
2402             m.clear();
2403         }
2404         public Object[] toArray()     { return toList(this).toArray();  }
2405         public  T[] toArray(T[] a) { return toList(this).toArray(a); }
2406     }
2407 
2408     static final class EntrySet extends AbstractSet> {
2409         private final ConcurrentNavigableMap m;
2410         EntrySet(ConcurrentNavigableMap map) {
2411             m = map;
2412         }
2413 
2414         public Iterator> iterator() {
2415             if (m instanceof ConcurrentSkipListMap)
2416                 return ((ConcurrentSkipListMap)m).entryIterator();
2417             else
2418                 return ((SubMap)m).entryIterator();
2419         }
2420 
2421         public boolean contains(Object o) {
2422             if (!(o instanceof Map.Entry))
2423                 return false;
2424             Map.Entry e = (Map.Entry)o;
2425             V1 v = m.get(e.getKey());
2426             return v != null && v.equals(e.getValue());
2427         }
2428         public boolean remove(Object o) {
2429             if (!(o instanceof Map.Entry))
2430                 return false;
2431             Map.Entry e = (Map.Entry)o;
2432             return m.remove(e.getKey(),
2433                             e.getValue());
2434         }
2435         public boolean isEmpty() {
2436             return m.isEmpty();
2437         }
2438         public int size() {
2439             return m.size();
2440         }
2441         public void clear() {
2442             m.clear();
2443         }
2444         public boolean equals(Object o) {
2445             if (o == this)
2446                 return true;
2447             if (!(o instanceof Set))
2448                 return false;
2449             Collection c = (Collection) o;
2450             try {
2451                 return containsAll(c) && c.containsAll(this);
2452             } catch (ClassCastException unused)   {
2453                 return false;
2454             } catch (NullPointerException unused) {
2455                 return false;
2456             }
2457         }
2458         public Object[] toArray()     { return toList(this).toArray();  }
2459         public  T[] toArray(T[] a) { return toList(this).toArray(a); }
2460     }
2461 
2462     /**
2463      * Submaps returned by {@link ConcurrentSkipListMap} submap operations
2464      * represent a subrange of mappings of their underlying
2465      * maps. Instances of this class support all methods of their
2466      * underlying maps, differing in that mappings outside their range are
2467      * ignored, and attempts to add mappings outside their ranges result
2468      * in {@link IllegalArgumentException}.  Instances of this class are
2469      * constructed only using the subMap, headMap, and
2470      * tailMap methods of their underlying maps.
2471      *
2472      * @serial include
2473      */
2474     static final class SubMap extends AbstractMap
2475         implements ConcurrentNavigableMap, Cloneable,
2476                    java.io.Serializable {
2477         private static final long serialVersionUID = -7647078645895051609L;
2478 
2479         /** Underlying map */
2480         private final ConcurrentSkipListMap m;
2481         /** lower bound key, or null if from start */
2482         private final K lo;
2483         /** upper bound key, or null if to end */
2484         private final K hi;
2485         /** inclusion flag for lo */
2486         private final boolean loInclusive;
2487         /** inclusion flag for hi */
2488         private final boolean hiInclusive;
2489         /** direction */
2490         private final boolean isDescending;
2491 
2492         // Lazily initialized view holders
2493         private transient KeySet keySetView;
2494         private transient Set> entrySetView;
2495         private transient Collection valuesView;
2496 
2497         /**
2498          * Creates a new submap, initializing all fields
2499          */
2500         SubMap(ConcurrentSkipListMap map,
2501                K fromKey, boolean fromInclusive,
2502                K toKey, boolean toInclusive,
2503                boolean isDescending) {
2504             if (fromKey != null && toKey != null &&
2505                 map.compare(fromKey, toKey) > 0)
2506                 throw new IllegalArgumentException("inconsistent range");
2507             this.m = map;
2508             this.lo = fromKey;
2509             this.hi = toKey;
2510             this.loInclusive = fromInclusive;
2511             this.hiInclusive = toInclusive;
2512             this.isDescending = isDescending;
2513         }
2514 
2515         /* ----------------  Utilities -------------- */
2516 
2517         private boolean tooLow(K key) {
2518             if (lo != null) {
2519                 int c = m.compare(key, lo);
2520                 if (c < 0 || (c == 0 && !loInclusive))
2521                     return true;
2522             }
2523             return false;
2524         }
2525 
2526         private boolean tooHigh(K key) {
2527             if (hi != null) {
2528                 int c = m.compare(key, hi);
2529                 if (c > 0 || (c == 0 && !hiInclusive))
2530                     return true;
2531             }
2532             return false;
2533         }
2534 
2535         private boolean inBounds(K key) {
2536             return !tooLow(key) && !tooHigh(key);
2537         }
2538 
2539         private void checkKeyBounds(K key) throws IllegalArgumentException {
2540             if (key == null)
2541                 throw new NullPointerException();
2542             if (!inBounds(key))
2543                 throw new IllegalArgumentException("key out of range");
2544         }
2545 
2546         /**
2547          * Returns true if node key is less than upper bound of range
2548          */
2549         private boolean isBeforeEnd(ConcurrentSkipListMap.Node n) {
2550             if (n == null)
2551                 return false;
2552             if (hi == null)
2553                 return true;
2554             K k = n.key;
2555             if (k == null) // pass by markers and headers
2556                 return true;
2557             int c = m.compare(k, hi);
2558             if (c > 0 || (c == 0 && !hiInclusive))
2559                 return false;
2560             return true;
2561         }
2562 
2563         /**
2564          * Returns lowest node. This node might not be in range, so
2565          * most usages need to check bounds
2566          */
2567         private ConcurrentSkipListMap.Node loNode() {
2568             if (lo == null)
2569                 return m.findFirst();
2570             else if (loInclusive)
2571                 return m.findNear(lo, m.GT|m.EQ);
2572             else
2573                 return m.findNear(lo, m.GT);
2574         }
2575 
2576         /**
2577          * Returns highest node. This node might not be in range, so
2578          * most usages need to check bounds
2579          */
2580         private ConcurrentSkipListMap.Node hiNode() {
2581             if (hi == null)
2582                 return m.findLast();
2583             else if (hiInclusive)
2584                 return m.findNear(hi, m.LT|m.EQ);
2585             else
2586                 return m.findNear(hi, m.LT);
2587         }
2588 
2589         /**
2590          * Returns lowest absolute key (ignoring directonality)
2591          */
2592         private K lowestKey() {
2593             ConcurrentSkipListMap.Node n = loNode();
2594             if (isBeforeEnd(n))
2595                 return n.key;
2596             else
2597                 throw new NoSuchElementException();
2598         }
2599 
2600         /**
2601          * Returns highest absolute key (ignoring directonality)
2602          */
2603         private K highestKey() {
2604             ConcurrentSkipListMap.Node n = hiNode();
2605             if (n != null) {
2606                 K last = n.key;
2607                 if (inBounds(last))
2608                     return last;
2609             }
2610             throw new NoSuchElementException();
2611         }
2612 
2613         private Map.Entry lowestEntry() {
2614             for (;;) {
2615                 ConcurrentSkipListMap.Node n = loNode();
2616                 if (!isBeforeEnd(n))
2617                     return null;
2618                 Map.Entry e = n.createSnapshot();
2619                 if (e != null)
2620                     return e;
2621             }
2622         }
2623 
2624         private Map.Entry highestEntry() {
2625             for (;;) {
2626                 ConcurrentSkipListMap.Node n = hiNode();
2627                 if (n == null || !inBounds(n.key))
2628                     return null;
2629                 Map.Entry e = n.createSnapshot();
2630                 if (e != null)
2631                     return e;
2632             }
2633         }
2634 
2635         private Map.Entry removeLowest() {
2636             for (;;) {
2637                 Node n = loNode();
2638                 if (n == null)
2639                     return null;
2640                 K k = n.key;
2641                 if (!inBounds(k))
2642                     return null;
2643                 V v = m.doRemove(k, null);
2644                 if (v != null)
2645                     return new AbstractMap.SimpleImmutableEntry(k, v);
2646             }
2647         }
2648 
2649         private Map.Entry removeHighest() {
2650             for (;;) {
2651                 Node n = hiNode();
2652                 if (n == null)
2653                     return null;
2654                 K k = n.key;
2655                 if (!inBounds(k))
2656                     return null;
2657                 V v = m.doRemove(k, null);
2658                 if (v != null)
2659                     return new AbstractMap.SimpleImmutableEntry(k, v);
2660             }
2661         }
2662 
2663         /**
2664          * Submap version of ConcurrentSkipListMap.getNearEntry
2665          */
2666         private Map.Entry getNearEntry(K key, int rel) {
2667             if (isDescending) { // adjust relation for direction
2668                 if ((rel & m.LT) == 0)
2669                     rel |= m.LT;
2670                 else
2671                     rel &= ~m.LT;
2672             }
2673             if (tooLow(key))
2674                 return ((rel & m.LT) != 0) ? null : lowestEntry();
2675             if (tooHigh(key))
2676                 return ((rel & m.LT) != 0) ? highestEntry() : null;
2677             for (;;) {
2678                 Node n = m.findNear(key, rel);
2679                 if (n == null || !inBounds(n.key))
2680                     return null;
2681                 K k = n.key;
2682                 V v = n.getValidValue();
2683                 if (v != null)
2684                     return new AbstractMap.SimpleImmutableEntry(k, v);
2685             }
2686         }
2687 
2688         // Almost the same as getNearEntry, except for keys
2689         private K getNearKey(K key, int rel) {
2690             if (isDescending) { // adjust relation for direction
2691                 if ((rel & m.LT) == 0)
2692                     rel |= m.LT;
2693                 else
2694                     rel &= ~m.LT;
2695             }
2696             if (tooLow(key)) {
2697                 if ((rel & m.LT) == 0) {
2698                     ConcurrentSkipListMap.Node n = loNode();
2699                     if (isBeforeEnd(n))
2700                         return n.key;
2701                 }
2702                 return null;
2703             }
2704             if (tooHigh(key)) {
2705                 if ((rel & m.LT) != 0) {
2706                     ConcurrentSkipListMap.Node n = hiNode();
2707                     if (n != null) {
2708                         K last = n.key;
2709                         if (inBounds(last))
2710                             return last;
2711                     }
2712                 }
2713                 return null;
2714             }
2715             for (;;) {
2716                 Node n = m.findNear(key, rel);
2717                 if (n == null || !inBounds(n.key))
2718                     return null;
2719                 K k = n.key;
2720                 V v = n.getValidValue();
2721                 if (v != null)
2722                     return k;
2723             }
2724         }
2725 
2726         /* ----------------  Map API methods -------------- */
2727 
2728         public boolean containsKey(Object key) {
2729             if (key == null) throw new NullPointerException();
2730             K k = (K)key;
2731             return inBounds(k) && m.containsKey(k);
2732         }
2733 
2734         public V get(Object key) {
2735             if (key == null) throw new NullPointerException();
2736             K k = (K)key;
2737             return ((!inBounds(k)) ? null : m.get(k));
2738         }
2739 
2740         public V put(K key, V value) {
2741             checkKeyBounds(key);
2742             return m.put(key, value);
2743         }
2744 
2745         public V remove(Object key) {
2746             K k = (K)key;
2747             return (!inBounds(k)) ? null : m.remove(k);
2748         }
2749 
2750         public int size() {
2751             long count = 0;
2752             for (ConcurrentSkipListMap.Node n = loNode();
2753                  isBeforeEnd(n);
2754                  n = n.next) {
2755                 if (n.getValidValue() != null)
2756                     ++count;
2757             }
2758             return count >= Integer.MAX_VALUE ? Integer.MAX_VALUE : (int)count;
2759         }
2760 
2761         public boolean isEmpty() {
2762             return !isBeforeEnd(loNode());
2763         }
2764 
2765         public boolean containsValue(Object value) {
2766             if (value == null)
2767                 throw new NullPointerException();
2768             for (ConcurrentSkipListMap.Node n = loNode();
2769                  isBeforeEnd(n);
2770                  n = n.next) {
2771                 V v = n.getValidValue();
2772                 if (v != null && value.equals(v))
2773                     return true;
2774             }
2775             return false;
2776         }
2777 
2778         public void clear() {
2779             for (ConcurrentSkipListMap.Node n = loNode();
2780                  isBeforeEnd(n);
2781                  n = n.next) {
2782                 if (n.getValidValue() != null)
2783                     m.remove(n.key);
2784             }
2785         }
2786 
2787         /* ----------------  ConcurrentMap API methods -------------- */
2788 
2789         public V putIfAbsent(K key, V value) {
2790             checkKeyBounds(key);
2791             return m.putIfAbsent(key, value);
2792         }
2793 
2794         public boolean remove(Object key, Object value) {
2795             K k = (K)key;
2796             return inBounds(k) && m.remove(k, value);
2797         }
2798 
2799         public boolean replace(K key, V oldValue, V newValue) {
2800             checkKeyBounds(key);
2801             return m.replace(key, oldValue, newValue);
2802         }
2803 
2804         public V replace(K key, V value) {
2805             checkKeyBounds(key);
2806             return m.replace(key, value);
2807         }
2808 
2809         /* ----------------  SortedMap API methods -------------- */
2810 
2811         public Comparatorsuper K> comparator() {
2812             Comparatorsuper K> cmp = m.comparator();
2813             if (isDescending)
2814                 return Collections.reverseOrder(cmp);
2815             else
2816                 return cmp;
2817         }
2818 
2819         /**
2820          * Utility to create submaps, where given bounds override
2821          * unbounded(null) ones and/or are checked against bounded ones.
2822          */
2823         private SubMap newSubMap(K fromKey,
2824                                       boolean fromInclusive,
2825                                       K toKey,
2826                                       boolean toInclusive) {
2827             if (isDescending) { // flip senses
2828                 K tk = fromKey;
2829                 fromKey = toKey;
2830                 toKey = tk;
2831                 boolean ti = fromInclusive;
2832                 fromInclusive = toInclusive;
2833                 toInclusive = ti;
2834             }
2835             if (lo != null) {
2836                 if (fromKey == null) {
2837                     fromKey = lo;
2838                     fromInclusive = loInclusive;
2839                 }
2840                 else {
2841                     int c = m.compare(fromKey, lo);
2842                     if (c < 0 || (c == 0 && !loInclusive && fromInclusive))
2843                         throw new IllegalArgumentException("key out of range");
2844                 }
2845             }
2846             if (hi != null) {
2847                 if (toKey == null) {
2848                     toKey = hi;
2849                     toInclusive = hiInclusive;
2850                 }
2851                 else {
2852                     int c = m.compare(toKey, hi);
2853                     if (c > 0 || (c == 0 && !hiInclusive && toInclusive))
2854                         throw new IllegalArgumentException("key out of range");
2855                 }
2856             }
2857             return new SubMap(m, fromKey, fromInclusive,
2858                                    toKey, toInclusive, isDescending);
2859         }
2860 
2861         public SubMap subMap(K fromKey,
2862                                   boolean fromInclusive,
2863                                   K toKey,
2864                                   boolean toInclusive) {
2865             if (fromKey == null || toKey == null)
2866                 throw new NullPointerException();
2867             return newSubMap(fromKey, fromInclusive, toKey, toInclusive);
2868         }
2869 
2870         public SubMap headMap(K toKey,
2871                                    boolean inclusive) {
2872             if (toKey == null)
2873                 throw new NullPointerException();
2874             return newSubMap(null, false, toKey, inclusive);
2875         }
2876 
2877         public SubMap tailMap(K fromKey,
2878                                    boolean inclusive) {
2879             if (fromKey == null)
2880                 throw new NullPointerException();
2881             return newSubMap(fromKey, inclusive, null, false);
2882         }
2883 
2884         public SubMap subMap(K fromKey, K toKey) {
2885             return subMap(fromKey, true, toKey, false);
2886         }
2887 
2888         public SubMap headMap(K toKey) {
2889             return headMap(toKey, false);
2890         }
2891 
2892         public SubMap tailMap(K fromKey) {
2893             return tailMap(fromKey, true);
2894         }
2895 
2896         public SubMap descendingMap() {
2897             return new SubMap(m, lo, loInclusive,
2898                                    hi, hiInclusive, !isDescending);
2899         }
2900 
2901         /* ----------------  Relational methods -------------- */
2902 
2903         public Map.Entry ceilingEntry(K key) {
2904             return getNearEntry(key, (m.GT|m.EQ));
2905         }
2906 
2907         public K ceilingKey(K key) {
2908             return getNearKey(key, (m.GT|m.EQ));
2909         }
2910 
2911         public Map.Entry lowerEntry(K key) {
2912             return getNearEntry(key, (m.LT));
2913         }
2914 
2915         public K lowerKey(K key) {
2916             return getNearKey(key, (m.LT));
2917         }
2918 
2919         public Map.Entry floorEntry(K key) {
2920             return getNearEntry(key, (m.LT|m.EQ));
2921         }
2922 
2923         public K floorKey(K key) {
2924             return getNearKey(key, (m.LT|m.EQ));
2925         }
2926 
2927         public Map.Entry higherEntry(K key) {
2928             return getNearEntry(key, (m.GT));
2929         }
2930 
2931         public K higherKey(K key) {
2932             return getNearKey(key, (m.GT));
2933         }
2934 
2935         public K firstKey() {
2936             return isDescending ? highestKey() : lowestKey();
2937         }
2938 
2939         public K lastKey() {
2940             return isDescending ? lowestKey() : highestKey();
2941         }
2942 
2943         public Map.Entry firstEntry() {
2944             return isDescending ? highestEntry() : lowestEntry();
2945         }
2946 
2947         public Map.Entry lastEntry() {
2948             return isDescending ? lowestEntry() : highestEntry();
2949         }
2950 
2951         public Map.Entry pollFirstEntry() {
2952             return isDescending ? removeHighest() : removeLowest();
2953         }
2954 
2955         public Map.Entry pollLastEntry() {
2956             return isDescending ? removeLowest() : removeHighest();
2957         }
2958 
2959         /* ---------------- Submap Views -------------- */
2960 
2961         public NavigableSet keySet() {
2962             KeySet ks = keySetView;
2963             return (ks != null) ? ks : (keySetView = new KeySet(this));
2964         }
2965 
2966         public NavigableSet navigableKeySet() {
2967             KeySet ks = keySetView;
2968             return (ks != null) ? ks : (keySetView = new KeySet(this));
2969         }
2970 
2971         public Collection values() {
2972             Collection vs = valuesView;
2973             return (vs != null) ? vs : (valuesView = new Values(this));
2974         }
2975 
2976         public Set> entrySet() {
2977             Set> es = entrySetView;
2978             return (es != null) ? es : (entrySetView = new EntrySet(this));
2979         }
2980 
2981         public NavigableSet descendingKeySet() {
2982             return descendingMap().navigableKeySet();
2983         }
2984 
2985         Iterator keyIterator() {
2986             return new SubMapKeyIterator();
2987         }
2988 
2989         Iterator valueIterator() {
2990             return new SubMapValueIterator();
2991         }
2992 
2993         Iterator> entryIterator() {
2994             return new SubMapEntryIterator();
2995         }
2996 
2997         /**
2998          * Variant of main Iter class to traverse through submaps.
2999          */
3000         abstract class SubMapIter implements Iterator {
3001             /** the last node returned by next() */
3002             Node lastReturned;
3003             /** the next node to return from next(); */
3004             Node next;
3005             /** Cache of next value field to maintain weak consistency */
3006             V nextValue;
3007 
3008             SubMapIter() {
3009                 for (;;) {
3010                     next = isDescending ? hiNode() : loNode();
3011                     if (next == null)
3012                         break;
3013                     Object x = next.value;
3014                     if (x != null && x != next) {
3015                         if (! inBounds(next.key))
3016                             next = null;
3017                         else
3018                             nextValue = (V) x;
3019                         break;
3020                     }
3021                 }
3022             }
3023 
3024             public final boolean hasNext() {
3025                 return next != null;
3026             }
3027 
3028             final void advance() {
3029                 if (next == null)
3030                     throw new NoSuchElementException();
3031                 lastReturned = next;
3032                 if (isDescending)
3033                     descend();
3034                 else
3035                     ascend();
3036             }
3037 
3038             private void ascend() {
3039                 for (;;) {
3040                     next = next.next;
3041                     if (next == null)
3042                         break;
3043                     Object x = next.value;
3044                     if (x != null && x != next) {
3045                         if (tooHigh(next.key))
3046                             next = null;
3047                         else
3048                             nextValue = (V) x;
3049                         break;
3050                     }
3051                 }
3052             }
3053 
3054             private void descend() {
3055                 for (;;) {
3056                     next = m.findNear(lastReturned.key, LT);
3057                     if (next == null)
3058                         break;
3059                     Object x = next.value;
3060                     if (x != null && x != next) {
3061                         if (tooLow(next.key))
3062                             next = null;
3063                         else
3064                             nextValue = (V) x;
3065                         break;
3066                     }
3067                 }
3068             }
3069 
3070             public void remove() {
3071                 Node l = lastReturned;
3072                 if (l == null)
3073                     throw new IllegalStateException();
3074                 m.remove(l.key);
3075                 lastReturned = null;
3076             }
3077 
3078         }
3079 
3080         final class SubMapValueIterator extends SubMapIter {
3081             public V next() {
3082                 V v = nextValue;
3083                 advance();
3084                 return v;
3085             }
3086         }
3087 
3088         final class SubMapKeyIterator extends SubMapIter {
3089             public K next() {
3090                 Node n = next;
3091                 advance();
3092                 return n.key;
3093             }
3094         }
3095 
3096         final class SubMapEntryIterator extends SubMapIter> {
3097             public Map.Entry next() {
3098                 Node n = next;
3099                 V v = nextValue;
3100                 advance();
3101                 return new AbstractMap.SimpleImmutableEntry(n.key, v);
3102             }
3103         }
3104     }
3105 
3106     // Unsafe mechanics
3107     private static final sun.misc.Unsafe UNSAFE;
3108     private static final long headOffset;
3109     static {
3110         try {
3111             UNSAFE = sun.misc.Unsafe.getUnsafe();
3112             Class k = ConcurrentSkipListMap.class;
3113             headOffset = UNSAFE.objectFieldOffset
3114                 (k.getDeclaredField("head"));
3115         } catch (Exception e) {
3116             throw new Error(e);
3117         }
3118     }
3119 }

 

下面从ConcurrentSkipListMap的添加，删除，获取这3个方面对它进行分析。

1. 添加

下面以put(K key, V value)为例，对ConcurrentSkipListMap的添加方法进行说明。

public V put(K key, V value) {
    if (value == null)
        throw new NullPointerException();
    return doPut(key, value, false);
}

实际上，put()是通过doPut()将key-value键值对添加到ConcurrentSkipListMap中的。

doPut()的源码如下：

private V doPut(K kkey, V value, boolean onlyIfAbsent) {
    Comparablesuper K> key = comparable(kkey);
    for (;;) {
        // 找到key的前继节点
        Node b = findPredecessor(key);
        // 设置n为“key的前继节点的后继节点”，即n应该是“插入节点”的“后继节点”
        Node n = b.next;
        for (;;) {
            if (n != null) {
                Node f = n.next;
                // 如果两次获得的b.next不是相同的Node，就跳转到”外层for循环“，重新获得b和n后再遍历。
                if (n != b.next)
                    break;
                // v是“n的值”
                Object v = n.value;
                // 当n的值为null(意味着其它线程删除了n)；此时删除b的下一个节点，然后跳转到”外层for循环“，重新获得b和n后再遍历。
                if (v == null) {               // n is deleted
                    n.helpDelete(b, f);
                    break;
                }
                // 如果其它线程删除了b；则跳转到”外层for循环“，重新获得b和n后再遍历。
                if (v == n || b.value == null) // b is deleted
                    break;
                // 比较key和n.key
                int c = key.compareTo(n.key);
                if (c > 0) {
                    b = n;
                    n = f;
                    continue;
                }
                if (c == 0) {
                    if (onlyIfAbsent || n.casValue(v, value))
                        return (V)v;
                    else
                        break; // restart if lost race to replace value
                }
                // else c < 0; fall through
            }

            // 新建节点(对应是“要插入的键值对”)
            Node z = new Node(kkey, value, n);
            // 设置“b的后继节点”为z
            if (!b.casNext(n, z))
                break;         // 多线程情况下，break才可能发生(其它线程对b进行了操作)
            // 随机获取一个level
            // 然后在“第1层”到“第level层”的链表中都插入新建节点
            int level = randomLevel();
            if (level > 0)
                insertIndex(z, level);
            return null;
        }
    }
}

说明：doPut() 的作用就是将键值对添加到“跳表”中。
要想搞清doPut()，首先要弄清楚它的主干部分 —— 我们先单纯的只考虑“单线程的情况下，将key-value添加到跳表中”，即忽略“多线程相关的内容”。它的流程如下：
第1步：找到“插入位置”。
即，找到“key的前继节点(b)”和“key的后继节点(n)”；key是要插入节点的键。
第2步：新建并插入节点。
即，新建节点z(key对应的节点)，并将新节点z插入到“跳表”中(设置“b的后继节点为z”，“z的后继节点为n”)。
第3步：更新跳表。
即，随机获取一个level，然后在“跳表”的第1层～第level层之间，每一层都插入节点z；在第level层之上就不再插入节点了。若level数值大于“跳表的层次”，则新建一层。
主干部分“对应的精简后的doPut()的代码”如下(仅供参考)：

private V doPut(K kkey, V value, boolean onlyIfAbsent) {
    Comparablesuper K> key = comparable(kkey);
    for (;;) {
        // 找到key的前继节点
        Node b = findPredecessor(key);
        // 设置n为key的后继节点
        Node n = b.next;
        for (;;) {
            
            // 新建节点(对应是“要被插入的键值对”)
            Node z = new Node(kkey, value, n);
            // 设置“b的后继节点”为z
            b.casNext(n, z);

            // 随机获取一个level
            // 然后在“第1层”到“第level层”的链表中都插入新建节点
            int level = randomLevel();
            if (level > 0)
                insertIndex(z, level);
            return null;
        }
    }
}

理清主干之后，剩余的工作就相对简单了。主要是上面几步的对应算法的具体实现，以及多线程相关情况的处理！

 

2. 删除

下面以remove(Object key)为例，对ConcurrentSkipListMap的删除方法进行说明。

public V remove(Object key) {
    return doRemove(key, null);
}

实际上，remove()是通过doRemove()将ConcurrentSkipListMap中的key对应的键值对删除的。

doRemove()的源码如下：

 

final V doRemove(Object okey, Object value) {
    Comparablesuper K> key = comparable(okey);
    for (;;) {
        // 找到“key的前继节点”
        Node b = findPredecessor(key);
        // 设置n为“b的后继节点”(即若key存在于“跳表中”，n就是key对应的节点)
        Node n = b.next;
        for (;;) {
            if (n == null)
                return null;
            // f是“当前节点n的后继节点”
            Node f = n.next;
            // 如果两次读取到的“b的后继节点”不同(其它线程操作了该跳表)，则返回到“外层for循环”重新遍历。
            if (n != b.next)                    // inconsistent read
                break;
            // 如果“当前节点n的值”变为null(其它线程操作了该跳表)，则返回到“外层for循环”重新遍历。
            Object v = n.value;
            if (v == null) {                    // n is deleted
                n.helpDelete(b, f);
                break;
            }
            // 如果“前继节点b”被删除(其它线程操作了该跳表)，则返回到“外层for循环”重新遍历。
            if (v == n || b.value == null)      // b is deleted
                break;
            int c = key.compareTo(n.key);
            if (c < 0)
                return null;
            if (c > 0) {
                b = n;
                n = f;
                continue;
            }

            // 以下是c=0的情况
            if (value != null && !value.equals(v))
                return null;
            // 设置“当前节点n”的值为null
            if (!n.casValue(v, null))
                break;
            // 设置“b的后继节点”为f
            if (!n.appendMarker(f) || !b.casNext(n, f))
                findNode(key);                  // Retry via findNode
            else {
                // 清除“跳表”中每一层的key节点
                findPredecessor(key);           // Clean index
                // 如果“表头的右索引为空”，则将“跳表的层次”-1。
                if (head.right == null)
                    tryReduceLevel();
            }
            return (V)v;
        }
    }
}

说明：doRemove()的作用是删除跳表中的节点。
和doPut()一样，我们重点看doRemove()的主干部分，了解主干部分之后，其余部分就非常容易理解了。下面是“单线程的情况下，删除跳表中键值对的步骤”：
第1步：找到“被删除节点的位置”。
即，找到“key的前继节点(b)”，“key所对应的节点(n)”，“n的后继节点f”；key是要删除节点的键。
第2步：删除节点。
即，将“key所对应的节点n”从跳表中移除 -- 将“b的后继节点”设为“f”！
第3步：更新跳表。
即，遍历跳表，删除每一层的“key节点”(如果存在的话)。如果删除“key节点”之后，跳表的层次需要-1；则执行相应的操作！
主干部分“对应的精简后的doRemove()的代码”如下(仅供参考)：

 

final V doRemove(Object okey, Object value) {
    Comparablesuper K> key = comparable(okey);
    for (;;) {
        // 找到“key的前继节点”
        Node b = findPredecessor(key);
        // 设置n为“b的后继节点”(即若key存在于“跳表中”，n就是key对应的节点)
        Node n = b.next;
        for (;;) {
            // f是“当前节点n的后继节点”
            Node f = n.next;

            // 设置“当前节点n”的值为null
            n.casValue(v, null);

            // 设置“b的后继节点”为f
            b.casNext(n, f);
            // 清除“跳表”中每一层的key节点
            findPredecessor(key);
            // 如果“表头的右索引为空”，则将“跳表的层次”-1。
            if (head.right == null)
                tryReduceLevel();
            return (V)v;
        }
    }
}

 

3. 获取

下面以get(Object key)为例，对ConcurrentSkipListMap的获取方法进行说明。

public V get(Object key) {
    return doGet(key);
}

doGet的源码如下：

private V doGet(Object okey) {
    Comparablesuper K> key = comparable(okey);
    for (;;) {
        // 找到“key对应的节点”
        Node n = findNode(key);
        if (n == null)
            return null;
        Object v = n.value;
        if (v != null)
            return (V)v;
    }
}

说明：doGet()是通过findNode()找到并返回节点的。

private Node findNode(Comparablesuper K> key) {
    for (;;) {
        // 找到key的前继节点
        Node b = findPredecessor(key);
        // 设置n为“b的后继节点”(即若key存在于“跳表中”，n就是key对应的节点)
        Node n = b.next;
        for (;;) {
            // 如果“n为null”，则跳转中不存在key对应的节点，直接返回null。
            if (n == null)
                return null;
            Node f = n.next;
            // 如果两次读取到的“b的后继节点”不同(其它线程操作了该跳表)，则返回到“外层for循环”重新遍历。
            if (n != b.next)                // inconsistent read
                break;
            Object v = n.value;
            // 如果“当前节点n的值”变为null(其它线程操作了该跳表)，则返回到“外层for循环”重新遍历。
            if (v == null) {                // n is deleted
                n.helpDelete(b, f);
                break;
            }
            if (v == n || b.value == null)  // b is deleted
                break;
            // 若n是当前节点，则返回n。
            int c = key.compareTo(n.key);
            if (c == 0)
                return n;
            // 若“节点n的key”小于“key”，则说明跳表中不存在key对应的节点，返回null
            if (c < 0)
                return null;
            // 若“节点n的key”大于“key”，则更新b和n，继续查找。
            b = n;
            n = f;
        }
    }
}

说明：findNode(key)的作用是在返回跳表中key对应的节点；存在则返回节点，不存在则返回null。
先弄清函数的主干部分，即抛开“多线程相关内容”，单纯的考虑单线程情况下，从跳表获取节点的算法。
第1步：找到“被删除节点的位置”。
根据findPredecessor()定位key所在的层次以及找到key的前继节点(b)，然后找到b的后继节点n。
第2步：根据“key的前继节点(b)”和“key的前继节点的后继节点(n)”来定位“key对应的节点”。
具体是通过比较“n的键值”和“key”的大小。如果相等，则n就是所要查找的键。

 

ConcurrentSkipListMap示例

import java.util.*;
import java.util.concurrent.*;

/*
 *   ConcurrentSkipListMap是“线程安全”的哈希表，而TreeMap是非线程安全的。
 *
 *   下面是“多个线程同时操作并且遍历map”的示例
 *   (01) 当map是ConcurrentSkipListMap对象时，程序能正常运行。
 *   (02) 当map是TreeMap对象时，程序会产生ConcurrentModificationException异常。
 *
 * @author skywang
 */
public class ConcurrentSkipListMapDemo1 {

    // TODO: map是TreeMap对象时，程序会出错。
    //private static Map map = new TreeMap();
    private static Map map = new ConcurrentSkipListMap();
    public static void main(String[] args) {
    
        // 同时启动两个线程对map进行操作！
        new MyThread("a").start();
        new MyThread("b").start();
    }

    private static void printAll() {
        String key, value;
        Iterator iter = map.entrySet().iterator();
        while(iter.hasNext()) {
            Map.Entry entry = (Map.Entry)iter.next();
            key = (String)entry.getKey();
            value = (String)entry.getValue();
            System.out.print("("+key+", "+value+"), ");
        }
        System.out.println();
    }

    private static class MyThread extends Thread {
        MyThread(String name) {
            super(name);
        }
        @Override
        public void run() {
                int i = 0;
            while (i++ < 6) {
                // “线程名” + "序号"
                String val = Thread.currentThread().getName()+i;
                map.put(val, "0");
                // 通过“Iterator”遍历map。
                printAll();
            }
        }
    }
}

(某一次)运行结果：

 

(a1, 0), (a1, 0), (b1, 0), (b1, 0),

(a1, 0), (b1, 0), (b2, 0), 
(a1, 0), (a1, 0), (a2, 0), (a2, 0), (b1, 0), (b1, 0), (b2, 0), (b2, 0), (b3, 0), 
(b3, 0), (a1, 0), 
(a2, 0), (a3, 0), (a1, 0), (b1, 0), (a2, 0), (b2, 0), (a3, 0), (b3, 0), (b1, 0), (b4, 0), 
(b2, 0), (a1, 0), (b3, 0), (a2, 0), (b4, 0), 
(a3, 0), (a1, 0), (a4, 0), (a2, 0), (b1, 0), (a3, 0), (b2, 0), (a4, 0), (b3, 0), (b1, 0), (b4, 0), (b2, 0), (b5, 0), 
(b3, 0), (a1, 0), (b4, 0), (a2, 0), (b5, 0), 
(a3, 0), (a1, 0), (a4, 0), (a2, 0), (a5, 0), (a3, 0), (b1, 0), (a4, 0), (b2, 0), (a5, 0), (b3, 0), (b1, 0), (b4, 0), (b2, 0), (b5, 0), (b3, 0), (b6, 0), 
(b4, 0), (a1, 0), (b5, 0), (a2, 0), (b6, 0), 
(a3, 0), (a4, 0), (a5, 0), (a6, 0), (b1, 0), (b2, 0), (b3, 0), (b4, 0), (b5, 0), (b6, 0), 

结果说明：
示例程序中，启动两个线程(线程a和线程b)分别对ConcurrentSkipListMap进行操作。以线程a而言，它会先获取“线程名”+“序号”，然后将该字符串作为key，将“0”作为value，插入到ConcurrentSkipListMap中；接着，遍历并输出ConcurrentSkipListMap中的全部元素。 线程b的操作和线程a一样，只不过线程b的名字和线程a的名字不同。
当map是ConcurrentSkipListMap对象时，程序能正常运行。如果将map改为TreeMap时，程序会产生ConcurrentModificationException异常。

 

 

概要

本章对Java.util.concurrent包中的ConcurrentSkipListMap类进行详细的介绍。内容包括：
ConcurrentSkipListMap介绍
ConcurrentSkipListMap原理和数据结构
ConcurrentSkipListMap函数列表
ConcurrentSkipListMap源码分析(JDK1.7.0_40版本)
ConcurrentSkipListMap示例

转载请注明出处：http://www.cnblogs.com/skywang12345/p/3498556.html

 

ConcurrentSkipListMap介绍

ConcurrentSkipListMap是线程安全的有序的哈希表，适用于高并发的场景。
ConcurrentSkipListMap和TreeMap，它们虽然都是有序的哈希表。但是，第一，它们的线程安全机制不同，TreeMap是非线程安全的，而ConcurrentSkipListMap是线程安全的。第二，ConcurrentSkipListMap是通过跳表实现的，而TreeMap是通过红黑树实现的。
关于跳表(Skip List)，它是平衡树的一种替代的数据结构，但是和红黑树不相同的是，跳表对于树的平衡的实现是基于一种随机化的算法的，这样也就是说跳表的插入和删除的工作是比较简单的。

 

ConcurrentSkipListMap原理和数据结构

ConcurrentSkipListMap的数据结构，如下图所示：

说明：

先以数据“7,14,21,32,37,71,85”序列为例，来对跳表进行简单说明。

跳表分为许多层(level)，每一层都可以看作是数据的索引，这些索引的意义就是加快跳表查找数据速度。每一层的数据都是有序的，上一层数据是下一层数据的子集，并且第一层(level 1)包含了全部的数据；层次越高，跳跃性越大，包含的数据越少。
跳表包含一个表头，它查找数据时，是从上往下，从左往右进行查找。现在“需要找出值为32的节点”为例，来对比说明跳表和普遍的链表。

情况1：链表中查找“32”节点
路径如下图1-02所示：

需要4步(红色部分表示路径)。

 

情况2：跳表中查找“32”节点
路径如下图1-03所示：

忽略索引垂直线路上路径的情况下，只需要2步(红色部分表示路径)。

下面说说Java中ConcurrentSkipListMap的数据结构。
(01) ConcurrentSkipListMap继承于AbstractMap类，也就意味着它是一个哈希表。
(02) Index是ConcurrentSkipListMap的内部类，它与“跳表中的索引相对应”。HeadIndex继承于Index，ConcurrentSkipListMap中含有一个HeadIndex的对象head，head是“跳表的表头”。
(03) Index是跳表中的索引，它包含“右索引的指针(right)”，“下索引的指针(down)”和“哈希表节点node”。node是Node的对象，Node也是ConcurrentSkipListMap中的内部类。

 

ConcurrentSkipListMap函数列表

// 构造一个新的空映射，该映射按照键的自然顺序进行排序。
ConcurrentSkipListMap()
// 构造一个新的空映射，该映射按照指定的比较器进行排序。
ConcurrentSkipListMap(Comparatorsuper K> comparator)
// 构造一个新映射，该映射所包含的映射关系与给定映射包含的映射关系相同，并按照键的自然顺序进行排序。
ConcurrentSkipListMap(Mapextends K,? extends V> m)
// 构造一个新映射，该映射所包含的映射关系与指定的有序映射包含的映射关系相同，使用的顺序也相同。
ConcurrentSkipListMap(SortedMapextends V> m)

// 返回与大于等于给定键的最小键关联的键-值映射关系；如果不存在这样的条目，则返回 null。
Map.Entry ceilingEntry(K key)
// 返回大于等于给定键的最小键；如果不存在这样的键，则返回 null。
K ceilingKey(K key)
// 从此映射中移除所有映射关系。
void clear()
// 返回此 ConcurrentSkipListMap 实例的浅表副本。
ConcurrentSkipListMap clone()
// 返回对此映射中的键进行排序的比较器；如果此映射使用键的自然顺序，则返回 null。
Comparatorsuper K> comparator()
// 如果此映射包含指定键的映射关系，则返回 true。
boolean containsKey(Object key)
// 如果此映射为指定值映射一个或多个键，则返回 true。
boolean containsValue(Object value)
// 返回此映射中所包含键的逆序 NavigableSet 视图。
NavigableSet descendingKeySet()
// 返回此映射中所包含映射关系的逆序视图。
ConcurrentNavigableMap descendingMap()
// 返回此映射中所包含的映射关系的 Set 视图。
Set> entrySet()
// 比较指定对象与此映射的相等性。
boolean equals(Object o)
// 返回与此映射中的最小键关联的键-值映射关系；如果该映射为空，则返回 null。
Map.Entry firstEntry()
// 返回此映射中当前第一个（最低）键。
K firstKey()
// 返回与小于等于给定键的最大键关联的键-值映射关系；如果不存在这样的键，则返回 null。
Map.Entry floorEntry(K key)
// 返回小于等于给定键的最大键；如果不存在这样的键，则返回 null。
K floorKey(K key)
// 返回指定键所映射到的值；如果此映射不包含该键的映射关系，则返回 null。
V get(Object key)
// 返回此映射的部分视图，其键值严格小于 toKey。
ConcurrentNavigableMap headMap(K toKey)
// 返回此映射的部分视图，其键小于（或等于，如果 inclusive 为 true）toKey。
ConcurrentNavigableMap headMap(K toKey, boolean inclusive)
// 返回与严格大于给定键的最小键关联的键-值映射关系；如果不存在这样的键，则返回 null。
Map.Entry higherEntry(K key)
// 返回严格大于给定键的最小键；如果不存在这样的键，则返回 null。
K higherKey(K key)
// 如果此映射未包含键-值映射关系，则返回 true。
boolean isEmpty()
// 返回此映射中所包含键的 NavigableSet 视图。
NavigableSet keySet()
// 返回与此映射中的最大键关联的键-值映射关系；如果该映射为空，则返回 null。
Map.Entry lastEntry()
// 返回映射中当前最后一个（最高）键。
K lastKey()
// 返回与严格小于给定键的最大键关联的键-值映射关系；如果不存在这样的键，则返回 null。
Map.Entry lowerEntry(K key)
// 返回严格小于给定键的最大键；如果不存在这样的键，则返回 null。
K lowerKey(K key)
// 返回此映射中所包含键的 NavigableSet 视图。
NavigableSet navigableKeySet()
// 移除并返回与此映射中的最小键关联的键-值映射关系；如果该映射为空，则返回 null。
Map.Entry pollFirstEntry()
// 移除并返回与此映射中的最大键关联的键-值映射关系；如果该映射为空，则返回 null。
Map.Entry pollLastEntry()
// 将指定值与此映射中的指定键关联。
V put(K key, V value)
// 如果指定键已经不再与某个值相关联，则将它与给定值关联。
V putIfAbsent(K key, V value)
// 从此映射中移除指定键的映射关系（如果存在）。
V remove(Object key)
// 只有目前将键的条目映射到给定值时，才移除该键的条目。
boolean remove(Object key, Object value)
// 只有目前将键的条目映射到某一值时，才替换该键的条目。
V replace(K key, V value)
// 只有目前将键的条目映射到给定值时，才替换该键的条目。
boolean replace(K key, V oldValue, V newValue)
// 返回此映射中的键-值映射关系数。
int size()
// 返回此映射的部分视图，其键的范围从 fromKey 到 toKey。
ConcurrentNavigableMap subMap(K fromKey, boolean fromInclusive, K toKey, boolean toInclusive)
// 返回此映射的部分视图，其键值的范围从 fromKey（包括）到 toKey（不包括）。
ConcurrentNavigableMap subMap(K fromKey, K toKey)
// 返回此映射的部分视图，其键大于等于 fromKey。
ConcurrentNavigableMap tailMap(K fromKey)
// 返回此映射的部分视图，其键大于（或等于，如果 inclusive 为 true）fromKey。
ConcurrentNavigableMap tailMap(K fromKey, boolean inclusive)
// 返回此映射中所包含值的 Collection 视图。
Collection values()

 

ConcurrentSkipListMap源码分析(JDK1.7.0_40版本)

ConcurrentSkipListMap.java的完整源码如下：

   1 /*
   2  * ORACLE PROPRIETARY/CONFIDENTIAL. Use is subject to license terms.
   3  *
   4  *
   5  *
   6  *
   7  *
   8  *
   9  *
  10  *
  11  *
  12  *
  13  *
  14  *
  15  *
  16  *
  17  *
  18  *
  19  *
  20  *
  21  *
  22  *
  23  */
  24 
  25 /*
  26  *
  27  *
  28  *
  29  *
  30  *
  31  * Written by Doug Lea with assistance from members of JCP JSR-166
  32  * Expert Group and released to the public domain, as explained at
  33  * http://creativecommons.org/publicdomain/zero/1.0/
  34  */
  35 
  36 package java.util.concurrent;
  37 import java.util.*;
  38 import java.util.concurrent.atomic.*;
  39 
  40 /**
  41  * A scalable concurrent {@link ConcurrentNavigableMap} implementation.
  42  * The map is sorted according to the {@linkplain Comparable natural
  43  * ordering} of its keys, or by a {@link Comparator} provided at map
  44  * creation time, depending on which constructor is used.
  45  *
  46  * 
This class implements a concurrent variant of   47  * href="http://en.wikipedia.org/wiki/Skip_list" target="_top">SkipLists
  48  * providing expected average log(n) time cost for the
  49  * containsKey, get, put and
  50  * remove operations and their variants.  Insertion, removal,
  51  * update, and access operations safely execute concurrently by
  52  * multiple threads.  Iterators are weakly consistent, returning
  53  * elements reflecting the state of the map at some point at or since
  54  * the creation of the iterator.  They do not throw {@link
  55  * ConcurrentModificationException}, and may proceed concurrently with
  56  * other operations. Ascending key ordered views and their iterators
  57  * are faster than descending ones.
  58  *
  59  * 
All Map.Entry pairs returned by methods in this class
  60  * and its views represent snapshots of mappings at the time they were
  61  * produced. They do not support the Entry.setValue
  62  * method. (Note however that it is possible to change mappings in the
  63  * associated map using put, putIfAbsent, or
  64  * replace, depending on exactly which effect you need.)
  65  *
  66  * 
Beware that, unlike in most collections, the size
  67  * method is not a constant-time operation. Because of the
  68  * asynchronous nature of these maps, determining the current number
  69  * of elements requires a traversal of the elements, and so may report
  70  * inaccurate results if this collection is modified during traversal.
  71  * Additionally, the bulk operations putAll, equals,
  72  * toArray, containsValue, and clear are
  73  * not guaranteed to be performed atomically. For example, an
  74  * iterator operating concurrently with a putAll operation
  75  * might view only some of the added elements.
  76  *
  77  * 
This class and its views and iterators implement all of the
  78  * optional methods of the {@link Map} and {@link Iterator}
  79  * interfaces. Like most other concurrent collections, this class does
  80  * not permit the use of null keys or values because some
  81  * null return values cannot be reliably distinguished from the absence of
  82  * elements.
  83  *
  84  * 
This class is a member of the
  85  * docRoot}/../technotes/guides/collections/index.html">
  86  * Java Collections Framework.
  87  *
  88  * @author Doug Lea
  89  * @param  the type of keys maintained by this map
  90  * @param  the type of mapped values
  91  * @since 1.6
  92  */
  93 public class ConcurrentSkipListMap extends AbstractMap
  94     implements ConcurrentNavigableMap,
  95                Cloneable,
  96                java.io.Serializable {
  97     /*
  98      * This class implements a tree-like two-dimensionally linked skip
  99      * list in which the index levels are represented in separate
 100      * nodes from the base nodes holding data.  There are two reasons
 101      * for taking this approach instead of the usual array-based
 102      * structure: 1) Array based implementations seem to encounter
 103      * more complexity and overhead 2) We can use cheaper algorithms
 104      * for the heavily-traversed index lists than can be used for the
 105      * base lists.  Here's a picture of some of the basics for a
 106      * possible list with 2 levels of index:
 107      *
 108      * Head nodes          Index nodes
 109      * +-+    right        +-+                      +-+
 110      * |2|---------------->| |--------------------->| |->null
 111      * +-+                 +-+                      +-+
 112      *  | down              |                        |
 113      *  v                   v                        v
 114      * +-+            +-+  +-+       +-+            +-+       +-+
 115      * |1|----------->| |->| |------>| |----------->| |------>| |->null
 116      * +-+            +-+  +-+       +-+            +-+       +-+
 117      *  v              |    |         |              |         |
 118      * Nodes  next     v    v         v              v         v
 119      * +-+  +-+  +-+  +-+  +-+  +-+  +-+  +-+  +-+  +-+  +-+  +-+
 120      * | |->|A|->|B|->|C|->|D|->|E|->|F|->|G|->|H|->|I|->|J|->|K|->null
 121      * +-+  +-+  +-+  +-+  +-+  +-+  +-+  +-+  +-+  +-+  +-+  +-+
 122      *
 123      * The base lists use a variant of the HM linked ordered set
 124      * algorithm. See Tim Harris, "A pragmatic implementation of
 125      * non-blocking linked lists"
 126      * http://www.cl.cam.ac.uk/~tlh20/publications.html and Maged
 127      * Michael "High Performance Dynamic Lock-Free Hash Tables and
 128      * List-Based Sets"
 129      * http://www.research.ibm.com/people/m/michael/pubs.htm.  The
 130      * basic idea in these lists is to mark the "next" pointers of
 131      * deleted nodes when deleting to avoid conflicts with concurrent
 132      * insertions, and when traversing to keep track of triples
 133      * (predecessor, node, successor) in order to detect when and how
 134      * to unlink these deleted nodes.
 135      *
 136      * Rather than using mark-bits to mark list deletions (which can
 137      * be slow and space-intensive using AtomicMarkedReference), nodes
 138      * use direct CAS'able next pointers.  On deletion, instead of
 139      * marking a pointer, they splice in another node that can be
 140      * thought of as standing for a marked pointer (indicating this by
 141      * using otherwise impossible field values).  Using plain nodes
 142      * acts roughly like "boxed" implementations of marked pointers,
 143      * but uses new nodes only when nodes are deleted, not for every
 144      * link.  This requires less space and supports faster
 145      * traversal. Even if marked references were better supported by
 146      * JVMs, traversal using this technique might still be faster
 147      * because any search need only read ahead one more node than
 148      * otherwise required (to check for trailing marker) rather than
 149      * unmasking mark bits or whatever on each read.
 150      *
 151      * This approach maintains the essential property needed in the HM
 152      * algorithm of changing the next-pointer of a deleted node so
 153      * that any other CAS of it will fail, but implements the idea by
 154      * changing the pointer to point to a different node, not by
 155      * marking it.  While it would be possible to further squeeze
 156      * space by defining marker nodes not to have key/value fields, it
 157      * isn't worth the extra type-testing overhead.  The deletion
 158      * markers are rarely encountered during traversal and are
 159      * normally quickly garbage collected. (Note that this technique
 160      * would not work well in systems without garbage collection.)
 161      *
 162      * In addition to using deletion markers, the lists also use
 163      * nullness of value fields to indicate deletion, in a style
 164      * similar to typical lazy-deletion schemes.  If a node's value is
 165      * null, then it is considered logically deleted and ignored even
 166      * though it is still reachable. This maintains proper control of
 167      * concurrent replace vs delete operations -- an attempted replace
 168      * must fail if a delete beat it by nulling field, and a delete
 169      * must return the last non-null value held in the field. (Note:
 170      * Null, rather than some special marker, is used for value fields
 171      * here because it just so happens to mesh with the Map API
 172      * requirement that method get returns null if there is no
 173      * mapping, which allows nodes to remain concurrently readable
 174      * even when deleted. Using any other marker value here would be
 175      * messy at best.)
 176      *
 177      * Here's the sequence of events for a deletion of node n with
 178      * predecessor b and successor f, initially:
 179      *
 180      *        +------+       +------+      +------+
 181      *   ...  |   b  |------>|   n  |----->|   f  | ...
 182      *        +------+       +------+      +------+
 183      *
 184      * 1. CAS n's value field from non-null to null.
 185      *    From this point on, no public operations encountering
 186      *    the node consider this mapping to exist. However, other
 187      *    ongoing insertions and deletions might still modify
 188      *    n's next pointer.
 189      *
 190      * 2. CAS n's next pointer to point to a new marker node.
 191      *    From this point on, no other nodes can be appended to n.
 192      *    which avoids deletion errors in CAS-based linked lists.
 193      *
 194      *        +------+       +------+      +------+       +------+
 195      *   ...  |   b  |------>|   n  |----->|marker|------>|   f  | ...
 196      *        +------+       +------+      +------+       +------+
 197      *
 198      * 3. CAS b's next pointer over both n and its marker.
 199      *    From this point on, no new traversals will encounter n,
 200      *    and it can eventually be GCed.
 201      *        +------+                                    +------+
 202      *   ...  |   b  |----------------------------------->|   f  | ...
 203      *        +------+                                    +------+
 204      *
 205      * A failure at step 1 leads to simple retry due to a lost race
 206      * with another operation. Steps 2-3 can fail because some other
 207      * thread noticed during a traversal a node with null value and
 208      * helped out by marking and/or unlinking.  This helping-out
 209      * ensures that no thread can become stuck waiting for progress of
 210      * the deleting thread.  The use of marker nodes slightly
 211      * complicates helping-out code because traversals must track
 212      * consistent reads of up to four nodes (b, n, marker, f), not
 213      * just (b, n, f), although the next field of a marker is
 214      * immutable, and once a next field is CAS'ed to point to a
 215      * marker, it never again changes, so this requires less care.
 216      *
 217      * Skip lists add indexing to this scheme, so that the base-level
 218      * traversals start close to the locations being found, inserted
 219      * or deleted -- usually base level traversals only traverse a few
 220      * nodes. This doesn't change the basic algorithm except for the
 221      * need to make sure base traversals start at predecessors (here,
 222      * b) that are not (structurally) deleted, otherwise retrying
 223      * after processing the deletion.
 224      *
 225      * Index levels are maintained as lists with volatile next fields,
 226      * using CAS to link and unlink.  Races are allowed in index-list
 227      * operations that can (rarely) fail to link in a new index node
 228      * or delete one. (We can't do this of course for data nodes.)
 229      * However, even when this happens, the index lists remain sorted,
 230      * so correctly serve as indices.  This can impact performance,
 231      * but since skip lists are probabilistic anyway, the net result
 232      * is that under contention, the effective "p" value may be lower
 233      * than its nominal value. And race windows are kept small enough
 234      * that in practice these failures are rare, even under a lot of
 235      * contention.
 236      *
 237      * The fact that retries (for both base and index lists) are
 238      * relatively cheap due to indexing allows some minor
 239      * simplifications of retry logic. Traversal restarts are
 240      * performed after most "helping-out" CASes. This isn't always
 241      * strictly necessary, but the implicit backoffs tend to help
 242      * reduce other downstream failed CAS's enough to outweigh restart
 243      * cost.  This worsens the worst case, but seems to improve even
 244      * highly contended cases.
 245      *
 246      * Unlike most skip-list implementations, index insertion and
 247      * deletion here require a separate traversal pass occuring after
 248      * the base-level action, to add or remove index nodes.  This adds
 249      * to single-threaded overhead, but improves contended
 250      * multithreaded performance by narrowing interference windows,
 251      * and allows deletion to ensure that all index nodes will be made
 252      * unreachable upon return from a public remove operation, thus
 253      * avoiding unwanted garbage retention. This is more important
 254      * here than in some other data structures because we cannot null
 255      * out node fields referencing user keys since they might still be
 256      * read by other ongoing traversals.
 257      *
 258      * Indexing uses skip list parameters that maintain good search
 259      * performance while using sparser-than-usual indices: The
 260      * hardwired parameters k=1, p=0.5 (see method randomLevel) mean
 261      * that about one-quarter of the nodes have indices. Of those that
 262      * do, half have one level, a quarter have two, and so on (see
 263      * Pugh's Skip List Cookbook, sec 3.4).  The expected total space
 264      * requirement for a map is slightly less than for the current
 265      * implementation of java.util.TreeMap.
 266      *
 267      * Changing the level of the index (i.e, the height of the
 268      * tree-like structure) also uses CAS. The head index has initial
 269      * level/height of one. Creation of an index with height greater
 270      * than the current level adds a level to the head index by
 271      * CAS'ing on a new top-most head. To maintain good performance
 272      * after a lot of removals, deletion methods heuristically try to
 273      * reduce the height if the topmost levels appear to be empty.
 274      * This may encounter races in which it possible (but rare) to
 275      * reduce and "lose" a level just as it is about to contain an
 276      * index (that will then never be encountered). This does no
 277      * structural harm, and in practice appears to be a better option
 278      * than allowing unrestrained growth of levels.
 279      *
 280      * The code for all this is more verbose than you'd like. Most
 281      * operations entail locating an element (or position to insert an
 282      * element). The code to do this can't be nicely factored out
 283      * because subsequent uses require a snapshot of predecessor
 284      * and/or successor and/or value fields which can't be returned
 285      * all at once, at least not without creating yet another object
 286      * to hold them -- creating such little objects is an especially
 287      * bad idea for basic internal search operations because it adds
 288      * to GC overhead.  (This is one of the few times I've wished Java
 289      * had macros.) Instead, some traversal code is interleaved within
 290      * insertion and removal operations.  The control logic to handle
 291      * all the retry conditions is sometimes twisty. Most search is
 292      * broken into 2 parts. findPredecessor() searches index nodes
 293      * only, returning a base-level predecessor of the key. findNode()
 294      * finishes out the base-level search. Even with this factoring,
 295      * there is a fair amount of near-duplication of code to handle
 296      * variants.
 297      *
 298      * For explanation of algorithms sharing at least a couple of
 299      * features with this one, see Mikhail Fomitchev's thesis
 300      * (http://www.cs.yorku.ca/~mikhail/), Keir Fraser's thesis
 301      * (http://www.cl.cam.ac.uk/users/kaf24/), and Hakan Sundell's
 302      * thesis (http://www.cs.chalmers.se/~phs/).
 303      *
 304      * Given the use of tree-like index nodes, you might wonder why
 305      * this doesn't use some kind of search tree instead, which would
 306      * support somewhat faster search operations. The reason is that
 307      * there are no known efficient lock-free insertion and deletion
 308      * algorithms for search trees. The immutability of the "down"
 309      * links of index nodes (as opposed to mutable "left" fields in
 310      * true trees) makes this tractable using only CAS operations.
 311      *
 312      * Notation guide for local variables
 313      * Node:         b, n, f    for  predecessor, node, successor
 314      * Index:        q, r, d    for index node, right, down.
 315      *               t          for another index node
 316      * Head:         h
 317      * Levels:       j
 318      * Keys:         k, key
 319      * Values:       v, value
 320      * Comparisons:  c
 321      */
 322 
 323     private static final long serialVersionUID = -8627078645895051609L;
 324 
 325     /**
 326      * Generates the initial random seed for the cheaper per-instance
 327      * random number generators used in randomLevel.
 328      */
 329     private static final Random seedGenerator = new Random();
 330 
 331     /**
 332      * Special value used to identify base-level header
 333      */
 334     private static final Object BASE_HEADER = new Object();
 335 
 336     /**
 337      * The topmost head index of the skiplist.
 338      */
 339     private transient volatile HeadIndex head;
 340 
 341     /**
 342      * The comparator used to maintain order in this map, or null
 343      * if using natural ordering.
 344      * @serial
 345      */
 346     private final Comparatorsuper K> comparator;
 347 
 348     /**
 349      * Seed for simple random number generator.  Not volatile since it
 350      * doesn't matter too much if different threads don't see updates.
 351      */
 352     private transient int randomSeed;
 353 
 354     /** Lazily initialized key set */
 355     private transient KeySet keySet;
 356     /** Lazily initialized entry set */
 357     private transient EntrySet entrySet;
 358     /** Lazily initialized values collection */
 359     private transient Values values;
 360     /** Lazily initialized descending key set */
 361     private transient ConcurrentNavigableMap descendingMap;
 362 
 363     /**
 364      * Initializes or resets state. Needed by constructors, clone,
 365      * clear, readObject. and ConcurrentSkipListSet.clone.
 366      * (Note that comparator must be separately initialized.)
 367      */
 368     final void initialize() {
 369         keySet = null;
 370         entrySet = null;
 371         values = null;
 372         descendingMap = null;
 373         randomSeed = seedGenerator.nextInt() | 0x0100; // ensure nonzero
 374         head = new HeadIndex(new Node(null, BASE_HEADER, null),
 375                                   null, null, 1);
 376     }
 377 
 378     /**
 379      * compareAndSet head node
 380      */
 381     private boolean casHead(HeadIndex cmp, HeadIndex val) {
 382         return UNSAFE.compareAndSwapObject(this, headOffset, cmp, val);
 383     }
 384 
 385     /* ---------------- Nodes -------------- */
 386 
 387     /**
 388      * Nodes hold keys and values, and are singly linked in sorted
 389      * order, possibly with some intervening marker nodes. The list is
 390      * headed by a dummy node accessible as head.node. The value field
 391      * is declared only as Object because it takes special non-V
 392      * values for marker and header nodes.
 393      */
 394     static final class Node {
 395         final K key;
 396         volatile Object value;
 397         volatile Node next;
 398 
 399         /**
 400          * Creates a new regular node.
 401          */
 402         Node(K key, Object value, Node next) {
 403             this.key = key;
 404             this.value = value;
 405             this.next = next;
 406         }
 407 
 408         /**
 409          * Creates a new marker node. A marker is distinguished by
 410          * having its value field point to itself.  Marker nodes also
 411          * have null keys, a fact that is exploited in a few places,
 412          * but this doesn't distinguish markers from the base-level
 413          * header node (head.node), which also has a null key.
 414          */
 415         Node(Node next) {
 416             this.key = null;
 417             this.value = this;
 418             this.next = next;
 419         }
 420 
 421         /**
 422          * compareAndSet value field
 423          */
 424         boolean casValue(Object cmp, Object val) {
 425             return UNSAFE.compareAndSwapObject(this, valueOffset, cmp, val);
 426         }
 427 
 428         /**
 429          * compareAndSet next field
 430          */
 431         boolean casNext(Node cmp, Node val) {
 432             return UNSAFE.compareAndSwapObject(this, nextOffset, cmp, val);
 433         }
 434 
 435         /**
 436          * Returns true if this node is a marker. This method isn't
 437          * actually called in any current code checking for markers
 438          * because callers will have already read value field and need
 439          * to use that read (not another done here) and so directly
 440          * test if value points to node.
 441          * @param n a possibly null reference to a node
 442          * @return true if this node is a marker node
 443          */
 444         boolean isMarker() {
 445             return value == this;
 446         }
 447 
 448         /**
 449          * Returns true if this node is the header of base-level list.
 450          * @return true if this node is header node
 451          */
 452         boolean isBaseHeader() {
 453             return value == BASE_HEADER;
 454         }
 455 
 456         /**
 457          * Tries to append a deletion marker to this node.
 458          * @param f the assumed current successor of this node
 459          * @return true if successful
 460          */
 461         boolean appendMarker(Node f) {
 462             return casNext(f, new Node(f));
 463         }
 464 
 465         /**
 466          * Helps out a deletion by appending marker or unlinking from
 467          * predecessor. This is called during traversals when value
 468          * field seen to be null.
 469          * @param b predecessor
 470          * @param f successor
 471          */
 472         void helpDelete(Node b, Node f) {
 473             /*
 474              * Rechecking links and then doing only one of the
 475              * help-out stages per call tends to minimize CAS
 476              * interference among helping threads.
 477              */
 478             if (f == next && this == b.next) {
 479                 if (f == null || f.value != f) // not already marked
 480                     appendMarker(f);
 481                 else
 482                     b.casNext(this, f.next);
 483             }
 484         }
 485 
 486         /**
 487          * Returns value if this node contains a valid key-value pair,
 488          * else null.
 489          * @return this node's value if it isn't a marker or header or
 490          * is deleted, else null.
 491          */
 492         V getValidValue() {
 493             Object v = value;
 494             if (v == this || v == BASE_HEADER)
 495                 return null;
 496             return (V)v;
 497         }
 498 
 499         /**
 500          * Creates and returns a new SimpleImmutableEntry holding current
 501          * mapping if this node holds a valid value, else null.
 502          * @return new entry or null
 503          */
 504         AbstractMap.SimpleImmutableEntry createSnapshot() {
 505             V v = getValidValue();
 506             if (v == null)
 507                 return null;
 508             return new AbstractMap.SimpleImmutableEntry(key, v);
 509         }
 510 
 511         // UNSAFE mechanics
 512 
 513         private static final sun.misc.Unsafe UNSAFE;
 514         private static final long valueOffset;
 515         private static final long nextOffset;
 516 
 517         static {
 518             try {
 519                 UNSAFE = sun.misc.Unsafe.getUnsafe();
 520                 Class k = Node.class;
 521                 valueOffset = UNSAFE.objectFieldOffset
 522                     (k.getDeclaredField("value"));
 523                 nextOffset = UNSAFE.objectFieldOffset
 524                     (k.getDeclaredField("next"));
 525             } catch (Exception e) {
 526                 throw new Error(e);
 527             }
 528         }
 529     }
 530 
 531     /* ---------------- Indexing -------------- */
 532 
 533     /**
 534      * Index nodes represent the levels of the skip list.  Note that
 535      * even though both Nodes and Indexes have forward-pointing
 536      * fields, they have different types and are handled in different
 537      * ways, that can't nicely be captured by placing field in a
 538      * shared abstract class.
 539      */
 540     static class Index {
 541         final Node node;
 542         final Index down;
 543         volatile Index right;
 544 
 545         /**
 546          * Creates index node with given values.
 547          */
 548         Index(Node node, Index down, Index right) {
 549             this.node = node;
 550             this.down = down;
 551             this.right = right;
 552         }
 553 
 554         /**
 555          * compareAndSet right field
 556          */
 557         final boolean casRight(Index cmp, Index val) {
 558             return UNSAFE.compareAndSwapObject(this, rightOffset, cmp, val);
 559         }
 560 
 561         /**
 562          * Returns true if the node this indexes has been deleted.
 563          * @return true if indexed node is known to be deleted
 564          */
 565         final boolean indexesDeletedNode() {
 566             return node.value == null;
 567         }
 568 
 569         /**
 570          * Tries to CAS newSucc as successor.  To minimize races with
 571          * unlink that may lose this index node, if the node being
 572          * indexed is known to be deleted, it doesn't try to link in.
 573          * @param succ the expected current successor
 574          * @param newSucc the new successor
 575          * @return true if successful
 576          */
 577         final boolean link(Index succ, Index newSucc) {
 578             Node n = node;
 579             newSucc.right = succ;
 580             return n.value != null && casRight(succ, newSucc);
 581         }
 582 
 583         /**
 584          * Tries to CAS right field to skip over apparent successor
 585          * succ.  Fails (forcing a retraversal by caller) if this node
 586          * is known to be deleted.
 587          * @param succ the expected current successor
 588          * @return true if successful
 589          */
 590         final boolean unlink(Index succ) {
 591             return !indexesDeletedNode() && casRight(succ, succ.right);
 592         }
 593 
 594         // Unsafe mechanics
 595         private static final sun.misc.Unsafe UNSAFE;
 596         private static final long rightOffset;
 597         static {
 598             try {
 599                 UNSAFE = sun.misc.Unsafe.getUnsafe();
 600                 Class k = Index.class;
 601                 rightOffset = UNSAFE.objectFieldOffset
 602                     (k.getDeclaredField("right"));
 603             } catch (Exception e) {
 604                 throw new Error(e);
 605             }
 606         }
 607     }
 608 
 609     /* ---------------- Head nodes -------------- */
 610 
 611     /**
 612      * Nodes heading each level keep track of their level.
 613      */
 614     static final class HeadIndex extends Index {
 615         final int level;
 616         HeadIndex(Node node, Index down, Index right, int level) {
 617             super(node, down, right);
 618             this.level = level;
 619         }
 620     }
 621 
 622     /* ---------------- Comparison utilities -------------- */
 623 
 624     /**
 625      * Represents a key with a comparator as a Comparable.
 626      *
 627      * Because most sorted collections seem to use natural ordering on
 628      * Comparables (Strings, Integers, etc), most internal methods are
 629      * geared to use them. This is generally faster than checking
 630      * per-comparison whether to use comparator or comparable because
 631      * it doesn't require a (Comparable) cast for each comparison.
 632      * (Optimizers can only sometimes remove such redundant checks
 633      * themselves.) When Comparators are used,
 634      * ComparableUsingComparators are created so that they act in the
 635      * same way as natural orderings. This penalizes use of
 636      * Comparators vs Comparables, which seems like the right
 637      * tradeoff.
 638      */
 639     static final class ComparableUsingComparator implements Comparable {
 640         final K actualKey;
 641         final Comparatorsuper K> cmp;
 642         ComparableUsingComparator(K key, Comparatorsuper K> cmp) {
 643             this.actualKey = key;
 644             this.cmp = cmp;
 645         }
 646         public int compareTo(K k2) {
 647             return cmp.compare(actualKey, k2);
 648         }
 649     }
 650 
 651     /**
 652      * If using comparator, return a ComparableUsingComparator, else
 653      * cast key as Comparable, which may cause ClassCastException,
 654      * which is propagated back to caller.
 655      */
 656     private Comparablesuper K> comparable(Object key)
 657             throws ClassCastException {
 658         if (key == null)
 659             throw new NullPointerException();
 660         if (comparator != null)
 661             return new ComparableUsingComparator((K)key, comparator);
 662         else
 663             return (Comparablesuper K>)key;
 664     }
 665 
 666     /**
 667      * Compares using comparator or natural ordering. Used when the
 668      * ComparableUsingComparator approach doesn't apply.
 669      */
 670     int compare(K k1, K k2) throws ClassCastException {
 671         Comparatorsuper K> cmp = comparator;
 672         if (cmp != null)
 673             return cmp.compare(k1, k2);
 674         else
 675             return ((Comparablesuper K>)k1).compareTo(k2);
 676     }
 677 
 678     /**
 679      * Returns true if given key greater than or equal to least and
 680      * strictly less than fence, bypassing either test if least or
 681      * fence are null. Needed mainly in submap operations.
 682      */
 683     boolean inHalfOpenRange(K key, K least, K fence) {
 684         if (key == null)
 685             throw new NullPointerException();
 686         return ((least == null || compare(key, least) >= 0) &&
 687                 (fence == null || compare(key, fence) <  0));
 688     }
 689 
 690     /**
 691      * Returns true if given key greater than or equal to least and less
 692      * or equal to fence. Needed mainly in submap operations.
 693      */
 694     boolean inOpenRange(K key, K least, K fence) {
 695         if (key == null)
 696             throw new NullPointerException();
 697         return ((least == null || compare(key, least) >= 0) &&
 698                 (fence == null || compare(key, fence) <= 0));
 699     }
 700 
 701     /* ---------------- Traversal -------------- */
 702 
 703     /**
 704      * Returns a base-level node with key strictly less than given key,
 705      * or the base-level header if there is no such node.  Also
 706      * unlinks indexes to deleted nodes found along the way.  Callers
 707      * rely on this side-effect of clearing indices to deleted nodes.
 708      * @param key the key
 709      * @return a predecessor of key
 710      */
 711     private Node findPredecessor(Comparablesuper K> key) {
 712         if (key == null)
 713             throw new NullPointerException(); // don't postpone errors
 714         for (;;) {
 715             Index q = head;
 716             Index r = q.right;
 717             for (;;) {
 718                 if (r != null) {
 719                     Node n = r.node;
 720                     K k = n.key;
 721                     if (n.value == null) {
 722                         if (!q.unlink(r))
 723                             break;           // restart
 724                         r = q.right;         // reread r
 725                         continue;
 726                     }
 727                     if (key.compareTo(k) > 0) {
 728                         q = r;
 729                         r = r.right;
 730                         continue;
 731                     }
 732                 }
 733                 Index d = q.down;
 734                 if (d != null) {
 735                     q = d;
 736                     r = d.right;
 737                 } else
 738                     return q.node;
 739             }
 740         }
 741     }
 742 
 743     /**
 744      * Returns node holding key or null if no such, clearing out any
 745      * deleted nodes seen along the way.  Repeatedly traverses at
 746      * base-level looking for key starting at predecessor returned
 747      * from findPredecessor, processing base-level deletions as
 748      * encountered. Some callers rely on this side-effect of clearing
 749      * deleted nodes.
 750      *
 751      * Restarts occur, at traversal step centered on node n, if:
 752      *
 753      *   (1) After reading n's next field, n is no longer assumed
 754      *       predecessor b's current successor, which means that
 755      *       we don't have a consistent 3-node snapshot and so cannot
 756      *       unlink any subsequent deleted nodes encountered.
 757      *
 758      *   (2) n's value field is null, indicating n is deleted, in
 759      *       which case we help out an ongoing structural deletion
 760      *       before retrying.  Even though there are cases where such
 761      *       unlinking doesn't require restart, they aren't sorted out
 762      *       here because doing so would not usually outweigh cost of
 763      *       restarting.
 764      *
 765      *   (3) n is a marker or n's predecessor's value field is null,
 766      *       indicating (among other possibilities) that
 767      *       findPredecessor returned a deleted node. We can't unlink
 768      *       the node because we don't know its predecessor, so rely
 769      *       on another call to findPredecessor to notice and return
 770      *       some earlier predecessor, which it will do. This check is
 771      *       only strictly needed at beginning of loop, (and the
 772      *       b.value check isn't strictly needed at all) but is done
 773      *       each iteration to help avoid contention with other
 774      *       threads by callers that will fail to be able to change
 775      *       links, and so will retry anyway.
 776      *
 777      * The traversal loops in doPut, doRemove, and findNear all
 778      * include the same three kinds of checks. And specialized
 779      * versions appear in findFirst, and findLast and their
 780      * variants. They can't easily share code because each uses the
 781      * reads of fields held in locals occurring in the orders they
 782      * were performed.
 783      *
 784      * @param key the key
 785      * @return node holding key, or null if no such
 786      */
 787     private Node findNode(Comparablesuper K> key) {
 788         for (;;) {
 789             Node b = findPredecessor(key);
 790             Node n = b.next;
 791             for (;;) {
 792                 if (n == null)
 793                     return null;
 794                 Node f = n.next;
 795                 if (n != b.next)                // inconsistent read
 796                     break;
 797                 Object v = n.value;
 798                 if (v == null) {                // n is deleted
 799                     n.helpDelete(b, f);
 800                     break;
 801                 }
 802                 if (v == n || b.value == null)  // b is deleted
 803                     break;
 804                 int c = key.compareTo(n.key);
 805                 if (c == 0)
 806                     return n;
 807                 if (c < 0)
 808                     return null;
 809                 b = n;
 810                 n = f;
 811             }
 812         }
 813     }
 814 
 815     /**
 816      * Gets value for key using findNode.
 817      * @param okey the key
 818      * @return the value, or null if absent
 819      */
 820     private V doGet(Object okey) {
 821         Comparablesuper K> key = comparable(okey);
 822         /*
 823          * Loop needed here and elsewhere in case value field goes
 824          * null just as it is about to be returned, in which case we
 825          * lost a race with a deletion, so must retry.
 826          */
 827         for (;;) {
 828             Node n = findNode(key);
 829             if (n == null)
 830                 return null;
 831             Object v = n.value;
 832             if (v != null)
 833                 return (V)v;
 834         }
 835     }
 836 
 837     /* ---------------- Insertion -------------- */
 838 
 839     /**
 840      * Main insertion method.  Adds element if not present, or
 841      * replaces value if present and onlyIfAbsent is false.
 842      * @param kkey the key
 843      * @param value  the value that must be associated with key
 844      * @param onlyIfAbsent if should not insert if already present
 845      * @return the old value, or null if newly inserted
 846      */
 847     private V doPut(K kkey, V value, boolean onlyIfAbsent) {
 848         Comparablesuper K> key = comparable(kkey);
 849         for (;;) {
 850             Node b = findPredecessor(key);
 851             Node n = b.next;
 852             for (;;) {
 853                 if (n != null) {
 854                     Node f = n.next;
 855                     if (n != b.next)               // inconsistent read
 856                         break;
 857                     Object v = n.value;
 858                     if (v == null) {               // n is deleted
 859                         n.helpDelete(b, f);
 860                         break;
 861                     }
 862                     if (v == n || b.value == null) // b is deleted
 863                         break;
 864                     int c = key.compareTo(n.key);
 865                     if (c > 0) {
 866                         b = n;
 867                         n = f;
 868                         continue;
 869                     }
 870                     if (c == 0) {
 871                         if (onlyIfAbsent || n.casValue(v, value))
 872                             return (V)v;
 873                         else
 874                             break; // restart if lost race to replace value
 875                     }
 876                     // else c < 0; fall through
 877                 }
 878 
 879                 Node z = new Node(kkey, value, n);
 880                 if (!b.casNext(n, z))
 881                     break;         // restart if lost race to append to b
 882                 int level = randomLevel();
 883                 if (level > 0)
 884                     insertIndex(z, level);
 885                 return null;
 886             }
 887         }
 888     }
 889 
 890     /**
 891      * Returns a random level for inserting a new node.
 892      * Hardwired to k=1, p=0.5, max 31 (see above and
 893      * Pugh's "Skip List Cookbook", sec 3.4).
 894      *
 895      * This uses the simplest of the generators described in George
 896      * Marsaglia's "Xorshift RNGs" paper.  This is not a high-quality
 897      * generator but is acceptable here.
 898      */
 899     private int randomLevel() {
 900         int x = randomSeed;
 901         x ^= x << 13;
 902         x ^= x >>> 17;
 903         randomSeed = x ^= x << 5;
 904         if ((x & 0x80000001) != 0) // test highest and lowest bits
 905             return 0;
 906         int level = 1;
 907         while (((x >>>= 1) & 1) != 0) ++level;
 908         return level;
 909     }
 910 
 911     /**
 912      * Creates and adds index nodes for the given node.
 913      * @param z the node
 914      * @param level the level of the index
 915      */
 916     private void insertIndex(Node z, int level) {
 917         HeadIndex h = head;
 918         int max = h.level;
 919 
 920         if (level <= max) {
 921             Index idx = null;
 922             for (int i = 1; i <= level; ++i)
 923                 idx = new Index(z, idx, null);
 924             addIndex(idx, h, level);
 925 
 926         } else { // Add a new level
 927             /*
 928              * To reduce interference by other threads checking for
 929              * empty levels in tryReduceLevel, new levels are added
 930              * with initialized right pointers. Which in turn requires
 931              * keeping levels in an array to access them while
 932              * creating new head index nodes from the opposite
 933              * direction.
 934              */
 935             level = max + 1;
 936             Index[] idxs = (Index[])new Index[level+1];
 937             Index idx = null;
 938             for (int i = 1; i <= level; ++i)
 939                 idxs[i] = idx = new Index(z, idx, null);
 940 
 941             HeadIndex oldh;
 942             int k;
 943             for (;;) {
 944                 oldh = head;
 945                 int oldLevel = oldh.level;
 946                 if (level <= oldLevel) { // lost race to add level
 947                     k = level;
 948                     break;
 949                 }
 950                 HeadIndex newh = oldh;
 951                 Node oldbase = oldh.node;
 952                 for (int j = oldLevel+1; j <= level; ++j)
 953                     newh = new HeadIndex(oldbase, newh, idxs[j], j);
 954                 if (casHead(oldh, newh)) {
 955                     k = oldLevel;
 956                     break