ViewVC Help

View File | Revision Log | Show Annotations | Download File | Root Listing

root/jsr166/jsr166/src/jsr166e/ConcurrentHashMapV8.java

Comparing jsr166/src/jsr166e/ConcurrentHashMapV8.java (file contents):
Revision 1.23 by jsr166, Sun Sep 11 04:25:00 2011 UTC vs.
Revision 1.43 by jsr166, Wed Jul 4 20:21:02 2012 UTC

#	Line 6 | Line 6
6
7		package jsr166e;
8		import jsr166e.LongAdder;
9	+	import java.util.Arrays;
10		import java.util.Map;
11		import java.util.Set;
12		import java.util.Collection;
#	Line 19 | Line 20 | import java.util.Enumeration;
20		import java.util.ConcurrentModificationException;
21		import java.util.NoSuchElementException;
22		import java.util.concurrent.ConcurrentMap;
23	+	import java.util.concurrent.ThreadLocalRandom;
24	+	import java.util.concurrent.locks.LockSupport;
25	+	import java.util.concurrent.locks.AbstractQueuedSynchronizer;
26		import java.io.Serializable;
27
28		/**
#	Line 54 | Line 58 | import java.io.Serializable;
58		* <p> The table is dynamically expanded when there are too many
59		* collisions (i.e., keys that have distinct hash codes but fall into
60		* the same slot modulo the table size), with the expected average
61	<	* effect of maintaining roughly two bins per mapping. There may be
62	<	* much variance around this average as mappings are added and
63	<	* removed, but overall, this maintains a commonly accepted time/space
64	<	* tradeoff for hash tables.  However, resizing this or any other kind
65	<	* of hash table may be a relatively slow operation. When possible, it
66	<	* is a good idea to provide a size estimate as an optional {@code
61	>	* effect of maintaining roughly two bins per mapping (corresponding
62	>	* to a 0.75 load factor threshold for resizing). There may be much
63	>	* variance around this average as mappings are added and removed, but
64	>	* overall, this maintains a commonly accepted time/space tradeoff for
65	>	* hash tables.  However, resizing this or any other kind of hash
66	>	* table may be a relatively slow operation. When possible, it is a
67	>	* good idea to provide a size estimate as an optional {@code
68		* initialCapacity} constructor argument. An additional optional
69		* {@code loadFactor} constructor argument provides a further means of
70		* customizing initial table capacity by specifying the table density
#	Line 68 | Line 73 | import java.io.Serializable;
73		* versions of this class, constructors may optionally specify an
74		* expected {@code concurrencyLevel} as an additional hint for
75		* internal sizing.  Note that using many keys with exactly the same
76	<	* {@code hashCode{}} is a sure way to slow down performance of any
76	>	* {@code hashCode()} is a sure way to slow down performance of any
77		* hash table.
78		*
79		* <p>This class and its views and iterators implement all of the
#	Line 95 | Line 100 | public class ConcurrentHashMapV8<K, V>
100		private static final long serialVersionUID = 7249069246763182397L;
101
102		/**
103	<	* A function computing a mapping from the given key to a value,
104	<	* or {@code null} if there is no mapping. This is a place-holder
100	<	* for an upcoming JDK8 interface.
103	>	* A function computing a mapping from the given key to a value.
104	>	* This is a place-holder for an upcoming JDK8 interface.
105		*/
106		public static interface MappingFunction<K, V> {
107		/**
108	<	* Returns a value for the given key, or null if there is no
105	<	* mapping. If this function throws an (unchecked) exception,
106	<	* the exception is rethrown to its caller, and no mapping is
107	<	* recorded.  Because this function is invoked within
108	<	* atomicity control, the computation should be short and
109	<	* simple. The most common usage is to construct a new object
110	<	* serving as an initial mapped value.
108	>	* Returns a value for the given key, or null if there is no mapping.
109		*
110		* @param key the (non-null) key
111	<	* @return a value, or null if none
111	>	* @return a value for the key, or null if none
112		*/
113		V map(K key);
114		}
115
116	+	/**
117	+	* A function computing a new mapping given a key and its current
118	+	* mapped value (or {@code null} if there is no current
119	+	* mapping). This is a place-holder for an upcoming JDK8
120	+	* interface.
121	+	*/
122	+	public static interface RemappingFunction<K, V> {
123	+	/**
124	+	* Returns a new value given a key and its current value.
125	+	*
126	+	* @param key the (non-null) key
127	+	* @param value the current value, or null if there is no mapping
128	+	* @return a value for the key, or null if none
129	+	*/
130	+	V remap(K key, V value);
131	+	}
132	+
133	+	/**
134	+	* A partitionable iterator. A Spliterator can be traversed
135	+	* directly, but can also be partitioned (before traversal) by
136	+	* creating another Spliterator that covers a non-overlapping
137	+	* portion of the elements, and so may be amenable to parallel
138	+	* execution.
139	+	*
140	+	* <p> This interface exports a subset of expected JDK8
141	+	* functionality.
142	+	*
143	+	* <p>Sample usage: Here is one (of the several) ways to compute
144	+	* the sum of the values held in a map using the ForkJoin
145	+	* framework. As illustrated here, Spliterators are well suited to
146	+	* designs in which a task repeatedly splits off half its work
147	+	* into forked subtasks until small enough to process directly,
148	+	* and then joins these subtasks. Variants of this style can be
149	+	* also be used in completion-based designs.
150	+	*
151	+	* <pre>
152	+	* {@code ConcurrentHashMapV8<String, Long> m = ...
153	+	* // Uses parallel depth of log2 of size / (parallelism * slack of 8).
154	+	* int depth = 32 - Integer.numberOfLeadingZeros(m.size() / (aForkJoinPool.getParallelism() * 8));
155	+	* long sum = aForkJoinPool.invoke(new SumValues(m.valueSpliterator(), depth, null));
156	+	* // ...
157	+	* static class SumValues extends RecursiveTask<Long> {
158	+	*   final Spliterator<Long> s;
159	+	*   final int depth;             // number of splits before processing
160	+	*   final SumValues nextJoin;    // records forked subtasks to join
161	+	*   SumValues(Spliterator<Long> s, int depth, SumValues nextJoin) {
162	+	*     this.s = s; this.depth = depth; this.nextJoin = nextJoin;
163	+	*   }
164	+	*   public Long compute() {
165	+	*     long sum = 0;
166	+	*     SumValues subtasks = null; // fork subtasks
167	+	*     for (int d = depth - 1; d >= 0; --d)
168	+	*       (subtasks = new SumValues(s.split(), d, subtasks)).fork();
169	+	*     while (s.hasNext())        // directly process remaining elements
170	+	*       sum += s.next();
171	+	*     for (SumValues t = subtasks; t != null; t = t.nextJoin)
172	+	*       sum += t.join();         // collect subtask results
173	+	*     return sum;
174	+	*   }
175	+	* }
176	+	* }</pre>
177	+	*/
178	+	public static interface Spliterator<T> extends Iterator<T> {
179	+	/**
180	+	* Returns a Spliterator covering approximately half of the
181	+	* elements, guaranteed not to overlap with those subsequently
182	+	* returned by this Spliterator.  After invoking this method,
183	+	* the current Spliterator will <em>not</em> produce any of
184	+	* the elements of the returned Spliterator, but the two
185	+	* Spliterators together will produce all of the elements that
186	+	* would have been produced by this Spliterator had this
187	+	* method not been called. The exact number of elements
188	+	* produced by the returned Spliterator is not guaranteed, and
189	+	* may be zero (i.e., with {@code hasNext()} reporting {@code
190	+	* false}) if this Spliterator cannot be further split.
191	+	*
192	+	* @return a Spliterator covering approximately half of the
193	+	* elements
194	+	* @throws IllegalStateException if this Spliterator has
195	+	* already commenced traversing elements.
196	+	*/
197	+	Spliterator<T> split();
198	+
199	+	/**
200	+	* Returns a Spliterator producing the same elements as this
201	+	* Spliterator. This method may be used for example to create
202	+	* a second Spliterator before a traversal, in order to later
203	+	* perform a second traversal.
204	+	*
205	+	* @return a Spliterator covering the same range as this Spliterator.
206	+	* @throws IllegalStateException if this Spliterator has
207	+	* already commenced traversing elements.
208	+	*/
209	+	Spliterator<T> clone();
210	+	}
211	+
212		/*
213		* Overview:
214		*
215		* The primary design goal of this hash table is to maintain
216		* concurrent readability (typically method get(), but also
217		* iterators and related methods) while minimizing update
218	<	* contention.
218	>	* contention. Secondary goals are to keep space consumption about
219	>	* the same or better than java.util.HashMap, and to support high
220	>	* initial insertion rates on an empty table by many threads.
221		*
222		* Each key-value mapping is held in a Node.  Because Node fields
223		* can contain special values, they are defined using plain Object
#	Line 129 | Line 225 | public class ConcurrentHashMapV8<K, V>
225		* work off Object types. And similarly, so do the internal
226		* methods of auxiliary iterator and view classes.  All public
227		* generic typed methods relay in/out of these internal methods,
228	<	* supplying null-checks and casts as needed.
228	>	* supplying null-checks and casts as needed. This also allows
229	>	* many of the public methods to be factored into a smaller number
230	>	* of internal methods (although sadly not so for the five
231	>	* variants of put-related operations). The validation-based
232	>	* approach explained below leads to a lot of code sprawl because
233	>	* retry-control precludes factoring into smaller methods.
234		*
235		* The table is lazily initialized to a power-of-two size upon the
236	<	* first insertion.  Each bin in the table contains a list of
237	<	* Nodes (most often, zero or one Node).  Table accesses require
238	<	* volatile/atomic reads, writes, and CASes.  Because there is no
239	<	* other way to arrange this without adding further indirections,
240	<	* we use intrinsics (sun.misc.Unsafe) operations.  The lists of
241	<	* nodes within bins are always accurately traversable under
242	<	* volatile reads, so long as lookups check hash code and
243	<	* non-nullness of value before checking key equality. (All valid
244	<	* hash codes are nonnegative. Negative values are reserved for
245	<	* special forwarding nodes; see below.)
236	>	* first insertion.  Each bin in the table normally contains a
237	>	* list of Nodes (most often, the list has only zero or one Node).
238	>	* Table accesses require volatile/atomic reads, writes, and
239	>	* CASes.  Because there is no other way to arrange this without
240	>	* adding further indirections, we use intrinsics
241	>	* (sun.misc.Unsafe) operations.  The lists of nodes within bins
242	>	* are always accurately traversable under volatile reads, so long
243	>	* as lookups check hash code and non-nullness of value before
244	>	* checking key equality.
245	>	*
246	>	* We use the top two bits of Node hash fields for control
247	>	* purposes -- they are available anyway because of addressing
248	>	* constraints.  As explained further below, these top bits are
249	>	* used as follows:
250	>	*  00 - Normal
251	>	*  01 - Locked
252	>	*  11 - Locked and may have a thread waiting for lock
253	>	*  10 - Node is a forwarding node
254	>	*
255	>	* The lower 30 bits of each Node's hash field contain a
256	>	* transformation of the key's hash code, except for forwarding
257	>	* nodes, for which the lower bits are zero (and so always have
258	>	* hash field == MOVED).
259		*
260	<	* Insertion (via put or putIfAbsent) of the first node in an
260	>	* Insertion (via put or its variants) of the first node in an
261		* empty bin is performed by just CASing it to the bin.  This is
262	<	* on average by far the most common case for put operations.
263	<	* Other update operations (insert, delete, and replace) require
264	<	* locks.  We do not want to waste the space required to associate
265	<	* a distinct lock object with each bin, so instead use the first
266	<	* node of a bin list itself as a lock, using plain "synchronized"
267	<	* locks. These save space and we can live with block-structured
268	<	* lock/unlock operations. Using the first node of a list as a
269	<	* lock does not by itself suffice though. When a node is locked,
270	<	* any update must first validate that it is still the first node,
271	<	* and retry if not. Because new nodes are always appended to
272	<	* lists, once a node is first in a bin, it remains first until
273	<	* deleted or the bin becomes invalidated.  However, operations
274	<	* that only conditionally update can and sometimes do inspect
275	<	* nodes until the point of update. This is a converse of sorts to
276	<	* the lazy locking technique described by Herlihy & Shavit.
262	>	* by far the most common case for put operations under most
263	>	* key/hash distributions.  Other update operations (insert,
264	>	* delete, and replace) require locks.  We do not want to waste
265	>	* the space required to associate a distinct lock object with
266	>	* each bin, so instead use the first node of a bin list itself as
267	>	* a lock. Blocking support for these locks relies on the builtin
268	>	* "synchronized" monitors.  However, we also need a tryLock
269	>	* construction, so we overlay these by using bits of the Node
270	>	* hash field for lock control (see above), and so normally use
271	>	* builtin monitors only for blocking and signalling using
272	>	* wait/notifyAll constructions. See Node.tryAwaitLock.
273	>	*
274	>	* Using the first node of a list as a lock does not by itself
275	>	* suffice though: When a node is locked, any update must first
276	>	* validate that it is still the first node after locking it, and
277	>	* retry if not. Because new nodes are always appended to lists,
278	>	* once a node is first in a bin, it remains first until deleted
279	>	* or the bin becomes invalidated (upon resizing).  However,
280	>	* operations that only conditionally update may inspect nodes
281	>	* until the point of update. This is a converse of sorts to the
282	>	* lazy locking technique described by Herlihy & Shavit.
283		*
284	<	* The main disadvantage of this approach is that most update
284	>	* The main disadvantage of per-bin locks is that other update
285		* operations on other nodes in a bin list protected by the same
286		* lock can stall, for example when user equals() or mapping
287	<	* functions take a long time.  However, statistically, this is
288	<	* not a common enough problem to outweigh the time/space overhead
289	<	* of alternatives: Under random hash codes, the frequency of
170	<	* nodes in bins follows a Poisson distribution
287	>	* functions take a long time.  However, statistically, under
288	>	* random hash codes, this is not a common problem.  Ideally, the
289	>	* frequency of nodes in bins follows a Poisson distribution
290		* (http://en.wikipedia.org/wiki/Poisson_distribution) with a
291		* parameter of about 0.5 on average, given the resizing threshold
292		* of 0.75, although with a large variance because of resizing
293		* granularity. Ignoring variance, the expected occurrences of
294		* list size k are (exp(-0.5) * pow(0.5, k) / factorial(k)). The
295	<	* first few values are:
295	>	* first values are:
296		*
297	<	* 0:    0.607
298	<	* 1:    0.303
299	<	* 2:    0.076
300	<	* 3:    0.012
301	<	* more: 0.002
297	>	* 0:    0.60653066
298	>	* 1:    0.30326533
299	>	* 2:    0.07581633
300	>	* 3:    0.01263606
301	>	* 4:    0.00157952
302	>	* 5:    0.00015795
303	>	* 6:    0.00001316
304	>	* 7:    0.00000094
305	>	* 8:    0.00000006
306	>	* more: less than 1 in ten million
307		*
308		* Lock contention probability for two threads accessing distinct
309	<	* elements is roughly 1 / (8 * #elements).  Function "spread"
310	<	* performs hashCode randomization that improves the likelihood
311	<	* that these assumptions hold unless users define exactly the
312	<	* same value for too many hashCodes.
313	<	*
314	<	* The table is resized when occupancy exceeds a threshold.  Only
315	<	* a single thread performs the resize (using field "resizing", to
316	<	* arrange exclusion), but the table otherwise remains usable for
317	<	* reads and updates. Resizing proceeds by transferring bins, one
318	<	* by one, from the table to the next table.  Upon transfer, the
319	<	* old table bin contains only a special forwarding node (with
320	<	* negative hash field) that contains the next table as its
321	<	* key. On encountering a forwarding node, access and update
322	<	* operations restart, using the new table. To ensure concurrent
323	<	* readability of traversals, transfers must proceed from the last
324	<	* bin (table.length - 1) up towards the first.  Upon seeing a
325	<	* forwarding node, traversals (see class InternalIterator)
326	<	* arrange to move to the new table for the rest of the traversal
327	<	* without revisiting nodes.  This constrains bin transfers to a
328	<	* particular order, and so can block indefinitely waiting for the
329	<	* next lock, and other threads cannot help with the transfer.
330	<	* However, expected stalls are infrequent enough to not warrant
331	<	* the additional overhead of access and iteration schemes that
332	<	* could admit out-of-order or concurrent bin transfers.
309	>	* elements is roughly 1 / (8 * #elements) under random hashes.
310	>	*
311	>	* Actual hash code distributions encountered in practice
312	>	* sometimes deviate significantly from uniform randomness.  This
313	>	* includes the case when N > (1<<30), so some keys MUST collide.
314	>	* Similarly for dumb or hostile usages in which multiple keys are
315	>	* designed to have identical hash codes. Also, although we guard
316	>	* against the worst effects of this (see method spread), sets of
317	>	* hashes may differ only in bits that do not impact their bin
318	>	* index for a given power-of-two mask.  So we use a secondary
319	>	* strategy that applies when the number of nodes in a bin exceeds
320	>	* a threshold, and at least one of the keys implements
321	>	* Comparable.  These TreeBins use a balanced tree to hold nodes
322	>	* (a specialized form of red-black trees), bounding search time
323	>	* to O(log N).  Each search step in a TreeBin is around twice as
324	>	* slow as in a regular list, but given that N cannot exceed
325	>	* (1<<64) (before running out of addresses) this bounds search
326	>	* steps, lock hold times, etc, to reasonable constants (roughly
327	>	* 100 nodes inspected per operation worst case) so long as keys
328	>	* are Comparable (which is very common -- String, Long, etc).
329	>	* TreeBin nodes (TreeNodes) also maintain the same "next"
330	>	* traversal pointers as regular nodes, so can be traversed in
331	>	* iterators in the same way.
332	>	*
333	>	* The table is resized when occupancy exceeds a percentage
334	>	* threshold (nominally, 0.75, but see below).  Only a single
335	>	* thread performs the resize (using field "sizeCtl", to arrange
336	>	* exclusion), but the table otherwise remains usable for reads
337	>	* and updates. Resizing proceeds by transferring bins, one by
338	>	* one, from the table to the next table.  Because we are using
339	>	* power-of-two expansion, the elements from each bin must either
340	>	* stay at same index, or move with a power of two offset. We
341	>	* eliminate unnecessary node creation by catching cases where old
342	>	* nodes can be reused because their next fields won't change.  On
343	>	* average, only about one-sixth of them need cloning when a table
344	>	* doubles. The nodes they replace will be garbage collectable as
345	>	* soon as they are no longer referenced by any reader thread that
346	>	* may be in the midst of concurrently traversing table.  Upon
347	>	* transfer, the old table bin contains only a special forwarding
348	>	* node (with hash field "MOVED") that contains the next table as
349	>	* its key. On encountering a forwarding node, access and update
350	>	* operations restart, using the new table.
351	>	*
352	>	* Each bin transfer requires its bin lock. However, unlike other
353	>	* cases, a transfer can skip a bin if it fails to acquire its
354	>	* lock, and revisit it later (unless it is a TreeBin). Method
355	>	* rebuild maintains a buffer of TRANSFER_BUFFER_SIZE bins that
356	>	* have been skipped because of failure to acquire a lock, and
357	>	* blocks only if none are available (i.e., only very rarely).
358	>	* The transfer operation must also ensure that all accessible
359	>	* bins in both the old and new table are usable by any traversal.
360	>	* When there are no lock acquisition failures, this is arranged
361	>	* simply by proceeding from the last bin (table.length - 1) up
362	>	* towards the first.  Upon seeing a forwarding node, traversals
363	>	* (see class InternalIterator) arrange to move to the new table
364	>	* without revisiting nodes.  However, when any node is skipped
365	>	* during a transfer, all earlier table bins may have become
366	>	* visible, so are initialized with a reverse-forwarding node back
367	>	* to the old table until the new ones are established. (This
368	>	* sometimes requires transiently locking a forwarding node, which
369	>	* is possible under the above encoding.) These more expensive
370	>	* mechanics trigger only when necessary.
371		*
372	<	* This traversal scheme also applies to partial traversals of
372	>	* The traversal scheme also applies to partial traversals of
373		* ranges of bins (via an alternate InternalIterator constructor)
374	<	* to support partitioned aggregate operations (that are not
375	<	* otherwise implemented yet).  Also, read-only operations give up
376	<	* if ever forwarded to a null table, which provides support for
377	<	* shutdown-style clearing, which is also not currently
216	<	* implemented.
374	>	* to support partitioned aggregate operations.  Also, read-only
375	>	* operations give up if ever forwarded to a null table, which
376	>	* provides support for shutdown-style clearing, which is also not
377	>	* currently implemented.
378		*
379		* Lazy table initialization minimizes footprint until first use,
380		* and also avoids resizings when the first operation is from a
381		* putAll, constructor with map argument, or deserialization.
382	<	* These cases attempt to override the targetCapacity used in
383	<	* growTable. These harmlessly fail to take effect in cases of
223	<	* races with other ongoing resizings. Uses of the threshold and
224	<	* targetCapacity during attempted initializations or resizings
225	<	* are racy but fall back on checks to preserve correctness.
382	>	* These cases attempt to override the initial capacity settings,
383	>	* but harmlessly fail to take effect in cases of races.
384		*
385		* The element count is maintained using a LongAdder, which avoids
386		* contention on updates but can encounter cache thrashing if read
387		* too frequently during concurrent access. To avoid reading so
388	<	* often, resizing is normally attempted only upon adding to a bin
389	<	* already holding two or more nodes. Under uniform hash
390	<	* distributions, the probability of this occurring at threshold
391	<	* is around 13%, meaning that only about 1 in 8 puts check
392	<	* threshold (and after resizing, many fewer do so). But this
393	<	* approximation has high variance for small table sizes, so we
394	<	* check on any collision for sizes <= 64.  Further, to increase
395	<	* the probability that a resize occurs soon enough, we offset the
396	<	* threshold (see THRESHOLD_OFFSET) by the expected number of puts
397	<	* between checks.
388	>	* often, resizing is attempted either when a bin lock is
389	>	* contended, or upon adding to a bin already holding two or more
390	>	* nodes (checked before adding in the xIfAbsent methods, after
391	>	* adding in others). Under uniform hash distributions, the
392	>	* probability of this occurring at threshold is around 13%,
393	>	* meaning that only about 1 in 8 puts check threshold (and after
394	>	* resizing, many fewer do so). But this approximation has high
395	>	* variance for small table sizes, so we check on any collision
396	>	* for sizes <= 64. The bulk putAll operation further reduces
397	>	* contention by only committing count updates upon these size
398	>	* checks.
399		*
400		* Maintaining API and serialization compatibility with previous
401		* versions of this class introduces several oddities. Mainly: We
402		* leave untouched but unused constructor arguments refering to
403	<	* concurrencyLevel. We also declare an unused "Segment" class
404	<	* that is instantiated in minimal form only when serializing.
403	>	* concurrencyLevel. We accept a loadFactor constructor argument,
404	>	* but apply it only to initial table capacity (which is the only
405	>	* time that we can guarantee to honor it.) We also declare an
406	>	* unused "Segment" class that is instantiated in minimal form
407	>	* only when serializing.
408		*/
409
410		/* ---------------- Constants -------------- */
#	Line 250 | Line 412 | public class ConcurrentHashMapV8<K, V>
412		/**
413		* The largest possible table capacity.  This value must be
414		* exactly 1<<30 to stay within Java array allocation and indexing
415	<	* bounds for power of two table sizes.
415	>	* bounds for power of two table sizes, and is further required
416	>	* because the top two bits of 32bit hash fields are used for
417	>	* control purposes.
418		*/
419		private static final int MAXIMUM_CAPACITY = 1 << 30;
420
#	Line 261 | Line 425 | public class ConcurrentHashMapV8<K, V>
425		private static final int DEFAULT_CAPACITY = 16;
426
427		/**
428	<	* The load factor for this table. Overrides of this value in
429	<	* constructors affect only the initial table capacity.  The
266	<	* actual floating point value isn't normally used, because it is
267	<	* simpler to rely on the expression {@code n - (n >>> 2)} for the
268	<	* associated resizing threshold.
428	>	* The largest possible (non-power of two) array size.
429	>	* Needed by toArray and related methods.
430		*/
431	<	private static final float LOAD_FACTOR = 0.75f;
431	>	static final int MAX_ARRAY_SIZE = Integer.MAX_VALUE - 8;
432
433		/**
434	<	* The count value to offset thresholds to compensate for checking
435	<	* for the need to resize only when inserting into bins with two
275	<	* or more elements. See above for explanation.
434	>	* The default concurrency level for this table. Unused but
435	>	* defined for compatibility with previous versions of this class.
436		*/
437	<	private static final int THRESHOLD_OFFSET = 8;
437	>	private static final int DEFAULT_CONCURRENCY_LEVEL = 16;
438
439		/**
440	<	* The default concurrency level for this table. Unused except as
441	<	* a sizing hint, but defined for compatibility with previous
442	<	* versions of this class.
440	>	* The load factor for this table. Overrides of this value in
441	>	* constructors affect only the initial table capacity.  The
442	>	* actual floating point value isn't normally used -- it is
443	>	* simpler to use expressions such as {@code n - (n >>> 2)} for
444	>	* the associated resizing threshold.
445		*/
446	<	private static final int DEFAULT_CONCURRENCY_LEVEL = 16;
285	<
286	<	/* ---------------- Nodes -------------- */
446	>	private static final float LOAD_FACTOR = 0.75f;
447
448		/**
449	<	* Key-value entry. Note that this is never exported out as a
450	<	* user-visible Map.Entry. Nodes with a negative hash field are
451	<	* special, and do not contain user keys or values.  Otherwise,
292	<	* keys are never null, and null val fields indicate that a node
293	<	* is in the process of being deleted or created. For purposes of
294	<	* read-only access, a key may be read before a val, but can only
295	<	* be used after checking val.  (For an update operation, when a
296	<	* lock is held on a node, order doesn't matter.)
449	>	* The buffer size for skipped bins during transfers. The
450	>	* value is arbitrary but should be large enough to avoid
451	>	* most locking stalls during resizes.
452		*/
453	<	static final class Node {
299	<	final int hash;
300	<	final Object key;
301	<	volatile Object val;
302	<	volatile Node next;
303	<
304	<	Node(int hash, Object key, Object val, Node next) {
305	<	this.hash = hash;
306	<	this.key = key;
307	<	this.val = val;
308	<	this.next = next;
309	<	}
310	<	}
453	>	private static final int TRANSFER_BUFFER_SIZE = 32;
454
455		/**
456	<	* Sign bit of node hash value indicating to use table in node.key.
456	>	* The bin count threshold for using a tree rather than list for a
457	>	* bin.  The value reflects the approximate break-even point for
458	>	* using tree-based operations.
459		*/
460	<	private static final int SIGN_BIT = 0x80000000;
460	>	private static final int TREE_THRESHOLD = 8;
461	>
462	>	/*
463	>	* Encodings for special uses of Node hash fields. See above for
464	>	* explanation.
465	>	*/
466	>	static final int MOVED     = 0x80000000; // hash field for forwarding nodes
467	>	static final int LOCKED    = 0x40000000; // set/tested only as a bit
468	>	static final int WAITING   = 0xc0000000; // both bits set/tested together
469	>	static final int HASH_BITS = 0x3fffffff; // usable bits of normal node hash
470
471		/* ---------------- Fields -------------- */
472
#	Line 322 | Line 476 | public class ConcurrentHashMapV8<K, V>
476		*/
477		transient volatile Node[] table;
478
479	<	/** The counter maintaining number of elements. */
479	>	/**
480	>	* The counter maintaining number of elements.
481	>	*/
482		private transient final LongAdder counter;
483	<	/** Nonzero when table is being initialized or resized. Updated via CAS. */
484	<	private transient volatile int resizing;
485	<	/** The next element count value upon which to resize the table. */
486	<	private transient int threshold;
487	<	/** The target capacity; volatile to cover initialization races. */
488	<	private transient volatile int targetCapacity;
483	>
484	>	/**
485	>	* Table initialization and resizing control.  When negative, the
486	>	* table is being initialized or resized. Otherwise, when table is
487	>	* null, holds the initial table size to use upon creation, or 0
488	>	* for default. After initialization, holds the next element count
489	>	* value upon which to resize the table.
490	>	*/
491	>	private transient volatile int sizeCtl;
492
493		// views
494		private transient KeySet<K,V> keySet;
#	Line 365 | Line 524 | public class ConcurrentHashMapV8<K, V>
524		UNSAFE.putObjectVolatile(tab, ((long)i<<ASHIFT)+ABASE, v);
525		}
526
527	<	/* ----------------Table Initialization and Resizing -------------- */
527	>	/* ---------------- Nodes -------------- */
528
529		/**
530	<	* Returns a power of two table size for the given desired capacity.
531	<	* See Hackers Delight, sec 3.2
530	>	* Key-value entry. Note that this is never exported out as a
531	>	* user-visible Map.Entry (see MapEntry below). Nodes with a hash
532	>	* field of MOVED are special, and do not contain user keys or
533	>	* values.  Otherwise, keys are never null, and null val fields
534	>	* indicate that a node is in the process of being deleted or
535	>	* created. For purposes of read-only access, a key may be read
536	>	* before a val, but can only be used after checking val to be
537	>	* non-null.
538		*/
539	<	private static final int tableSizeFor(int c) {
540	<	int n = c - 1;
541	<	n |= n >>> 1;
542	<	n |= n >>> 2;
543	<	n |= n >>> 4;
379	<	n |= n >>> 8;
380	<	n |= n >>> 16;
381	<	return (n < 0) ? 1 : (n >= MAXIMUM_CAPACITY) ? MAXIMUM_CAPACITY : n + 1;
382	<	}
539	>	static class Node {
540	>	volatile int hash;
541	>	final Object key;
542	>	volatile Object val;
543	>	volatile Node next;
544
545	<	/**
546	<	* If not already resizing, initializes or creates next table and
547	<	* transfers bins. Initial table size uses the capacity recorded
548	<	* in targetCapacity.  Rechecks occupancy after a transfer to see
549	<	* if another resize is already needed because resizings are
550	<	* lagging additions.
551	<	*
552	<	* @return current table
553	<	*/
554	<	private final Node[] growTable() {
555	<	if (resizing == 0 &&
556	<	UNSAFE.compareAndSwapInt(this, resizingOffset, 0, 1)) {
557	<	try {
558	<	for (;;) {
559	<	Node[] tab = table;
560	<	int n, c, m;
561	<	if (tab == null)
562	<	n = (c = targetCapacity) > 0 ? c : DEFAULT_CAPACITY;
563	<	else if ((m = tab.length) < MAXIMUM_CAPACITY &&
564	<	counter.sum() >= (long)threshold)
565	<	n = m << 1;
566	<	else
567	<	break;
568	<	threshold = n - (n >>> 2) - THRESHOLD_OFFSET;
569	<	Node[] nextTab = new Node[n];
570	<	if (tab != null)
571	<	transfer(tab, nextTab,
572	<	new Node(SIGN_BIT, nextTab, null, null));
573	<	table = nextTab;
574	<	if (tab == null)
545	>	Node(int hash, Object key, Object val, Node next) {
546	>	this.hash = hash;
547	>	this.key = key;
548	>	this.val = val;
549	>	this.next = next;
550	>	}
551	>
552	>	/** CompareAndSet the hash field */
553	>	final boolean casHash(int cmp, int val) {
554	>	return UNSAFE.compareAndSwapInt(this, hashOffset, cmp, val);
555	>	}
556	>
557	>	/** The number of spins before blocking for a lock */
558	>	static final int MAX_SPINS =
559	>	Runtime.getRuntime().availableProcessors() > 1 ? 64 : 1;
560	>
561	>	/**
562	>	* Spins a while if LOCKED bit set and this node is the first
563	>	* of its bin, and then sets WAITING bits on hash field and
564	>	* blocks (once) if they are still set.  It is OK for this
565	>	* method to return even if lock is not available upon exit,
566	>	* which enables these simple single-wait mechanics.
567	>	*
568	>	* The corresponding signalling operation is performed within
569	>	* callers: Upon detecting that WAITING has been set when
570	>	* unlocking lock (via a failed CAS from non-waiting LOCKED
571	>	* state), unlockers acquire the sync lock and perform a
572	>	* notifyAll.
573	>	*/
574	>	final void tryAwaitLock(Node[] tab, int i) {
575	>	if (tab != null && i >= 0 && i < tab.length) { // bounds check
576	>	int r = ThreadLocalRandom.current().nextInt(); // randomize spins
577	>	int spins = MAX_SPINS, h;
578	>	while (tabAt(tab, i) == this && ((h = hash) & LOCKED) != 0) {
579	>	if (spins >= 0) {
580	>	r ^= r << 1; r ^= r >>> 3; r ^= r << 10; // xorshift
581	>	if (r >= 0 && --spins == 0)
582	>	Thread.yield();  // yield before block
583	>	}
584	>	else if (casHash(h, h | WAITING)) {
585	>	synchronized (this) {
586	>	if (tabAt(tab, i) == this &&
587	>	(hash & WAITING) == WAITING) {
588	>	try {
589	>	wait();
590	>	} catch (InterruptedException ie) {
591	>	Thread.currentThread().interrupt();
592	>	}
593	>	}
594	>	else
595	>	notifyAll(); // possibly won race vs signaller
596	>	}
597		break;
598	+	}
599		}
416	–	} finally {
417	–	resizing = 0;
600		}
601		}
602	<	else if (table == null)
603	<	Thread.yield(); // lost initialization race; just spin
604	<	return table;
602	>
603	>	// Unsafe mechanics for casHash
604	>	private static final sun.misc.Unsafe UNSAFE;
605	>	private static final long hashOffset;
606	>
607	>	static {
608	>	try {
609	>	UNSAFE = getUnsafe();
610	>	Class<?> k = Node.class;
611	>	hashOffset = UNSAFE.objectFieldOffset
612	>	(k.getDeclaredField("hash"));
613	>	} catch (Exception e) {
614	>	throw new Error(e);
615	>	}
616	>	}
617		}
618
619	+	/* ---------------- TreeBins -------------- */
620	+
621		/**
622	<	* Reclassifies nodes in each bin to new table.  Because we are
427	<	* using power-of-two expansion, the elements from each bin must
428	<	* either stay at same index, or move with a power of two
429	<	* offset. We eliminate unnecessary node creation by catching
430	<	* cases where old nodes can be reused because their next fields
431	<	* won't change.  Statistically, only about one-sixth of them need
432	<	* cloning when a table doubles. The nodes they replace will be
433	<	* garbage collectable as soon as they are no longer referenced by
434	<	* any reader thread that may be in the midst of concurrently
435	<	* traversing table.
436	<	*
437	<	* Transfers are done from the bottom up to preserve iterator
438	<	* traversability. On each step, the old bin is locked,
439	<	* moved/copied, and then replaced with a forwarding node.
622	>	* Nodes for use in TreeBins
623		*/
624	<	private static final void transfer(Node[] tab, Node[] nextTab, Node fwd) {
625	<	int n = tab.length;
626	<	Node ignore = nextTab[n + n - 1]; // force bounds check
627	<	for (int i = n - 1; i >= 0; --i) {
628	<	for (Node e;;) {
629	<	if ((e = tabAt(tab, i)) != null) {
630	<	boolean validated = false;
631	<	synchronized (e) {
632	<	if (tabAt(tab, i) == e) {
633	<	validated = true;
634	<	Node lo = null, hi = null, lastRun = e;
635	<	int runBit = e.hash & n;
636	<	for (Node p = e.next; p != null; p = p.next) {
637	<	int b = p.hash & n;
638	<	if (b != runBit) {
639	<	runBit = b;
640	<	lastRun = p;
624	>	static final class TreeNode extends Node {
625	>	TreeNode parent;  // red-black tree links
626	>	TreeNode left;
627	>	TreeNode right;
628	>	TreeNode prev;    // needed to unlink next upon deletion
629	>	boolean red;
630	>
631	>	TreeNode(int hash, Object key, Object val, Node next, TreeNode parent) {
632	>	super(hash, key, val, next);
633	>	this.parent = parent;
634	>	}
635	>	}
636	>
637	>	/**
638	>	* A specialized form of red-black tree for use in bins
639	>	* whose size exceeds a threshold.
640	>	*
641	>	* TreeBins use a special form of comparison for search and
642	>	* related operations (which is the main reason we cannot use
643	>	* existing collections such as TreeMaps). TreeBins contain
644	>	* Comparable elements, but may contain others, as well as
645	>	* elements that are Comparable but not necessarily Comparable<T>
646	>	* for the same T, so we cannot invoke compareTo among them. To
647	>	* handle this, the tree is ordered primarily by hash value, then
648	>	* by getClass().getName() order, and then by Comparator order
649	>	* among elements of the same class.  On lookup at a node, if
650	>	* elements are not comparable or compare as 0, both left and
651	>	* right children may need to be searched in the case of tied hash
652	>	* values. (This corresponds to the full list search that would be
653	>	* necessary if all elements were non-Comparable and had tied
654	>	* hashes.)  The red-black balancing code is updated from
655	>	* pre-jdk-collections
656	>	* (http://gee.cs.oswego.edu/dl/classes/collections/RBCell.java)
657	>	* based in turn on Cormen, Leiserson, and Rivest "Introduction to
658	>	* Algorithms" (CLR).
659	>	*
660	>	* TreeBins also maintain a separate locking discipline than
661	>	* regular bins. Because they are forwarded via special MOVED
662	>	* nodes at bin heads (which can never change once established),
663	>	* we cannot use use those nodes as locks. Instead, TreeBin
664	>	* extends AbstractQueuedSynchronizer to support a simple form of
665	>	* read-write lock. For update operations and table validation,
666	>	* the exclusive form of lock behaves in the same way as bin-head
667	>	* locks. However, lookups use shared read-lock mechanics to allow
668	>	* multiple readers in the absence of writers.  Additionally,
669	>	* these lookups do not ever block: While the lock is not
670	>	* available, they proceed along the slow traversal path (via
671	>	* next-pointers) until the lock becomes available or the list is
672	>	* exhausted, whichever comes first. (These cases are not fast,
673	>	* but maximize aggregate expected throughput.)  The AQS mechanics
674	>	* for doing this are straightforward.  The lock state is held as
675	>	* AQS getState().  Read counts are negative; the write count (1)
676	>	* is positive.  There are no signalling preferences among readers
677	>	* and writers. Since we don't need to export full Lock API, we
678	>	* just override the minimal AQS methods and use them directly.
679	>	*/
680	>	static final class TreeBin extends AbstractQueuedSynchronizer {
681	>	private static final long serialVersionUID = 2249069246763182397L;
682	>	transient TreeNode root;  // root of tree
683	>	transient TreeNode first; // head of next-pointer list
684	>
685	>	/* AQS overrides */
686	>	public final boolean isHeldExclusively() { return getState() > 0; }
687	>	public final boolean tryAcquire(int ignore) {
688	>	if (compareAndSetState(0, 1)) {
689	>	setExclusiveOwnerThread(Thread.currentThread());
690	>	return true;
691	>	}
692	>	return false;
693	>	}
694	>	public final boolean tryRelease(int ignore) {
695	>	setExclusiveOwnerThread(null);
696	>	setState(0);
697	>	return true;
698	>	}
699	>	public final int tryAcquireShared(int ignore) {
700	>	for (int c;;) {
701	>	if ((c = getState()) > 0)
702	>	return -1;
703	>	if (compareAndSetState(c, c -1))
704	>	return 1;
705	>	}
706	>	}
707	>	public final boolean tryReleaseShared(int ignore) {
708	>	int c;
709	>	do {} while (!compareAndSetState(c = getState(), c + 1));
710	>	return c == -1;
711	>	}
712	>
713	>	/** From CLR */
714	>	private void rotateLeft(TreeNode p) {
715	>	if (p != null) {
716	>	TreeNode r = p.right, pp, rl;
717	>	if ((rl = p.right = r.left) != null)
718	>	rl.parent = p;
719	>	if ((pp = r.parent = p.parent) == null)
720	>	root = r;
721	>	else if (pp.left == p)
722	>	pp.left = r;
723	>	else
724	>	pp.right = r;
725	>	r.left = p;
726	>	p.parent = r;
727	>	}
728	>	}
729	>
730	>	/** From CLR */
731	>	private void rotateRight(TreeNode p) {
732	>	if (p != null) {
733	>	TreeNode l = p.left, pp, lr;
734	>	if ((lr = p.left = l.right) != null)
735	>	lr.parent = p;
736	>	if ((pp = l.parent = p.parent) == null)
737	>	root = l;
738	>	else if (pp.right == p)
739	>	pp.right = l;
740	>	else
741	>	pp.left = l;
742	>	l.right = p;
743	>	p.parent = l;
744	>	}
745	>	}
746	>
747	>	/**
748	>	* Return the TreeNode (or null if not found) for the given key
749	>	* starting at given root.
750	>	*/
751	>	@SuppressWarnings("unchecked") // suppress Comparable cast warning
752	>	final TreeNode getTreeNode(int h, Object k, TreeNode p) {
753	>	Class<?> c = k.getClass();
754	>	while (p != null) {
755	>	int dir, ph;  Object pk; Class<?> pc;
756	>	if ((ph = p.hash) == h) {
757	>	if ((pk = p.key) == k || k.equals(pk))
758	>	return p;
759	>	if (c != (pc = pk.getClass()) ||
760	>	!(k instanceof Comparable) ||
761	>	(dir = ((Comparable)k).compareTo((Comparable)pk)) == 0) {
762	>	dir = (c == pc) ? 0 : c.getName().compareTo(pc.getName());
763	>	TreeNode r = null, s = null, pl, pr;
764	>	if (dir >= 0) {
765	>	if ((pl = p.left) != null && h <= pl.hash)
766	>	s = pl;
767	>	}
768	>	else if ((pr = p.right) != null && h >= pr.hash)
769	>	s = pr;
770	>	if (s != null && (r = getTreeNode(h, k, s)) != null)
771	>	return r;
772	>	}
773	>	}
774	>	else
775	>	dir = (h < ph) ? -1 : 1;
776	>	p = (dir > 0) ? p.right : p.left;
777	>	}
778	>	return null;
779	>	}
780	>
781	>	/**
782	>	* Wrapper for getTreeNode used by CHM.get. Tries to obtain
783	>	* read-lock to call getTreeNode, but during failure to get
784	>	* lock, searches along next links.
785	>	*/
786	>	final Object getValue(int h, Object k) {
787	>	Node r = null;
788	>	int c = getState(); // Must read lock state first
789	>	for (Node e = first; e != null; e = e.next) {
790	>	if (c <= 0 && compareAndSetState(c, c - 1)) {
791	>	try {
792	>	r = getTreeNode(h, k, root);
793	>	} finally {
794	>	releaseShared(0);
795	>	}
796	>	break;
797	>	}
798	>	else if ((e.hash & HASH_BITS) == h && k.equals(e.key)) {
799	>	r = e;
800	>	break;
801	>	}
802	>	else
803	>	c = getState();
804	>	}
805	>	return r == null ? null : r.val;
806	>	}
807	>
808	>	/**
809	>	* Find or add a node
810	>	* @return null if added
811	>	*/
812	>	@SuppressWarnings("unchecked") // suppress Comparable cast warning
813	>	final TreeNode putTreeNode(int h, Object k, Object v) {
814	>	Class<?> c = k.getClass();
815	>	TreeNode pp = root, p = null;
816	>	int dir = 0;
817	>	while (pp != null) { // find existing node or leaf to insert at
818	>	int ph;  Object pk; Class<?> pc;
819	>	p = pp;
820	>	if ((ph = p.hash) == h) {
821	>	if ((pk = p.key) == k || k.equals(pk))
822	>	return p;
823	>	if (c != (pc = pk.getClass()) ||
824	>	!(k instanceof Comparable) ||
825	>	(dir = ((Comparable)k).compareTo((Comparable)pk)) == 0) {
826	>	dir = (c == pc) ? 0 : c.getName().compareTo(pc.getName());
827	>	TreeNode r = null, s = null, pl, pr;
828	>	if (dir >= 0) {
829	>	if ((pl = p.left) != null && h <= pl.hash)
830	>	s = pl;
831	>	}
832	>	else if ((pr = p.right) != null && h >= pr.hash)
833	>	s = pr;
834	>	if (s != null && (r = getTreeNode(h, k, s)) != null)
835	>	return r;
836	>	}
837	>	}
838	>	else
839	>	dir = (h < ph) ? -1 : 1;
840	>	pp = (dir > 0) ? p.right : p.left;
841	>	}
842	>
843	>	TreeNode f = first;
844	>	TreeNode x = first = new TreeNode(h, k, v, f, p);
845	>	if (p == null)
846	>	root = x;
847	>	else { // attach and rebalance; adapted from CLR
848	>	TreeNode xp, xpp;
849	>	if (f != null)
850	>	f.prev = x;
851	>	if (dir <= 0)
852	>	p.left = x;
853	>	else
854	>	p.right = x;
855	>	x.red = true;
856	>	while (x != null && (xp = x.parent) != null && xp.red &&
857	>	(xpp = xp.parent) != null) {
858	>	TreeNode xppl = xpp.left;
859	>	if (xp == xppl) {
860	>	TreeNode y = xpp.right;
861	>	if (y != null && y.red) {
862	>	y.red = false;
863	>	xp.red = false;
864	>	xpp.red = true;
865	>	x = xpp;
866	>	}
867	>	else {
868	>	if (x == xp.right) {
869	>	rotateLeft(x = xp);
870	>	xpp = (xp = x.parent) == null ? null : xp.parent;
871	>	}
872	>	if (xp != null) {
873	>	xp.red = false;
874	>	if (xpp != null) {
875	>	xpp.red = true;
876	>	rotateRight(xpp);
877		}
878		}
879	<	if (runBit == 0)
880	<	lo = lastRun;
881	<	else
882	<	hi = lastRun;
883	<	for (Node p = e; p != lastRun; p = p.next) {
884	<	int ph = p.hash;
885	<	Object pk = p.key, pv = p.val;
886	<	if ((ph & n) == 0)
887	<	lo = new Node(ph, pk, pv, lo);
888	<	else
889	<	hi = new Node(ph, pk, pv, hi);
879	>	}
880	>	}
881	>	else {
882	>	TreeNode y = xppl;
883	>	if (y != null && y.red) {
884	>	y.red = false;
885	>	xp.red = false;
886	>	xpp.red = true;
887	>	x = xpp;
888	>	}
889	>	else {
890	>	if (x == xp.left) {
891	>	rotateRight(x = xp);
892	>	xpp = (xp = x.parent) == null ? null : xp.parent;
893	>	}
894	>	if (xp != null) {
895	>	xp.red = false;
896	>	if (xpp != null) {
897	>	xpp.red = true;
898	>	rotateLeft(xpp);
899	>	}
900		}
472	–	setTabAt(nextTab, i, lo);
473	–	setTabAt(nextTab, i + n, hi);
474	–	setTabAt(tab, i, fwd);
901		}
902		}
903	<	if (validated)
903	>	}
904	>	TreeNode r = root;
905	>	if (r != null && r.red)
906	>	r.red = false;
907	>	}
908	>	return null;
909	>	}
910	>
911	>	/**
912	>	* Removes the given node, that must be present before this
913	>	* call.  This is messier than typical red-black deletion code
914	>	* because we cannot swap the contents of an interior node
915	>	* with a leaf successor that is pinned by "next" pointers
916	>	* that are accessible independently of lock. So instead we
917	>	* swap the tree linkages.
918	>	*/
919	>	final void deleteTreeNode(TreeNode p) {
920	>	TreeNode next = (TreeNode)p.next; // unlink traversal pointers
921	>	TreeNode pred = p.prev;
922	>	if (pred == null)
923	>	first = next;
924	>	else
925	>	pred.next = next;
926	>	if (next != null)
927	>	next.prev = pred;
928	>	TreeNode replacement;
929	>	TreeNode pl = p.left;
930	>	TreeNode pr = p.right;
931	>	if (pl != null && pr != null) {
932	>	TreeNode s = pr, sl;
933	>	while ((sl = s.left) != null) // find successor
934	>	s = sl;
935	>	boolean c = s.red; s.red = p.red; p.red = c; // swap colors
936	>	TreeNode sr = s.right;
937	>	TreeNode pp = p.parent;
938	>	if (s == pr) { // p was s's direct parent
939	>	p.parent = s;
940	>	s.right = p;
941	>	}
942	>	else {
943	>	TreeNode sp = s.parent;
944	>	if ((p.parent = sp) != null) {
945	>	if (s == sp.left)
946	>	sp.left = p;
947	>	else
948	>	sp.right = p;
949	>	}
950	>	if ((s.right = pr) != null)
951	>	pr.parent = s;
952	>	}
953	>	p.left = null;
954	>	if ((p.right = sr) != null)
955	>	sr.parent = p;
956	>	if ((s.left = pl) != null)
957	>	pl.parent = s;
958	>	if ((s.parent = pp) == null)
959	>	root = s;
960	>	else if (p == pp.left)
961	>	pp.left = s;
962	>	else
963	>	pp.right = s;
964	>	replacement = sr;
965	>	}
966	>	else
967	>	replacement = (pl != null) ? pl : pr;
968	>	TreeNode pp = p.parent;
969	>	if (replacement == null) {
970	>	if (pp == null) {
971	>	root = null;
972	>	return;
973	>	}
974	>	replacement = p;
975	>	}
976	>	else {
977	>	replacement.parent = pp;
978	>	if (pp == null)
979	>	root = replacement;
980	>	else if (p == pp.left)
981	>	pp.left = replacement;
982	>	else
983	>	pp.right = replacement;
984	>	p.left = p.right = p.parent = null;
985	>	}
986	>	if (!p.red) { // rebalance, from CLR
987	>	TreeNode x = replacement;
988	>	while (x != null) {
989	>	TreeNode xp, xpl;
990	>	if (x.red || (xp = x.parent) == null) {
991	>	x.red = false;
992		break;
993	+	}
994	+	if (x == (xpl = xp.left)) {
995	+	TreeNode sib = xp.right;
996	+	if (sib != null && sib.red) {
997	+	sib.red = false;
998	+	xp.red = true;
999	+	rotateLeft(xp);
1000	+	sib = (xp = x.parent) == null ? null : xp.right;
1001	+	}
1002	+	if (sib == null)
1003	+	x = xp;
1004	+	else {
1005	+	TreeNode sl = sib.left, sr = sib.right;
1006	+	if ((sr == null || !sr.red) &&
1007	+	(sl == null || !sl.red)) {
1008	+	sib.red = true;
1009	+	x = xp;
1010	+	}
1011	+	else {
1012	+	if (sr == null || !sr.red) {
1013	+	if (sl != null)
1014	+	sl.red = false;
1015	+	sib.red = true;
1016	+	rotateRight(sib);
1017	+	sib = (xp = x.parent) == null ? null : xp.right;
1018	+	}
1019	+	if (sib != null) {
1020	+	sib.red = (xp == null) ? false : xp.red;
1021	+	if ((sr = sib.right) != null)
1022	+	sr.red = false;
1023	+	}
1024	+	if (xp != null) {
1025	+	xp.red = false;
1026	+	rotateLeft(xp);
1027	+	}
1028	+	x = root;
1029	+	}
1030	+	}
1031	+	}
1032	+	else { // symmetric
1033	+	TreeNode sib = xpl;
1034	+	if (sib != null && sib.red) {
1035	+	sib.red = false;
1036	+	xp.red = true;
1037	+	rotateRight(xp);
1038	+	sib = (xp = x.parent) == null ? null : xp.left;
1039	+	}
1040	+	if (sib == null)
1041	+	x = xp;
1042	+	else {
1043	+	TreeNode sl = sib.left, sr = sib.right;
1044	+	if ((sl == null || !sl.red) &&
1045	+	(sr == null || !sr.red)) {
1046	+	sib.red = true;
1047	+	x = xp;
1048	+	}
1049	+	else {
1050	+	if (sl == null || !sl.red) {
1051	+	if (sr != null)
1052	+	sr.red = false;
1053	+	sib.red = true;
1054	+	rotateLeft(sib);
1055	+	sib = (xp = x.parent) == null ? null : xp.left;
1056	+	}
1057	+	if (sib != null) {
1058	+	sib.red = (xp == null) ? false : xp.red;
1059	+	if ((sl = sib.left) != null)
1060	+	sl.red = false;
1061	+	}
1062	+	if (xp != null) {
1063	+	xp.red = false;
1064	+	rotateRight(xp);
1065	+	}
1066	+	x = root;
1067	+	}
1068	+	}
1069	+	}
1070		}
1071	<	else if (casTabAt(tab, i, e, fwd))
1072	<	break;
1071	>	}
1072	>	if (p == replacement && (pp = p.parent) != null) {
1073	>	if (p == pp.left) // detach pointers
1074	>	pp.left = null;
1075	>	else if (p == pp.right)
1076	>	pp.right = null;
1077	>	p.parent = null;
1078		}
1079		}
1080		}
1081
1082	<	/* ---------------- Internal access and update methods -------------- */
1082	>	/* ---------------- Collision reduction methods -------------- */
1083
1084		/**
1085	<	* Applies a supplemental hash function to a given hashCode, which
1086	<	* defends against poor quality hash functions.  The result must
1087	<	* be non-negative, and for reasonable performance must have good
1088	<	* avalanche properties; i.e., that each bit of the argument
1089	<	* affects each bit (except sign bit) of the result.
1085	>	* Spreads higher bits to lower, and also forces top 2 bits to 0.
1086	>	* Because the table uses power-of-two masking, sets of hashes
1087	>	* that vary only in bits above the current mask will always
1088	>	* collide. (Among known examples are sets of Float keys holding
1089	>	* consecutive whole numbers in small tables.)  To counter this,
1090	>	* we apply a transform that spreads the impact of higher bits
1091	>	* downward. There is a tradeoff between speed, utility, and
1092	>	* quality of bit-spreading. Because many common sets of hashes
1093	>	* are already reasonably distributed across bits (so don't benefit
1094	>	* from spreading), and because we use trees to handle large sets
1095	>	* of collisions in bins, we don't need excessively high quality.
1096		*/
1097		private static final int spread(int h) {
1098	<	// Apply base step of MurmurHash; see http://code.google.com/p/smhasher/
1099	<	h ^= h >>> 16;
1100	<	h *= 0x85ebca6b;
1101	<	h ^= h >>> 13;
1102	<	h *= 0xc2b2ae35;
1103	<	return (h >>> 16) ^ (h & 0x7fffffff); // mask out sign bit
1098	>	h ^= (h >>> 18) ^ (h >>> 12);
1099	>	return (h ^ (h >>> 10)) & HASH_BITS;
1100	>	}
1101	>
1102	>	/**
1103	>	* Replaces a list bin with a tree bin. Call only when locked.
1104	>	* Fails to replace if the given key is non-comparable or table
1105	>	* is, or needs, resizing.
1106	>	*/
1107	>	private final void replaceWithTreeBin(Node[] tab, int index, Object key) {
1108	>	if ((key instanceof Comparable) &&
1109	>	(tab.length >= MAXIMUM_CAPACITY || counter.sum() < (long)sizeCtl)) {
1110	>	TreeBin t = new TreeBin();
1111	>	for (Node e = tabAt(tab, index); e != null; e = e.next)
1112	>	t.putTreeNode(e.hash & HASH_BITS, e.key, e.val);
1113	>	setTabAt(tab, index, new Node(MOVED, t, null, null));
1114	>	}
1115		}
1116
1117	+	/* ---------------- Internal access and update methods -------------- */
1118	+
1119		/** Implementation for get and containsKey */
1120		private final Object internalGet(Object k) {
1121		int h = spread(k.hashCode());
1122		retry: for (Node[] tab = table; tab != null;) {
1123	<	Node e; Object ek, ev; int eh;  // locals to read fields once
1123	>	Node e, p; Object ek, ev; int eh;      // locals to read fields once
1124		for (e = tabAt(tab, (tab.length - 1) & h); e != null; e = e.next) {
1125	<	if ((eh = e.hash) == h) {
1126	<	if ((ev = e.val) != null &&
1127	<	((ek = e.key) == k || k.equals(ek)))
1128	<	return ev;
1129	<	}
1130	<	else if (eh < 0) {          // sign bit set
1131	<	tab = (Node[])e.key;    // bin was moved during resize
517	<	continue retry;
1125	>	if ((eh = e.hash) == MOVED) {
1126	>	if ((ek = e.key) instanceof TreeBin)  // search TreeBin
1127	>	return ((TreeBin)ek).getValue(h, k);
1128	>	else {                        // restart with new table
1129	>	tab = (Node[])ek;
1130	>	continue retry;
1131	>	}
1132		}
1133	+	else if ((eh & HASH_BITS) == h && (ev = e.val) != null &&
1134	+	((ek = e.key) == k || k.equals(ek)))
1135	+	return ev;
1136		}
1137		break;
1138		}
1139		return null;
1140		}
1141
1142	<	/** Implementation for put and putIfAbsent */
1143	<	private final Object internalPut(Object k, Object v, boolean replace) {
1142	>	/**
1143	>	* Implementation for the four public remove/replace methods:
1144	>	* Replaces node value with v, conditional upon match of cv if
1145	>	* non-null.  If resulting value is null, delete.
1146	>	*/
1147	>	private final Object internalReplace(Object k, Object v, Object cv) {
1148	>	int h = spread(k.hashCode());
1149	>	Object oldVal = null;
1150	>	for (Node[] tab = table;;) {
1151	>	Node f; int i, fh; Object fk;
1152	>	if (tab == null ||
1153	>	(f = tabAt(tab, i = (tab.length - 1) & h)) == null)
1154	>	break;
1155	>	else if ((fh = f.hash) == MOVED) {
1156	>	if ((fk = f.key) instanceof TreeBin) {
1157	>	TreeBin t = (TreeBin)fk;
1158	>	boolean validated = false;
1159	>	boolean deleted = false;
1160	>	t.acquire(0);
1161	>	try {
1162	>	if (tabAt(tab, i) == f) {
1163	>	validated = true;
1164	>	TreeNode p = t.getTreeNode(h, k, t.root);
1165	>	if (p != null) {
1166	>	Object pv = p.val;
1167	>	if (cv == null || cv == pv || cv.equals(pv)) {
1168	>	oldVal = pv;
1169	>	if ((p.val = v) == null) {
1170	>	deleted = true;
1171	>	t.deleteTreeNode(p);
1172	>	}
1173	>	}
1174	>	}
1175	>	}
1176	>	} finally {
1177	>	t.release(0);
1178	>	}
1179	>	if (validated) {
1180	>	if (deleted)
1181	>	counter.add(-1L);
1182	>	break;
1183	>	}
1184	>	}
1185	>	else
1186	>	tab = (Node[])fk;
1187	>	}
1188	>	else if ((fh & HASH_BITS) != h && f.next == null) // precheck
1189	>	break;                          // rules out possible existence
1190	>	else if ((fh & LOCKED) != 0) {
1191	>	checkForResize();               // try resizing if can't get lock
1192	>	f.tryAwaitLock(tab, i);
1193	>	}
1194	>	else if (f.casHash(fh, fh | LOCKED)) {
1195	>	boolean validated = false;
1196	>	boolean deleted = false;
1197	>	try {
1198	>	if (tabAt(tab, i) == f) {
1199	>	validated = true;
1200	>	for (Node e = f, pred = null;;) {
1201	>	Object ek, ev;
1202	>	if ((e.hash & HASH_BITS) == h &&
1203	>	((ev = e.val) != null) &&
1204	>	((ek = e.key) == k || k.equals(ek))) {
1205	>	if (cv == null || cv == ev || cv.equals(ev)) {
1206	>	oldVal = ev;
1207	>	if ((e.val = v) == null) {
1208	>	deleted = true;
1209	>	Node en = e.next;
1210	>	if (pred != null)
1211	>	pred.next = en;
1212	>	else
1213	>	setTabAt(tab, i, en);
1214	>	}
1215	>	}
1216	>	break;
1217	>	}
1218	>	pred = e;
1219	>	if ((e = e.next) == null)
1220	>	break;
1221	>	}
1222	>	}
1223	>	} finally {
1224	>	if (!f.casHash(fh | LOCKED, fh)) {
1225	>	f.hash = fh;
1226	>	synchronized (f) { f.notifyAll(); };
1227	>	}
1228	>	}
1229	>	if (validated) {
1230	>	if (deleted)
1231	>	counter.add(-1L);
1232	>	break;
1233	>	}
1234	>	}
1235	>	}
1236	>	return oldVal;
1237	>	}
1238	>
1239	>	/*
1240	>	* Internal versions of the five insertion methods, each a
1241	>	* little more complicated than the last. All have
1242	>	* the same basic structure as the first (internalPut):
1243	>	*  1. If table uninitialized, create
1244	>	*  2. If bin empty, try to CAS new node
1245	>	*  3. If bin stale, use new table
1246	>	*  4. if bin converted to TreeBin, validate and relay to TreeBin methods
1247	>	*  5. Lock and validate; if valid, scan and add or update
1248	>	*
1249	>	* The others interweave other checks and/or alternative actions:
1250	>	*  * Plain put checks for and performs resize after insertion.
1251	>	*  * putIfAbsent prescans for mapping without lock (and fails to add
1252	>	*    if present), which also makes pre-emptive resize checks worthwhile.
1253	>	*  * computeIfAbsent extends form used in putIfAbsent with additional
1254	>	*    mechanics to deal with, calls, potential exceptions and null
1255	>	*    returns from function call.
1256	>	*  * compute uses the same function-call mechanics, but without
1257	>	*    the prescans
1258	>	*  * putAll attempts to pre-allocate enough table space
1259	>	*    and more lazily performs count updates and checks.
1260	>	*
1261	>	* Someday when details settle down a bit more, it might be worth
1262	>	* some factoring to reduce sprawl.
1263	>	*/
1264	>
1265	>	/** Implementation for put */
1266	>	private final Object internalPut(Object k, Object v) {
1267		int h = spread(k.hashCode());
1268	<	Object oldVal = null;               // previous value or null if none
1268	>	int count = 0;
1269		for (Node[] tab = table;;) {
1270	<	Node e; int i; Object ek, ev;
1270	>	int i; Node f; int fh; Object fk;
1271		if (tab == null)
1272	<	tab = growTable();
1273	<	else if ((e = tabAt(tab, i = (tab.length - 1) & h)) == null) {
1272	>	tab = initTable();
1273	>	else if ((f = tabAt(tab, i = (tab.length - 1) & h)) == null) {
1274		if (casTabAt(tab, i, null, new Node(h, k, v, null)))
1275		break;                   // no lock when adding to empty bin
1276		}
1277	<	else if (e.hash < 0)             // resized -- restart with new table
1278	<	tab = (Node[])e.key;
1279	<	else if (!replace && e.hash == h && (ev = e.val) != null &&
1280	<	((ek = e.key) == k || k.equals(ek))) {
1281	<	if (tabAt(tab, i) == e) {    // inspect and validate 1st node
1282	<	oldVal = ev;             // without lock for putIfAbsent
1283	<	break;
1277	>	else if ((fh = f.hash) == MOVED) {
1278	>	if ((fk = f.key) instanceof TreeBin) {
1279	>	TreeBin t = (TreeBin)fk;
1280	>	Object oldVal = null;
1281	>	t.acquire(0);
1282	>	try {
1283	>	if (tabAt(tab, i) == f) {
1284	>	count = 2;
1285	>	TreeNode p = t.putTreeNode(h, k, v);
1286	>	if (p != null) {
1287	>	oldVal = p.val;
1288	>	p.val = v;
1289	>	}
1290	>	}
1291	>	} finally {
1292	>	t.release(0);
1293	>	}
1294	>	if (count != 0) {
1295	>	if (oldVal != null)
1296	>	return oldVal;
1297	>	break;
1298	>	}
1299		}
1300	+	else
1301	+	tab = (Node[])fk;
1302		}
1303	<	else {
1304	<	boolean validated = false;
1305	<	boolean checkSize = false;
1306	<	synchronized (e) {           // lock the 1st node of bin list
1307	<	if (tabAt(tab, i) == e) {
1308	<	validated = true;    // retry if 1st already deleted
1309	<	for (Node first = e;;) {
1310	<	if (e.hash == h &&
1311	<	((ek = e.key) == k || k.equals(ek)) &&
1312	<	(ev = e.val) != null) {
1303	>	else if ((fh & LOCKED) != 0) {
1304	>	checkForResize();
1305	>	f.tryAwaitLock(tab, i);
1306	>	}
1307	>	else if (f.casHash(fh, fh | LOCKED)) {
1308	>	Object oldVal = null;
1309	>	try {                        // needed in case equals() throws
1310	>	if (tabAt(tab, i) == f) {
1311	>	count = 1;
1312	>	for (Node e = f;; ++count) {
1313	>	Object ek, ev;
1314	>	if ((e.hash & HASH_BITS) == h &&
1315	>	(ev = e.val) != null &&
1316	>	((ek = e.key) == k || k.equals(ek))) {
1317		oldVal = ev;
1318	<	if (replace)
558	<	e.val = v;
1318	>	e.val = v;
1319		break;
1320		}
1321		Node last = e;
1322		if ((e = e.next) == null) {
1323		last.next = new Node(h, k, v, null);
1324	<	if (last != first || tab.length <= 64)
1325	<	checkSize = true;
1324	>	if (count >= TREE_THRESHOLD)
1325	>	replaceWithTreeBin(tab, i, k);
1326		break;
1327		}
1328		}
1329		}
1330	+	} finally {                  // unlock and signal if needed
1331	+	if (!f.casHash(fh | LOCKED, fh)) {
1332	+	f.hash = fh;
1333	+	synchronized (f) { f.notifyAll(); };
1334	+	}
1335		}
1336	<	if (validated) {
1337	<	if (checkSize && tab.length < MAXIMUM_CAPACITY &&
1338	<	resizing == 0 && counter.sum() >= (long)threshold)
1339	<	growTable();
1336	>	if (count != 0) {
1337	>	if (oldVal != null)
1338	>	return oldVal;
1339	>	if (tab.length <= 64)
1340	>	count = 2;
1341		break;
1342		}
1343		}
1344		}
1345	<	if (oldVal == null)
1346	<	counter.increment();             // update counter outside of locks
1347	<	return oldVal;
1345	>	counter.add(1L);
1346	>	if (count > 1)
1347	>	checkForResize();
1348	>	return null;
1349		}
1350
1351	<	/**
1352	<	* Implementation for the four public remove/replace methods:
586	<	* Replaces node value with v, conditional upon match of cv if
587	<	* non-null.  If resulting value is null, delete.
588	<	*/
589	<	private final Object internalReplace(Object k, Object v, Object cv) {
1351	>	/** Implementation for putIfAbsent */
1352	>	private final Object internalPutIfAbsent(Object k, Object v) {
1353		int h = spread(k.hashCode());
1354	+	int count = 0;
1355		for (Node[] tab = table;;) {
1356	<	Node e; int i;
1357	<	if (tab == null ||
1358	<	(e = tabAt(tab, i = (tab.length - 1) & h)) == null)
1359	<	return null;
1360	<	else if (e.hash < 0)
1361	<	tab = (Node[])e.key;
1356	>	int i; Node f; int fh; Object fk, fv;
1357	>	if (tab == null)
1358	>	tab = initTable();
1359	>	else if ((f = tabAt(tab, i = (tab.length - 1) & h)) == null) {
1360	>	if (casTabAt(tab, i, null, new Node(h, k, v, null)))
1361	>	break;
1362	>	}
1363	>	else if ((fh = f.hash) == MOVED) {
1364	>	if ((fk = f.key) instanceof TreeBin) {
1365	>	TreeBin t = (TreeBin)fk;
1366	>	Object oldVal = null;
1367	>	t.acquire(0);
1368	>	try {
1369	>	if (tabAt(tab, i) == f) {
1370	>	count = 2;
1371	>	TreeNode p = t.putTreeNode(h, k, v);
1372	>	if (p != null)
1373	>	oldVal = p.val;
1374	>	}
1375	>	} finally {
1376	>	t.release(0);
1377	>	}
1378	>	if (count != 0) {
1379	>	if (oldVal != null)
1380	>	return oldVal;
1381	>	break;
1382	>	}
1383	>	}
1384	>	else
1385	>	tab = (Node[])fk;
1386	>	}
1387	>	else if ((fh & HASH_BITS) == h && (fv = f.val) != null &&
1388	>	((fk = f.key) == k || k.equals(fk)))
1389	>	return fv;
1390		else {
1391	<	Object oldVal = null;
1392	<	boolean validated = false;
1393	<	boolean deleted = false;
1394	<	synchronized (e) {
1395	<	if (tabAt(tab, i) == e) {
1396	<	validated = true;
1397	<	Node pred = null;
1398	<	do {
1399	<	Object ek, ev;
1400	<	if (e.hash == h &&
1401	<	((ek = e.key) == k || k.equals(ek)) &&
1402	<	((ev = e.val) != null)) {
1403	<	if (cv == null || cv == ev || cv.equals(ev)) {
1391	>	Node g = f.next;
1392	>	if (g != null) { // at least 2 nodes -- search and maybe resize
1393	>	for (Node e = g;;) {
1394	>	Object ek, ev;
1395	>	if ((e.hash & HASH_BITS) == h && (ev = e.val) != null &&
1396	>	((ek = e.key) == k || k.equals(ek)))
1397	>	return ev;
1398	>	if ((e = e.next) == null) {
1399	>	checkForResize();
1400	>	break;
1401	>	}
1402	>	}
1403	>	}
1404	>	if (((fh = f.hash) & LOCKED) != 0) {
1405	>	checkForResize();
1406	>	f.tryAwaitLock(tab, i);
1407	>	}
1408	>	else if (tabAt(tab, i) == f && f.casHash(fh, fh | LOCKED)) {
1409	>	Object oldVal = null;
1410	>	try {
1411	>	if (tabAt(tab, i) == f) {
1412	>	count = 1;
1413	>	for (Node e = f;; ++count) {
1414	>	Object ek, ev;
1415	>	if ((e.hash & HASH_BITS) == h &&
1416	>	(ev = e.val) != null &&
1417	>	((ek = e.key) == k || k.equals(ek))) {
1418		oldVal = ev;
1419	<	if ((e.val = v) == null) {
1420	<	deleted = true;
1421	<	Node en = e.next;
1422	<	if (pred != null)
1423	<	pred.next = en;
1424	<	else
1425	<	setTabAt(tab, i, en);
1426	<	}
1419	>	break;
1420	>	}
1421	>	Node last = e;
1422	>	if ((e = e.next) == null) {
1423	>	last.next = new Node(h, k, v, null);
1424	>	if (count >= TREE_THRESHOLD)
1425	>	replaceWithTreeBin(tab, i, k);
1426	>	break;
1427		}
622	–	break;
1428		}
1429	<	} while ((e = (pred = e).next) != null);
1429	>	}
1430	>	} finally {
1431	>	if (!f.casHash(fh | LOCKED, fh)) {
1432	>	f.hash = fh;
1433	>	synchronized (f) { f.notifyAll(); };
1434	>	}
1435	>	}
1436	>	if (count != 0) {
1437	>	if (oldVal != null)
1438	>	return oldVal;
1439	>	if (tab.length <= 64)
1440	>	count = 2;
1441	>	break;
1442		}
626	–	}
627	–	if (validated) {
628	–	if (deleted)
629	–	counter.decrement();
630	–	return oldVal;
1443		}
1444		}
1445		}
1446	+	counter.add(1L);
1447	+	if (count > 1)
1448	+	checkForResize();
1449	+	return null;
1450		}
1451
1452	<	/** Implementation for computeIfAbsent and compute. Like put, but messier. */
1453	<	@SuppressWarnings("unchecked")
1454	<	private final V internalCompute(K k,
639	<	MappingFunction<? super K, ? extends V> f,
640	<	boolean replace) {
1452	>	/** Implementation for computeIfAbsent */
1453	>	private final Object internalComputeIfAbsent(K k,
1454	>	MappingFunction<? super K, ?> mf) {
1455		int h = spread(k.hashCode());
1456	<	V val = null;
1457	<	boolean added = false;
1458	<	Node[] tab = table;
1459	<	outer:for (;;) {
646	<	Node e; int i; Object ek, ev;
1456	>	Object val = null;
1457	>	int count = 0;
1458	>	for (Node[] tab = table;;) {
1459	>	Node f; int i, fh; Object fk, fv;
1460		if (tab == null)
1461	<	tab = growTable();
1462	<	else if ((e = tabAt(tab, i = (tab.length - 1) & h)) == null) {
1463	<	Node node = new Node(h, k, null, null);
1464	<	boolean validated = false;
1465	<	synchronized (node) {  // must lock while computing value
1466	<	if (casTabAt(tab, i, null, node)) {
1467	<	validated = true;
1468	<	try {
1469	<	val = f.map(k);
1470	<	if (val != null) {
1471	<	node.val = val;
1461	>	tab = initTable();
1462	>	else if ((f = tabAt(tab, i = (tab.length - 1) & h)) == null) {
1463	>	Node node = new Node(fh = h | LOCKED, k, null, null);
1464	>	if (casTabAt(tab, i, null, node)) {
1465	>	count = 1;
1466	>	try {
1467	>	if ((val = mf.map(k)) != null)
1468	>	node.val = val;
1469	>	} finally {
1470	>	if (val == null)
1471	>	setTabAt(tab, i, null);
1472	>	if (!node.casHash(fh, h)) {
1473	>	node.hash = h;
1474	>	synchronized (node) { node.notifyAll(); };
1475	>	}
1476	>	}
1477	>	}
1478	>	if (count != 0)
1479	>	break;
1480	>	}
1481	>	else if ((fh = f.hash) == MOVED) {
1482	>	if ((fk = f.key) instanceof TreeBin) {
1483	>	TreeBin t = (TreeBin)fk;
1484	>	boolean added = false;
1485	>	t.acquire(0);
1486	>	try {
1487	>	if (tabAt(tab, i) == f) {
1488	>	count = 1;
1489	>	TreeNode p = t.getTreeNode(h, k, t.root);
1490	>	if (p != null)
1491	>	val = p.val;
1492	>	else if ((val = mf.map(k)) != null) {
1493		added = true;
1494	+	count = 2;
1495	+	t.putTreeNode(h, k, val);
1496		}
661	–	} finally {
662	–	if (!added)
663	–	setTabAt(tab, i, null);
1497		}
1498	+	} finally {
1499	+	t.release(0);
1500	+	}
1501	+	if (count != 0) {
1502	+	if (!added)
1503	+	return val;
1504	+	break;
1505	+	}
1506	+	}
1507	+	else
1508	+	tab = (Node[])fk;
1509	+	}
1510	+	else if ((fh & HASH_BITS) == h && (fv = f.val) != null &&
1511	+	((fk = f.key) == k || k.equals(fk)))
1512	+	return fv;
1513	+	else {
1514	+	Node g = f.next;
1515	+	if (g != null) {
1516	+	for (Node e = g;;) {
1517	+	Object ek, ev;
1518	+	if ((e.hash & HASH_BITS) == h && (ev = e.val) != null &&
1519	+	((ek = e.key) == k || k.equals(ek)))
1520	+	return ev;
1521	+	if ((e = e.next) == null) {
1522	+	checkForResize();
1523	+	break;
1524	+	}
1525	+	}
1526	+	}
1527	+	if (((fh = f.hash) & LOCKED) != 0) {
1528	+	checkForResize();
1529	+	f.tryAwaitLock(tab, i);
1530	+	}
1531	+	else if (tabAt(tab, i) == f && f.casHash(fh, fh | LOCKED)) {
1532	+	boolean added = false;
1533	+	try {
1534	+	if (tabAt(tab, i) == f) {
1535	+	count = 1;
1536	+	for (Node e = f;; ++count) {
1537	+	Object ek, ev;
1538	+	if ((e.hash & HASH_BITS) == h &&
1539	+	(ev = e.val) != null &&
1540	+	((ek = e.key) == k || k.equals(ek))) {
1541	+	val = ev;
1542	+	break;
1543	+	}
1544	+	Node last = e;
1545	+	if ((e = e.next) == null) {
1546	+	if ((val = mf.map(k)) != null) {
1547	+	added = true;
1548	+	last.next = new Node(h, k, val, null);
1549	+	if (count >= TREE_THRESHOLD)
1550	+	replaceWithTreeBin(tab, i, k);
1551	+	}
1552	+	break;
1553	+	}
1554	+	}
1555	+	}
1556	+	} finally {
1557	+	if (!f.casHash(fh | LOCKED, fh)) {
1558	+	f.hash = fh;
1559	+	synchronized (f) { f.notifyAll(); };
1560	+	}
1561	+	}
1562	+	if (count != 0) {
1563	+	if (!added)
1564	+	return val;
1565	+	if (tab.length <= 64)
1566	+	count = 2;
1567	+	break;
1568		}
1569		}
667	–	if (validated)
668	–	break;
1570		}
1571	<	else if (e.hash < 0)
1572	<	tab = (Node[])e.key;
1573	<	else if (!replace && e.hash == h && (ev = e.val) != null &&
1574	<	((ek = e.key) == k || k.equals(ek))) {
1575	<	if (tabAt(tab, i) == e) {
1576	<	val = (V)ev;
1571	>	}
1572	>	if (val != null) {
1573	>	counter.add(1L);
1574	>	if (count > 1)
1575	>	checkForResize();
1576	>	}
1577	>	return val;
1578	>	}
1579	>
1580	>	/** Implementation for compute */
1581	>	@SuppressWarnings("unchecked")
1582	>	private final Object internalCompute(K k,
1583	>	RemappingFunction<? super K, V> mf) {
1584	>	int h = spread(k.hashCode());
1585	>	Object val = null;
1586	>	int delta = 0;
1587	>	int count = 0;
1588	>	for (Node[] tab = table;;) {
1589	>	Node f; int i, fh; Object fk;
1590	>	if (tab == null)
1591	>	tab = initTable();
1592	>	else if ((f = tabAt(tab, i = (tab.length - 1) & h)) == null) {
1593	>	Node node = new Node(fh = h | LOCKED, k, null, null);
1594	>	if (casTabAt(tab, i, null, node)) {
1595	>	try {
1596	>	count = 1;
1597	>	if ((val = mf.remap(k, null)) != null) {
1598	>	node.val = val;
1599	>	delta = 1;
1600	>	}
1601	>	} finally {
1602	>	if (delta == 0)
1603	>	setTabAt(tab, i, null);
1604	>	if (!node.casHash(fh, h)) {
1605	>	node.hash = h;
1606	>	synchronized (node) { node.notifyAll(); };
1607	>	}
1608	>	}
1609	>	}
1610	>	if (count != 0)
1611		break;
1612	+	}
1613	+	else if ((fh = f.hash) == MOVED) {
1614	+	if ((fk = f.key) instanceof TreeBin) {
1615	+	TreeBin t = (TreeBin)fk;
1616	+	t.acquire(0);
1617	+	try {
1618	+	if (tabAt(tab, i) == f) {
1619	+	count = 1;
1620	+	TreeNode p = t.getTreeNode(h, k, t.root);
1621	+	Object pv = (p == null) ? null : p.val;
1622	+	if ((val = mf.remap(k, (V)pv)) != null) {
1623	+	if (p != null)
1624	+	p.val = val;
1625	+	else {
1626	+	count = 2;
1627	+	delta = 1;
1628	+	t.putTreeNode(h, k, val);
1629	+	}
1630	+	}
1631	+	else if (p != null) {
1632	+	delta = -1;
1633	+	t.deleteTreeNode(p);
1634	+	}
1635	+	}
1636	+	} finally {
1637	+	t.release(0);
1638	+	}
1639	+	if (count != 0)
1640	+	break;
1641		}
1642	+	else
1643	+	tab = (Node[])fk;
1644		}
1645	<	else if (Thread.holdsLock(e))
1646	<	throw new IllegalStateException("Recursive map computation");
1647	<	else {
1648	<	boolean validated = false;
1649	<	boolean checkSize = false;
1650	<	synchronized (e) {
1651	<	if (tabAt(tab, i) == e) {
1652	<	validated = true;
1653	<	for (Node first = e;;) {
1654	<	if (e.hash == h &&
1655	<	((ek = e.key) == k || k.equals(ek)) &&
1656	<	((ev = e.val) != null)) {
1657	<	Object fv;
1658	<	if (replace && (fv = f.map(k)) != null)
1659	<	ev = e.val = fv;
1660	<	val = (V)ev;
1645	>	else if ((fh & LOCKED) != 0) {
1646	>	checkForResize();
1647	>	f.tryAwaitLock(tab, i);
1648	>	}
1649	>	else if (f.casHash(fh, fh | LOCKED)) {
1650	>	try {
1651	>	if (tabAt(tab, i) == f) {
1652	>	count = 1;
1653	>	for (Node e = f, pred = null;; ++count) {
1654	>	Object ek, ev;
1655	>	if ((e.hash & HASH_BITS) == h &&
1656	>	(ev = e.val) != null &&
1657	>	((ek = e.key) == k || k.equals(ek))) {
1658	>	val = mf.remap(k, (V)ev);
1659	>	if (val != null)
1660	>	e.val = val;
1661	>	else {
1662	>	delta = -1;
1663	>	Node en = e.next;
1664	>	if (pred != null)
1665	>	pred.next = en;
1666	>	else
1667	>	setTabAt(tab, i, en);
1668	>	}
1669		break;
1670		}
1671	<	Node last = e;
1671	>	pred = e;
1672		if ((e = e.next) == null) {
1673	<	if ((val = f.map(k)) != null) {
1674	<	last.next = new Node(h, k, val, null);
1675	<	added = true;
1676	<	if (last != first || tab.length <= 64)
1677	<	checkSize = true;
1673	>	if ((val = mf.remap(k, null)) != null) {
1674	>	pred.next = new Node(h, k, val, null);
1675	>	delta = 1;
1676	>	if (count >= TREE_THRESHOLD)
1677	>	replaceWithTreeBin(tab, i, k);
1678		}
1679		break;
1680		}
1681		}
1682		}
1683	+	} finally {
1684	+	if (!f.casHash(fh | LOCKED, fh)) {
1685	+	f.hash = fh;
1686	+	synchronized (f) { f.notifyAll(); };
1687	+	}
1688		}
1689	<	if (validated) {
1690	<	if (checkSize && tab.length < MAXIMUM_CAPACITY &&
1691	<	resizing == 0 && counter.sum() >= (long)threshold)
713	<	growTable();
1689	>	if (count != 0) {
1690	>	if (tab.length <= 64)
1691	>	count = 2;
1692		break;
1693		}
1694		}
1695		}
1696	<	if (added)
1697	<	counter.increment();
1696	>	if (delta != 0) {
1697	>	counter.add((long)delta);
1698	>	if (count > 1)
1699	>	checkForResize();
1700	>	}
1701		return val;
1702		}
1703
1704	+	/** Implementation for putAll */
1705	+	private final void internalPutAll(Map<?, ?> m) {
1706	+	tryPresize(m.size());
1707	+	long delta = 0L;     // number of uncommitted additions
1708	+	boolean npe = false; // to throw exception on exit for nulls
1709	+	try {                // to clean up counts on other exceptions
1710	+	for (Map.Entry<?, ?> entry : m.entrySet()) {
1711	+	Object k, v;
1712	+	if (entry == null || (k = entry.getKey()) == null ||
1713	+	(v = entry.getValue()) == null) {
1714	+	npe = true;
1715	+	break;
1716	+	}
1717	+	int h = spread(k.hashCode());
1718	+	for (Node[] tab = table;;) {
1719	+	int i; Node f; int fh; Object fk;
1720	+	if (tab == null)
1721	+	tab = initTable();
1722	+	else if ((f = tabAt(tab, i = (tab.length - 1) & h)) == null){
1723	+	if (casTabAt(tab, i, null, new Node(h, k, v, null))) {
1724	+	++delta;
1725	+	break;
1726	+	}
1727	+	}
1728	+	else if ((fh = f.hash) == MOVED) {
1729	+	if ((fk = f.key) instanceof TreeBin) {
1730	+	TreeBin t = (TreeBin)fk;
1731	+	boolean validated = false;
1732	+	t.acquire(0);
1733	+	try {
1734	+	if (tabAt(tab, i) == f) {
1735	+	validated = true;
1736	+	TreeNode p = t.getTreeNode(h, k, t.root);
1737	+	if (p != null)
1738	+	p.val = v;
1739	+	else {
1740	+	t.putTreeNode(h, k, v);
1741	+	++delta;
1742	+	}
1743	+	}
1744	+	} finally {
1745	+	t.release(0);
1746	+	}
1747	+	if (validated)
1748	+	break;
1749	+	}
1750	+	else
1751	+	tab = (Node[])fk;
1752	+	}
1753	+	else if ((fh & LOCKED) != 0) {
1754	+	counter.add(delta);
1755	+	delta = 0L;
1756	+	checkForResize();
1757	+	f.tryAwaitLock(tab, i);
1758	+	}
1759	+	else if (f.casHash(fh, fh | LOCKED)) {
1760	+	int count = 0;
1761	+	try {
1762	+	if (tabAt(tab, i) == f) {
1763	+	count = 1;
1764	+	for (Node e = f;; ++count) {
1765	+	Object ek, ev;
1766	+	if ((e.hash & HASH_BITS) == h &&
1767	+	(ev = e.val) != null &&
1768	+	((ek = e.key) == k || k.equals(ek))) {
1769	+	e.val = v;
1770	+	break;
1771	+	}
1772	+	Node last = e;
1773	+	if ((e = e.next) == null) {
1774	+	++delta;
1775	+	last.next = new Node(h, k, v, null);
1776	+	if (count >= TREE_THRESHOLD)
1777	+	replaceWithTreeBin(tab, i, k);
1778	+	break;
1779	+	}
1780	+	}
1781	+	}
1782	+	} finally {
1783	+	if (!f.casHash(fh | LOCKED, fh)) {
1784	+	f.hash = fh;
1785	+	synchronized (f) { f.notifyAll(); };
1786	+	}
1787	+	}
1788	+	if (count != 0) {
1789	+	if (count > 1) {
1790	+	counter.add(delta);
1791	+	delta = 0L;
1792	+	checkForResize();
1793	+	}
1794	+	break;
1795	+	}
1796	+	}
1797	+	}
1798	+	}
1799	+	} finally {
1800	+	if (delta != 0)
1801	+	counter.add(delta);
1802	+	}
1803	+	if (npe)
1804	+	throw new NullPointerException();
1805	+	}
1806	+
1807	+	/* ---------------- Table Initialization and Resizing -------------- */
1808	+
1809	+	/**
1810	+	* Returns a power of two table size for the given desired capacity.
1811	+	* See Hackers Delight, sec 3.2
1812	+	*/
1813	+	private static final int tableSizeFor(int c) {
1814	+	int n = c - 1;
1815	+	n |= n >>> 1;
1816	+	n |= n >>> 2;
1817	+	n |= n >>> 4;
1818	+	n |= n >>> 8;
1819	+	n |= n >>> 16;
1820	+	return (n < 0) ? 1 : (n >= MAXIMUM_CAPACITY) ? MAXIMUM_CAPACITY : n + 1;
1821	+	}
1822	+
1823	+	/**
1824	+	* Initializes table, using the size recorded in sizeCtl.
1825	+	*/
1826	+	private final Node[] initTable() {
1827	+	Node[] tab; int sc;
1828	+	while ((tab = table) == null) {
1829	+	if ((sc = sizeCtl) < 0)
1830	+	Thread.yield(); // lost initialization race; just spin
1831	+	else if (UNSAFE.compareAndSwapInt(this, sizeCtlOffset, sc, -1)) {
1832	+	try {
1833	+	if ((tab = table) == null) {
1834	+	int n = (sc > 0) ? sc : DEFAULT_CAPACITY;
1835	+	tab = table = new Node[n];
1836	+	sc = n - (n >>> 2);
1837	+	}
1838	+	} finally {
1839	+	sizeCtl = sc;
1840	+	}
1841	+	break;
1842	+	}
1843	+	}
1844	+	return tab;
1845	+	}
1846	+
1847	+	/**
1848	+	* If table is too small and not already resizing, creates next
1849	+	* table and transfers bins.  Rechecks occupancy after a transfer
1850	+	* to see if another resize is already needed because resizings
1851	+	* are lagging additions.
1852	+	*/
1853	+	private final void checkForResize() {
1854	+	Node[] tab; int n, sc;
1855	+	while ((tab = table) != null &&
1856	+	(n = tab.length) < MAXIMUM_CAPACITY &&
1857	+	(sc = sizeCtl) >= 0 && counter.sum() >= (long)sc &&
1858	+	UNSAFE.compareAndSwapInt(this, sizeCtlOffset, sc, -1)) {
1859	+	try {
1860	+	if (tab == table) {
1861	+	table = rebuild(tab);
1862	+	sc = (n << 1) - (n >>> 1);
1863	+	}
1864	+	} finally {
1865	+	sizeCtl = sc;
1866	+	}
1867	+	}
1868	+	}
1869	+
1870		/**
1871	<	* Implementation for clear. Steps through each bin, removing all nodes.
1871	>	* Tries to presize table to accommodate the given number of elements.
1872	>	*
1873	>	* @param size number of elements (doesn't need to be perfectly accurate)
1874	>	*/
1875	>	private final void tryPresize(int size) {
1876	>	int c = (size >= (MAXIMUM_CAPACITY >>> 1)) ? MAXIMUM_CAPACITY :
1877	>	tableSizeFor(size + (size >>> 1) + 1);
1878	>	int sc;
1879	>	while ((sc = sizeCtl) >= 0) {
1880	>	Node[] tab = table; int n;
1881	>	if (tab == null || (n = tab.length) == 0) {
1882	>	n = (sc > c) ? sc : c;
1883	>	if (UNSAFE.compareAndSwapInt(this, sizeCtlOffset, sc, -1)) {
1884	>	try {
1885	>	if (table == tab) {
1886	>	table = new Node[n];
1887	>	sc = n - (n >>> 2);
1888	>	}
1889	>	} finally {
1890	>	sizeCtl = sc;
1891	>	}
1892	>	}
1893	>	}
1894	>	else if (c <= sc || n >= MAXIMUM_CAPACITY)
1895	>	break;
1896	>	else if (UNSAFE.compareAndSwapInt(this, sizeCtlOffset, sc, -1)) {
1897	>	try {
1898	>	if (table == tab) {
1899	>	table = rebuild(tab);
1900	>	sc = (n << 1) - (n >>> 1);
1901	>	}
1902	>	} finally {
1903	>	sizeCtl = sc;
1904	>	}
1905	>	}
1906	>	}
1907	>	}
1908	>
1909	>	/*
1910	>	* Moves and/or copies the nodes in each bin to new table. See
1911	>	* above for explanation.
1912	>	*
1913	>	* @return the new table
1914	>	*/
1915	>	private static final Node[] rebuild(Node[] tab) {
1916	>	int n = tab.length;
1917	>	Node[] nextTab = new Node[n << 1];
1918	>	Node fwd = new Node(MOVED, nextTab, null, null);
1919	>	int[] buffer = null;       // holds bins to revisit; null until needed
1920	>	Node rev = null;           // reverse forwarder; null until needed
1921	>	int nbuffered = 0;         // the number of bins in buffer list
1922	>	int bufferIndex = 0;       // buffer index of current buffered bin
1923	>	int bin = n - 1;           // current non-buffered bin or -1 if none
1924	>
1925	>	for (int i = bin;;) {      // start upwards sweep
1926	>	int fh; Node f;
1927	>	if ((f = tabAt(tab, i)) == null) {
1928	>	if (bin >= 0) {    // no lock needed (or available)
1929	>	if (!casTabAt(tab, i, f, fwd))
1930	>	continue;
1931	>	}
1932	>	else {             // transiently use a locked forwarding node
1933	>	Node g = new Node(MOVED|LOCKED, nextTab, null, null);
1934	>	if (!casTabAt(tab, i, f, g))
1935	>	continue;
1936	>	setTabAt(nextTab, i, null);
1937	>	setTabAt(nextTab, i + n, null);
1938	>	setTabAt(tab, i, fwd);
1939	>	if (!g.casHash(MOVED|LOCKED, MOVED)) {
1940	>	g.hash = MOVED;
1941	>	synchronized (g) { g.notifyAll(); }
1942	>	}
1943	>	}
1944	>	}
1945	>	else if ((fh = f.hash) == MOVED) {
1946	>	Object fk = f.key;
1947	>	if (fk instanceof TreeBin) {
1948	>	TreeBin t = (TreeBin)fk;
1949	>	boolean validated = false;
1950	>	t.acquire(0);
1951	>	try {
1952	>	if (tabAt(tab, i) == f) {
1953	>	validated = true;
1954	>	splitTreeBin(nextTab, i, t);
1955	>	setTabAt(tab, i, fwd);
1956	>	}
1957	>	} finally {
1958	>	t.release(0);
1959	>	}
1960	>	if (!validated)
1961	>	continue;
1962	>	}
1963	>	}
1964	>	else if ((fh & LOCKED) == 0 && f.casHash(fh, fh|LOCKED)) {
1965	>	boolean validated = false;
1966	>	try {              // split to lo and hi lists; copying as needed
1967	>	if (tabAt(tab, i) == f) {
1968	>	validated = true;
1969	>	splitBin(nextTab, i, f);
1970	>	setTabAt(tab, i, fwd);
1971	>	}
1972	>	} finally {
1973	>	if (!f.casHash(fh | LOCKED, fh)) {
1974	>	f.hash = fh;
1975	>	synchronized (f) { f.notifyAll(); };
1976	>	}
1977	>	}
1978	>	if (!validated)
1979	>	continue;
1980	>	}
1981	>	else {
1982	>	if (buffer == null) // initialize buffer for revisits
1983	>	buffer = new int[TRANSFER_BUFFER_SIZE];
1984	>	if (bin < 0 && bufferIndex > 0) {
1985	>	int j = buffer[--bufferIndex];
1986	>	buffer[bufferIndex] = i;
1987	>	i = j;         // swap with another bin
1988	>	continue;
1989	>	}
1990	>	if (bin < 0 || nbuffered >= TRANSFER_BUFFER_SIZE) {
1991	>	f.tryAwaitLock(tab, i);
1992	>	continue;      // no other options -- block
1993	>	}
1994	>	if (rev == null)   // initialize reverse-forwarder
1995	>	rev = new Node(MOVED, tab, null, null);
1996	>	if (tabAt(tab, i) != f || (f.hash & LOCKED) == 0)
1997	>	continue;      // recheck before adding to list
1998	>	buffer[nbuffered++] = i;
1999	>	setTabAt(nextTab, i, rev);     // install place-holders
2000	>	setTabAt(nextTab, i + n, rev);
2001	>	}
2002	>
2003	>	if (bin > 0)
2004	>	i = --bin;
2005	>	else if (buffer != null && nbuffered > 0) {
2006	>	bin = -1;
2007	>	i = buffer[bufferIndex = --nbuffered];
2008	>	}
2009	>	else
2010	>	return nextTab;
2011	>	}
2012	>	}
2013	>
2014	>	/**
2015	>	* Split a normal bin with list headed by e into lo and hi parts;
2016	>	* install in given table
2017	>	*/
2018	>	private static void splitBin(Node[] nextTab, int i, Node e) {
2019	>	int bit = nextTab.length >>> 1; // bit to split on
2020	>	int runBit = e.hash & bit;
2021	>	Node lastRun = e, lo = null, hi = null;
2022	>	for (Node p = e.next; p != null; p = p.next) {
2023	>	int b = p.hash & bit;
2024	>	if (b != runBit) {
2025	>	runBit = b;
2026	>	lastRun = p;
2027	>	}
2028	>	}
2029	>	if (runBit == 0)
2030	>	lo = lastRun;
2031	>	else
2032	>	hi = lastRun;
2033	>	for (Node p = e; p != lastRun; p = p.next) {
2034	>	int ph = p.hash & HASH_BITS;
2035	>	Object pk = p.key, pv = p.val;
2036	>	if ((ph & bit) == 0)
2037	>	lo = new Node(ph, pk, pv, lo);
2038	>	else
2039	>	hi = new Node(ph, pk, pv, hi);
2040	>	}
2041	>	setTabAt(nextTab, i, lo);
2042	>	setTabAt(nextTab, i + bit, hi);
2043	>	}
2044	>
2045	>	/**
2046	>	* Split a tree bin into lo and hi parts; install in given table
2047	>	*/
2048	>	private static void splitTreeBin(Node[] nextTab, int i, TreeBin t) {
2049	>	int bit = nextTab.length >>> 1;
2050	>	TreeBin lt = new TreeBin();
2051	>	TreeBin ht = new TreeBin();
2052	>	int lc = 0, hc = 0;
2053	>	for (Node e = t.first; e != null; e = e.next) {
2054	>	int h = e.hash & HASH_BITS;
2055	>	Object k = e.key, v = e.val;
2056	>	if ((h & bit) == 0) {
2057	>	++lc;
2058	>	lt.putTreeNode(h, k, v);
2059	>	}
2060	>	else {
2061	>	++hc;
2062	>	ht.putTreeNode(h, k, v);
2063	>	}
2064	>	}
2065	>	Node ln, hn; // throw away trees if too small
2066	>	if (lc <= (TREE_THRESHOLD >>> 1)) {
2067	>	ln = null;
2068	>	for (Node p = lt.first; p != null; p = p.next)
2069	>	ln = new Node(p.hash, p.key, p.val, ln);
2070	>	}
2071	>	else
2072	>	ln = new Node(MOVED, lt, null, null);
2073	>	setTabAt(nextTab, i, ln);
2074	>	if (hc <= (TREE_THRESHOLD >>> 1)) {
2075	>	hn = null;
2076	>	for (Node p = ht.first; p != null; p = p.next)
2077	>	hn = new Node(p.hash, p.key, p.val, hn);
2078	>	}
2079	>	else
2080	>	hn = new Node(MOVED, ht, null, null);
2081	>	setTabAt(nextTab, i + bit, hn);
2082	>	}
2083	>
2084	>	/**
2085	>	* Implementation for clear. Steps through each bin, removing all
2086	>	* nodes.
2087		*/
2088		private final void internalClear() {
2089		long delta = 0L; // negative number of deletions
2090		int i = 0;
2091		Node[] tab = table;
2092		while (tab != null && i < tab.length) {
2093	<	Node e = tabAt(tab, i);
2094	<	if (e == null)
2093	>	int fh; Object fk;
2094	>	Node f = tabAt(tab, i);
2095	>	if (f == null)
2096		++i;
2097	<	else if (e.hash < 0)
2098	<	tab = (Node[])e.key;
2099	<	else {
2100	<	boolean validated = false;
2101	<	synchronized (e) {
2102	<	if (tabAt(tab, i) == e) {
2103	<	validated = true;
2104	<	Node en;
742	<	do {
743	<	en = e.next;
744	<	if (e.val != null) { // currently always true
745	<	e.val = null;
2097	>	else if ((fh = f.hash) == MOVED) {
2098	>	if ((fk = f.key) instanceof TreeBin) {
2099	>	TreeBin t = (TreeBin)fk;
2100	>	t.acquire(0);
2101	>	try {
2102	>	if (tabAt(tab, i) == f) {
2103	>	for (Node p = t.first; p != null; p = p.next) {
2104	>	p.val = null;
2105		--delta;
2106		}
2107	<	} while ((e = en) != null);
2107	>	t.first = null;
2108	>	t.root = null;
2109	>	++i;
2110	>	}
2111	>	} finally {
2112	>	t.release(0);
2113	>	}
2114	>	}
2115	>	else
2116	>	tab = (Node[])fk;
2117	>	}
2118	>	else if ((fh & LOCKED) != 0) {
2119	>	counter.add(delta); // opportunistically update count
2120	>	delta = 0L;
2121	>	f.tryAwaitLock(tab, i);
2122	>	}
2123	>	else if (f.casHash(fh, fh | LOCKED)) {
2124	>	try {
2125	>	if (tabAt(tab, i) == f) {
2126	>	for (Node e = f; e != null; e = e.next) {
2127	>	e.val = null;
2128	>	--delta;
2129	>	}
2130		setTabAt(tab, i, null);
2131	+	++i;
2132	+	}
2133	+	} finally {
2134	+	if (!f.casHash(fh | LOCKED, fh)) {
2135	+	f.hash = fh;
2136	+	synchronized (f) { f.notifyAll(); };
2137		}
2138		}
752	–	if (validated)
753	–	++i;
2139		}
2140		}
2141	<	counter.add(delta);
2141	>	if (delta != 0)
2142	>	counter.add(delta);
2143		}
2144
2145		/* ----------------Table Traversal -------------- */
#	Line 764 | Line 2150 | public class ConcurrentHashMapV8<K, V>
2150		*
2151		* At each step, the iterator snapshots the key ("nextKey") and
2152		* value ("nextVal") of a valid node (i.e., one that, at point of
2153	<	* snapshot, has a nonnull user value). Because val fields can
2153	>	* snapshot, has a non-null user value). Because val fields can
2154		* change (including to null, indicating deletion), field nextVal
2155		* might not be accurate at point of use, but still maintains the
2156		* weak consistency property of holding a value that was once
2157		* valid.
2158		*
2159		* Internal traversals directly access these fields, as in:
2160	<	* {@code while (it.next != null) { process(it.nextKey); it.advance(); }}
2160	>	* {@code while (it.advance() != null) { process(it.nextKey); }}
2161		*
2162	<	* Exported iterators (subclasses of ViewIterator) extract key,
2163	<	* value, or key-value pairs as return values of Iterator.next(),
2164	<	* and encapsulate the it.next check as hasNext();
2165	<	*
2166	<	* The iterator visits each valid node that was reachable upon
2167	<	* iterator construction once. It might miss some that were added
2168	<	* to a bin after the bin was visited, which is OK wrt consistency
2169	<	* guarantees. Maintaining this property in the face of possible
2170	<	* ongoing resizes requires a fair amount of bookkeeping state
2171	<	* that is difficult to optimize away amidst volatile accesses.
2172	<	* Even so, traversal maintains reasonable throughput.
2162	>	* Exported iterators must track whether the iterator has advanced
2163	>	* (in hasNext vs next) (by setting/checking/nulling field
2164	>	* nextVal), and then extract key, value, or key-value pairs as
2165	>	* return values of next().
2166	>	*
2167	>	* The iterator visits once each still-valid node that was
2168	>	* reachable upon iterator construction. It might miss some that
2169	>	* were added to a bin after the bin was visited, which is OK wrt
2170	>	* consistency guarantees. Maintaining this property in the face
2171	>	* of possible ongoing resizes requires a fair amount of
2172	>	* bookkeeping state that is difficult to optimize away amidst
2173	>	* volatile accesses.  Even so, traversal maintains reasonable
2174	>	* throughput.
2175		*
2176		* Normally, iteration proceeds bin-by-bin traversing lists.
2177		* However, if the table has been resized, then all future steps
#	Line 792 | Line 2180 | public class ConcurrentHashMapV8<K, V>
2180		* paranoically cope with potential sharing by users of iterators
2181		* across threads, iteration terminates if a bounds checks fails
2182		* for a table read.
795	–	*
796	–	* The range-based constructor enables creation of parallel
797	–	* range-splitting traversals. (Not yet implemented.)
2183		*/
2184	<	static class InternalIterator {
2184	>	static class InternalIterator<K,V> {
2185	>	final ConcurrentHashMapV8<K, V> map;
2186		Node next;           // the next entry to use
2187		Node last;           // the last entry used
2188		Object nextKey;      // cached key field of next
#	Line 804 | Line 2190 | public class ConcurrentHashMapV8<K, V>
2190		Node[] tab;          // current table; updated if resized
2191		int index;           // index of bin to use next
2192		int baseIndex;       // current index of initial table
2193	<	final int baseLimit; // index bound for initial table
2193	>	int baseLimit;       // index bound for initial table
2194		final int baseSize;  // initial table size
2195
2196		/** Creates iterator for all entries in the table. */
2197	<	InternalIterator(Node[] tab) {
2198	<	this.tab = tab;
2197	>	InternalIterator(ConcurrentHashMapV8<K, V> map) {
2198	>	this.tab = (this.map = map).table;
2199		baseLimit = baseSize = (tab == null) ? 0 : tab.length;
814	–	index = baseIndex = 0;
815	–	next = null;
816	–	advance();
817	–	}
818	–
819	–	/** Creates iterator for the given range of the table */
820	–	InternalIterator(Node[] tab, int lo, int hi) {
821	–	this.tab = tab;
822	–	baseSize = (tab == null) ? 0 : tab.length;
823	–	baseLimit = (hi <= baseSize) ? hi : baseSize;
824	–	index = baseIndex = lo;
825	–	next = null;
826	–	advance();
2200		}
2201
2202	<	/** Advances next. See above for explanation. */
2203	<	final void advance() {
2202	>	/** Creates iterator for clone() and split() methods */
2203	>	InternalIterator(InternalIterator<K,V> it, boolean split) {
2204	>	this.map = it.map;
2205	>	this.tab = it.tab;
2206	>	this.baseSize = it.baseSize;
2207	>	int lo = it.baseIndex;
2208	>	int hi = this.baseLimit = it.baseLimit;
2209	>	this.index = this.baseIndex =
2210	>	(split) ? (it.baseLimit = (lo + hi + 1) >>> 1) : lo;
2211	>	}
2212	>
2213	>	/**
2214	>	* Advances next; returns nextVal or null if terminated
2215	>	* See above for explanation.
2216	>	*/
2217	>	final Object advance() {
2218		Node e = last = next;
2219	+	Object ev = null;
2220		outer: do {
2221	<	if (e != null)                   // pass used or skipped node
2221	>	if (e != null)                  // advance past used/skipped node
2222		e = e.next;
2223	<	while (e == null) {              // get to next non-null bin
2224	<	Node[] t; int b, i, n;       // checks must use locals
2223	>	while (e == null) {             // get to next non-null bin
2224	>	Node[] t; int b, i, n; Object ek; // checks must use locals
2225		if ((b = baseIndex) >= baseLimit || (i = index) < 0 ||
2226		(t = tab) == null || i >= (n = t.length))
2227		break outer;
2228	<	else if ((e = tabAt(t, i)) != null && e.hash < 0)
2229	<	tab = (Node[])e.key;     // restarts due to null val
2230	<	else                         // visit upper slots if present
2231	<	index = (i += baseSize) < n ? i : (baseIndex = b + 1);
2228	>	else if ((e = tabAt(t, i)) != null && e.hash == MOVED) {
2229	>	if ((ek = e.key) instanceof TreeBin)
2230	>	e = ((TreeBin)ek).first;
2231	>	else {
2232	>	tab = (Node[])ek;
2233	>	continue;           // restarts due to null val
2234	>	}
2235	>	}                           // visit upper slots if present
2236	>	index = (i += baseSize) < n ? i : (baseIndex = b + 1);
2237		}
2238		nextKey = e.key;
2239	<	} while ((nextVal = e.val) == null); // skip deleted or special nodes
2239	>	} while ((ev = e.val) == null);    // skip deleted or special nodes
2240		next = e;
2241	+	return nextVal = ev;
2242	+	}
2243	+
2244	+	public final void remove() {
2245	+	if (nextVal == null)
2246	+	advance();
2247	+	Node e = last;
2248	+	if (e == null)
2249	+	throw new IllegalStateException();
2250	+	last = null;
2251	+	map.remove(e.key);
2252	+	}
2253	+
2254	+	public final boolean hasNext() {
2255	+	return nextVal != null || advance() != null;
2256		}
2257	+
2258	+	public final boolean hasMoreElements() { return hasNext(); }
2259		}
2260
2261		/* ---------------- Public operations -------------- */
#	Line 855 | Line 2265 | public class ConcurrentHashMapV8<K, V>
2265		*/
2266		public ConcurrentHashMapV8() {
2267		this.counter = new LongAdder();
858	–	this.targetCapacity = DEFAULT_CAPACITY;
2268		}
2269
2270		/**
#	Line 875 | Line 2284 | public class ConcurrentHashMapV8<K, V>
2284		MAXIMUM_CAPACITY :
2285		tableSizeFor(initialCapacity + (initialCapacity >>> 1) + 1));
2286		this.counter = new LongAdder();
2287	<	this.targetCapacity = cap;
2287	>	this.sizeCtl = cap;
2288		}
2289
2290		/**
#	Line 885 | Line 2294 | public class ConcurrentHashMapV8<K, V>
2294		*/
2295		public ConcurrentHashMapV8(Map<? extends K, ? extends V> m) {
2296		this.counter = new LongAdder();
2297	<	this.targetCapacity = DEFAULT_CAPACITY;
2298	<	putAll(m);
2297	>	this.sizeCtl = DEFAULT_CAPACITY;
2298	>	internalPutAll(m);
2299		}
2300
2301		/**
#	Line 933 | Line 2342 | public class ConcurrentHashMapV8<K, V>
2342		if (initialCapacity < concurrencyLevel)   // Use at least as many bins
2343		initialCapacity = concurrencyLevel;   // as estimated threads
2344		long size = (long)(1.0 + (long)initialCapacity / loadFactor);
2345	<	int cap =  ((size >= (long)MAXIMUM_CAPACITY) ?
2346	<	MAXIMUM_CAPACITY: tableSizeFor((int)size));
2345	>	int cap = ((size >= (long)MAXIMUM_CAPACITY) ?
2346	>	MAXIMUM_CAPACITY: tableSizeFor((int)size));
2347		this.counter = new LongAdder();
2348	<	this.targetCapacity = cap;
2348	>	this.sizeCtl = cap;
2349		}
2350
2351		/**
#	Line 956 | Line 2365 | public class ConcurrentHashMapV8<K, V>
2365		(int)n);
2366		}
2367
2368	+	final long longSize() { // accurate version of size needed for views
2369	+	long n = counter.sum();
2370	+	return (n < 0L) ? 0L : n;
2371	+	}
2372	+
2373		/**
2374		* Returns the value to which the specified key is mapped,
2375		* or {@code null} if this map contains no mapping for the key.
#	Line 1003 | Line 2417 | public class ConcurrentHashMapV8<K, V>
2417		if (value == null)
2418		throw new NullPointerException();
2419		Object v;
2420	<	InternalIterator it = new InternalIterator(table);
2421	<	while (it.next != null) {
2422	<	if ((v = it.nextVal) == value || value.equals(v))
2420	>	InternalIterator<K,V> it = new InternalIterator<K,V>(this);
2421	>	while ((v = it.advance()) != null) {
2422	>	if (v == value || value.equals(v))
2423		return true;
1010	–	it.advance();
2424		}
2425		return false;
2426		}
#	Line 1048 | Line 2461 | public class ConcurrentHashMapV8<K, V>
2461		public V put(K key, V value) {
2462		if (key == null || value == null)
2463		throw new NullPointerException();
2464	<	return (V)internalPut(key, value, true);
2464	>	return (V)internalPut(key, value);
2465		}
2466
2467		/**
#	Line 1062 | Line 2475 | public class ConcurrentHashMapV8<K, V>
2475		public V putIfAbsent(K key, V value) {
2476		if (key == null || value == null)
2477		throw new NullPointerException();
2478	<	return (V)internalPut(key, value, false);
2478	>	return (V)internalPutIfAbsent(key, value);
2479		}
2480
2481		/**
#	Line 1073 | Line 2486 | public class ConcurrentHashMapV8<K, V>
2486		* @param m mappings to be stored in this map
2487		*/
2488		public void putAll(Map<? extends K, ? extends V> m) {
2489	<	if (m == null)
1077	<	throw new NullPointerException();
1078	<	/*
1079	<	* If uninitialized, try to adjust targetCapacity to
1080	<	* accommodate the given number of elements.
1081	<	*/
1082	<	if (table == null) {
1083	<	int size = m.size();
1084	<	int cap = (size >= (MAXIMUM_CAPACITY >>> 1)) ? MAXIMUM_CAPACITY :
1085	<	tableSizeFor(size + (size >>> 1) + 1);
1086	<	if (cap > targetCapacity)
1087	<	targetCapacity = cap;
1088	<	}
1089	<	for (Map.Entry<? extends K, ? extends V> e : m.entrySet())
1090	<	put(e.getKey(), e.getValue());
2489	>	internalPutAll(m);
2490		}
2491
2492		/**
2493		* If the specified key is not already associated with a value,
2494	<	* computes its value using the given mappingFunction, and if
2495	<	* non-null, enters it into the map.  This is equivalent to
2496	<	*  <pre> {@code
2494	>	* computes its value using the given mappingFunction and enters
2495	>	* it into the map unless null.  This is equivalent to
2496	>	* <pre> {@code
2497		* if (map.containsKey(key))
2498		*   return map.get(key);
2499		* value = mappingFunction.map(key);
#	Line 1102 | Line 2501 | public class ConcurrentHashMapV8<K, V>
2501		*   map.put(key, value);
2502		* return value;}</pre>
2503		*
2504	<	* except that the action is performed atomically.  Some attempted
2505	<	* update operations on this map by other threads may be blocked
2506	<	* while computation is in progress, so the computation should be
2507	<	* short and simple, and must not attempt to update any other
2508	<	* mappings of this Map. The most appropriate usage is to
2504	>	* except that the action is performed atomically.  If the
2505	>	* function returns {@code null} no mapping is recorded. If the
2506	>	* function itself throws an (unchecked) exception, the exception
2507	>	* is rethrown to its caller, and no mapping is recorded.  Some
2508	>	* attempted update operations on this map by other threads may be
2509	>	* blocked while computation is in progress, so the computation
2510	>	* should be short and simple, and must not attempt to update any
2511	>	* other mappings of this Map. The most appropriate usage is to
2512		* construct a new object serving as an initial mapped value, or
2513		* memoized result, as in:
2514	+	*
2515		*  <pre> {@code
2516		* map.computeIfAbsent(key, new MappingFunction<K, V>() {
2517		*   public V map(K k) { return new Value(f(k)); }});}</pre>
#	Line 1116 | Line 2519 | public class ConcurrentHashMapV8<K, V>
2519		* @param key key with which the specified value is to be associated
2520		* @param mappingFunction the function to compute a value
2521		* @return the current (existing or computed) value associated with
2522	<	*         the specified key, or {@code null} if the computation
1120	<	*         returned {@code null}
2522	>	*         the specified key, or null if the computed value is null.
2523		* @throws NullPointerException if the specified key or mappingFunction
2524		*         is null
2525		* @throws IllegalStateException if the computation detectably
#	Line 1126 | Line 2528 | public class ConcurrentHashMapV8<K, V>
2528		* @throws RuntimeException or Error if the mappingFunction does so,
2529		*         in which case the mapping is left unestablished
2530		*/
2531	+	@SuppressWarnings("unchecked")
2532		public V computeIfAbsent(K key, MappingFunction<? super K, ? extends V> mappingFunction) {
2533		if (key == null || mappingFunction == null)
2534		throw new NullPointerException();
2535	<	return internalCompute(key, mappingFunction, false);
2535	>	return (V)internalComputeIfAbsent(key, mappingFunction);
2536		}
2537
2538		/**
2539	<	* Computes the value associated with the given key using the given
2540	<	* mappingFunction, and if non-null, enters it into the map.  This
2541	<	* is equivalent to
2539	>	* Computes a new mapping value given a key and
2540	>	* its current mapped value (or {@code null} if there is no current
2541	>	* mapping). This is equivalent to
2542		*  <pre> {@code
2543	<	* value = mappingFunction.map(key);
2544	<	* if (value != null)
2545	<	*   map.put(key, value);
2546	<	* else
2547	<	*   value = map.get(key);
2548	<	* return value;}</pre>
2549	<	*
2550	<	* except that the action is performed atomically.  Some attempted
2551	<	* update operations on this map by other threads may be blocked
2552	<	* while computation is in progress, so the computation should be
2553	<	* short and simple, and must not attempt to update any other
2554	<	* mappings of this Map.
2543	>	*   value = remappingFunction.remap(key, map.get(key));
2544	>	*   if (value != null)
2545	>	*     map.put(key, value);
2546	>	*   else
2547	>	*     map.remove(key);
2548	>	* }</pre>
2549	>	*
2550	>	* except that the action is performed atomically.  If the
2551	>	* function returns {@code null}, the mapping is removed.  If the
2552	>	* function itself throws an (unchecked) exception, the exception
2553	>	* is rethrown to its caller, and the current mapping is left
2554	>	* unchanged.  Some attempted update operations on this map by
2555	>	* other threads may be blocked while computation is in progress,
2556	>	* so the computation should be short and simple, and must not
2557	>	* attempt to update any other mappings of this Map. For example,
2558	>	* to either create or append new messages to a value mapping:
2559	>	*
2560	>	* <pre> {@code
2561	>	* Map<Key, String> map = ...;
2562	>	* final String msg = ...;
2563	>	* map.compute(key, new RemappingFunction<Key, String>() {
2564	>	*   public String remap(Key k, String v) {
2565	>	*    return (v == null) ? msg : v + msg;});}}</pre>
2566		*
2567		* @param key key with which the specified value is to be associated
2568	<	* @param mappingFunction the function to compute a value
2569	<	* @return the current value associated with
2570	<	*         the specified key, or {@code null} if the computation
2571	<	*         returned {@code null} and the value was not otherwise present
1158	<	* @throws NullPointerException if the specified key or mappingFunction
2568	>	* @param remappingFunction the function to compute a value
2569	>	* @return the new value associated with
2570	>	*         the specified key, or null if none.
2571	>	* @throws NullPointerException if the specified key or remappingFunction
2572		*         is null
2573		* @throws IllegalStateException if the computation detectably
2574		*         attempts a recursive update to this map that would
2575		*         otherwise never complete
2576	<	* @throws RuntimeException or Error if the mappingFunction does so,
2576	>	* @throws RuntimeException or Error if the remappingFunction does so,
2577		*         in which case the mapping is unchanged
2578		*/
2579	<	public V compute(K key, MappingFunction<? super K, ? extends V> mappingFunction) {
2580	<	if (key == null || mappingFunction == null)
2579	>	@SuppressWarnings("unchecked")
2580	>	public V compute(K key, RemappingFunction<? super K, V> remappingFunction) {
2581	>	if (key == null || remappingFunction == null)
2582		throw new NullPointerException();
2583	<	return internalCompute(key, mappingFunction, true);
2583	>	return (V)internalCompute(key, remappingFunction);
2584		}
2585
2586		/**
#	Line 1314 | Line 2728 | public class ConcurrentHashMapV8<K, V>
2728		}
2729
2730		/**
2731	+	* Returns a partionable iterator of the keys in this map.
2732	+	*
2733	+	* @return a partionable iterator of the keys in this map
2734	+	*/
2735	+	public Spliterator<K> keySpliterator() {
2736	+	return new KeyIterator<K,V>(this);
2737	+	}
2738	+
2739	+	/**
2740	+	* Returns a partionable iterator of the values in this map.
2741	+	*
2742	+	* @return a partionable iterator of the values in this map
2743	+	*/
2744	+	public Spliterator<V> valueSpliterator() {
2745	+	return new ValueIterator<K,V>(this);
2746	+	}
2747	+
2748	+	/**
2749	+	* Returns a partionable iterator of the entries in this map.
2750	+	*
2751	+	* @return a partionable iterator of the entries in this map
2752	+	*/
2753	+	public Spliterator<Map.Entry<K,V>> entrySpliterator() {
2754	+	return new EntryIterator<K,V>(this);
2755	+	}
2756	+
2757	+	/**
2758		* Returns the hash code value for this {@link Map}, i.e.,
2759		* the sum of, for each key-value pair in the map,
2760		* {@code key.hashCode() ^ value.hashCode()}.
#	Line 1322 | Line 2763 | public class ConcurrentHashMapV8<K, V>
2763		*/
2764		public int hashCode() {
2765		int h = 0;
2766	<	InternalIterator it = new InternalIterator(table);
2767	<	while (it.next != null) {
2768	<	h += it.nextKey.hashCode() ^ it.nextVal.hashCode();
2769	<	it.advance();
2766	>	InternalIterator<K,V> it = new InternalIterator<K,V>(this);
2767	>	Object v;
2768	>	while ((v = it.advance()) != null) {
2769	>	h += it.nextKey.hashCode() ^ v.hashCode();
2770		}
2771		return h;
2772		}
#	Line 1342 | Line 2783 | public class ConcurrentHashMapV8<K, V>
2783		* @return a string representation of this map
2784		*/
2785		public String toString() {
2786	<	InternalIterator it = new InternalIterator(table);
2786	>	InternalIterator<K,V> it = new InternalIterator<K,V>(this);
2787		StringBuilder sb = new StringBuilder();
2788		sb.append('{');
2789	<	if (it.next != null) {
2789	>	Object v;
2790	>	if ((v = it.advance()) != null) {
2791		for (;;) {
2792	<	Object k = it.nextKey, v = it.nextVal;
2792	>	Object k = it.nextKey;
2793		sb.append(k == this ? "(this Map)" : k);
2794		sb.append('=');
2795		sb.append(v == this ? "(this Map)" : v);
2796	<	it.advance();
1355	<	if (it.next == null)
2796	>	if ((v = it.advance()) == null)
2797		break;
2798		sb.append(',').append(' ');
2799		}
#	Line 1375 | Line 2816 | public class ConcurrentHashMapV8<K, V>
2816		if (!(o instanceof Map))
2817		return false;
2818		Map<?,?> m = (Map<?,?>) o;
2819	<	InternalIterator it = new InternalIterator(table);
2820	<	while (it.next != null) {
2821	<	Object val = it.nextVal;
2819	>	InternalIterator<K,V> it = new InternalIterator<K,V>(this);
2820	>	Object val;
2821	>	while ((val = it.advance()) != null) {
2822		Object v = m.get(it.nextKey);
2823		if (v == null || (v != val && !v.equals(val)))
2824		return false;
1384	–	it.advance();
2825		}
2826		for (Map.Entry<?,?> e : m.entrySet()) {
2827		Object mk, mv, v;
#	Line 1397 | Line 2837 | public class ConcurrentHashMapV8<K, V>
2837
2838		/* ----------------Iterators -------------- */
2839
2840	<	/**
2841	<	* Base class for key, value, and entry iterators.  Adds a map
2842	<	* reference to InternalIterator to support Iterator.remove.
2843	<	*/
2844	<	static abstract class ViewIterator<K,V> extends InternalIterator {
1405	<	final ConcurrentHashMapV8<K, V> map;
1406	<	ViewIterator(ConcurrentHashMapV8<K, V> map) {
1407	<	super(map.table);
1408	<	this.map = map;
2840	>	static final class KeyIterator<K,V> extends InternalIterator<K,V>
2841	>	implements Spliterator<K>, Enumeration<K> {
2842	>	KeyIterator(ConcurrentHashMapV8<K, V> map) { super(map); }
2843	>	KeyIterator(InternalIterator<K,V> it, boolean split) {
2844	>	super(it, split);
2845		}
2846	<
2847	<	public final void remove() {
1412	<	if (last == null)
2846	>	public KeyIterator<K,V> split() {
2847	>	if (last != null || (next != null && nextVal == null))
2848		throw new IllegalStateException();
2849	<	map.remove(last.key);
2850	<	last = null;
2849	>	return new KeyIterator<K,V>(this, true);
2850	>	}
2851	>	public KeyIterator<K,V> clone() {
2852	>	if (last != null || (next != null && nextVal == null))
2853	>	throw new IllegalStateException();
2854	>	return new KeyIterator<K,V>(this, false);
2855		}
1417	–
1418	–	public final boolean hasNext()         { return next != null; }
1419	–	public final boolean hasMoreElements() { return next != null; }
1420	–	}
1421	–
1422	–	static final class KeyIterator<K,V> extends ViewIterator<K,V>
1423	–	implements Iterator<K>, Enumeration<K> {
1424	–	KeyIterator(ConcurrentHashMapV8<K, V> map) { super(map); }
2856
2857		@SuppressWarnings("unchecked")
2858		public final K next() {
2859	<	if (next == null)
2859	>	if (nextVal == null && advance() == null)
2860		throw new NoSuchElementException();
2861		Object k = nextKey;
2862	<	advance();
2863	<	return (K)k;
2862	>	nextVal = null;
2863	>	return (K) k;
2864		}
2865
2866		public final K nextElement() { return next(); }
2867		}
2868
2869	<	static final class ValueIterator<K,V> extends ViewIterator<K,V>
2870	<	implements Iterator<V>, Enumeration<V> {
2869	>	static final class ValueIterator<K,V> extends InternalIterator<K,V>
2870	>	implements Spliterator<V>, Enumeration<V> {
2871		ValueIterator(ConcurrentHashMapV8<K, V> map) { super(map); }
2872	+	ValueIterator(InternalIterator<K,V> it, boolean split) {
2873	+	super(it, split);
2874	+	}
2875	+	public ValueIterator<K,V> split() {
2876	+	if (last != null || (next != null && nextVal == null))
2877	+	throw new IllegalStateException();
2878	+	return new ValueIterator<K,V>(this, true);
2879	+	}
2880	+
2881	+	public ValueIterator<K,V> clone() {
2882	+	if (last != null || (next != null && nextVal == null))
2883	+	throw new IllegalStateException();
2884	+	return new ValueIterator<K,V>(this, false);
2885	+	}
2886
2887		@SuppressWarnings("unchecked")
2888		public final V next() {
2889	<	if (next == null)
2889	>	Object v;
2890	>	if ((v = nextVal) == null && (v = advance()) == null)
2891		throw new NoSuchElementException();
2892	<	Object v = nextVal;
2893	<	advance();
1448	<	return (V)v;
2892	>	nextVal = null;
2893	>	return (V) v;
2894		}
2895
2896		public final V nextElement() { return next(); }
2897		}
2898
2899	<	static final class EntryIterator<K,V> extends ViewIterator<K,V>
2900	<	implements Iterator<Map.Entry<K,V>> {
2899	>	static final class EntryIterator<K,V> extends InternalIterator<K,V>
2900	>	implements Spliterator<Map.Entry<K,V>> {
2901		EntryIterator(ConcurrentHashMapV8<K, V> map) { super(map); }
2902	+	EntryIterator(InternalIterator<K,V> it, boolean split) {
2903	+	super(it, split);
2904	+	}
2905	+	public EntryIterator<K,V> split() {
2906	+	if (last != null || (next != null && nextVal == null))
2907	+	throw new IllegalStateException();
2908	+	return new EntryIterator<K,V>(this, true);
2909	+	}
2910	+	public EntryIterator<K,V> clone() {
2911	+	if (last != null || (next != null && nextVal == null))
2912	+	throw new IllegalStateException();
2913	+	return new EntryIterator<K,V>(this, false);
2914	+	}
2915
2916		@SuppressWarnings("unchecked")
2917		public final Map.Entry<K,V> next() {
2918	<	if (next == null)
2918	>	Object v;
2919	>	if ((v = nextVal) == null && (v = advance()) == null)
2920		throw new NoSuchElementException();
2921		Object k = nextKey;
2922	<	Object v = nextVal;
2923	<	advance();
1465	<	return new WriteThroughEntry<K,V>(map, (K)k, (V)v);
2922	>	nextVal = null;
2923	>	return new MapEntry<K,V>((K)k, (V)v, map);
2924		}
2925		}
2926
2927		/**
2928	<	* Custom Entry class used by EntryIterator.next(), that relays
1471	<	* setValue changes to the underlying map.
2928	>	* Exported Entry for iterators
2929		*/
2930	<	static final class WriteThroughEntry<K,V> implements Map.Entry<K, V> {
1474	<	final ConcurrentHashMapV8<K, V> map;
2930	>	static final class MapEntry<K,V> implements Map.Entry<K, V> {
2931		final K key; // non-null
2932		V val;       // non-null
2933	<	WriteThroughEntry(ConcurrentHashMapV8<K, V> map, K key, V val) {
2934	<	this.map = map; this.key = key; this.val = val;
2933	>	final ConcurrentHashMapV8<K, V> map;
2934	>	MapEntry(K key, V val, ConcurrentHashMapV8<K, V> map) {
2935	>	this.key = key;
2936	>	this.val = val;
2937	>	this.map = map;
2938		}
1480	–
2939		public final K getKey()       { return key; }
2940		public final V getValue()     { return val; }
2941		public final int hashCode()   { return key.hashCode() ^ val.hashCode(); }
#	Line 1494 | Line 2952 | public class ConcurrentHashMapV8<K, V>
2952
2953		/**
2954		* Sets our entry's value and writes through to the map. The
2955	<	* value to return is somewhat arbitrary here. Since a
2956	<	* WriteThroughEntry does not necessarily track asynchronous
2957	<	* changes, the most recent "previous" value could be
2958	<	* different from what we return (or could even have been
2959	<	* removed in which case the put will re-establish). We do not
1502	<	* and cannot guarantee more.
2955	>	* value to return is somewhat arbitrary here. Since a we do
2956	>	* not necessarily track asynchronous changes, the most recent
2957	>	* "previous" value could be different from what we return (or
2958	>	* could even have been removed in which case the put will
2959	>	* re-establish). We do not and cannot guarantee more.
2960		*/
2961		public final V setValue(V value) {
2962		if (value == null) throw new NullPointerException();
#	Line 1512 | Line 2969 | public class ConcurrentHashMapV8<K, V>
2969
2970		/* ----------------Views -------------- */
2971
2972	<	/*
2973	<	* These currently just extend java.util.AbstractX classes, but
1517	<	* may need a new custom base to support partitioned traversal.
2972	>	/**
2973	>	* Base class for views.
2974		*/
2975	<
1520	<	static final class KeySet<K,V> extends AbstractSet<K> {
2975	>	static abstract class MapView<K, V> {
2976		final ConcurrentHashMapV8<K, V> map;
2977	<	KeySet(ConcurrentHashMapV8<K, V> map)   { this.map = map; }
1523	<
2977	>	MapView(ConcurrentHashMapV8<K, V> map)  { this.map = map; }
2978		public final int size()                 { return map.size(); }
2979		public final boolean isEmpty()          { return map.isEmpty(); }
2980		public final void clear()               { map.clear(); }
2981	+
2982	+	// implementations below rely on concrete classes supplying these
2983	+	abstract public Iterator<?> iterator();
2984	+	abstract public boolean contains(Object o);
2985	+	abstract public boolean remove(Object o);
2986	+
2987	+	private static final String oomeMsg = "Required array size too large";
2988	+
2989	+	public final Object[] toArray() {
2990	+	long sz = map.longSize();
2991	+	if (sz > (long)(MAX_ARRAY_SIZE))
2992	+	throw new OutOfMemoryError(oomeMsg);
2993	+	int n = (int)sz;
2994	+	Object[] r = new Object[n];
2995	+	int i = 0;
2996	+	Iterator<?> it = iterator();
2997	+	while (it.hasNext()) {
2998	+	if (i == n) {
2999	+	if (n >= MAX_ARRAY_SIZE)
3000	+	throw new OutOfMemoryError(oomeMsg);
3001	+	if (n >= MAX_ARRAY_SIZE - (MAX_ARRAY_SIZE >>> 1) - 1)
3002	+	n = MAX_ARRAY_SIZE;
3003	+	else
3004	+	n += (n >>> 1) + 1;
3005	+	r = Arrays.copyOf(r, n);
3006	+	}
3007	+	r[i++] = it.next();
3008	+	}
3009	+	return (i == n) ? r : Arrays.copyOf(r, i);
3010	+	}
3011	+
3012	+	@SuppressWarnings("unchecked")
3013	+	public final <T> T[] toArray(T[] a) {
3014	+	long sz = map.longSize();
3015	+	if (sz > (long)(MAX_ARRAY_SIZE))
3016	+	throw new OutOfMemoryError(oomeMsg);
3017	+	int m = (int)sz;
3018	+	T[] r = (a.length >= m) ? a :
3019	+	(T[])java.lang.reflect.Array
3020	+	.newInstance(a.getClass().getComponentType(), m);
3021	+	int n = r.length;
3022	+	int i = 0;
3023	+	Iterator<?> it = iterator();
3024	+	while (it.hasNext()) {
3025	+	if (i == n) {
3026	+	if (n >= MAX_ARRAY_SIZE)
3027	+	throw new OutOfMemoryError(oomeMsg);
3028	+	if (n >= MAX_ARRAY_SIZE - (MAX_ARRAY_SIZE >>> 1) - 1)
3029	+	n = MAX_ARRAY_SIZE;
3030	+	else
3031	+	n += (n >>> 1) + 1;
3032	+	r = Arrays.copyOf(r, n);
3033	+	}
3034	+	r[i++] = (T)it.next();
3035	+	}
3036	+	if (a == r && i < n) {
3037	+	r[i] = null; // null-terminate
3038	+	return r;
3039	+	}
3040	+	return (i == n) ? r : Arrays.copyOf(r, i);
3041	+	}
3042	+
3043	+	public final int hashCode() {
3044	+	int h = 0;
3045	+	for (Iterator<?> it = iterator(); it.hasNext();)
3046	+	h += it.next().hashCode();
3047	+	return h;
3048	+	}
3049	+
3050	+	public final String toString() {
3051	+	StringBuilder sb = new StringBuilder();
3052	+	sb.append('[');
3053	+	Iterator<?> it = iterator();
3054	+	if (it.hasNext()) {
3055	+	for (;;) {
3056	+	Object e = it.next();
3057	+	sb.append(e == this ? "(this Collection)" : e);
3058	+	if (!it.hasNext())
3059	+	break;
3060	+	sb.append(',').append(' ');
3061	+	}
3062	+	}
3063	+	return sb.append(']').toString();
3064	+	}
3065	+
3066	+	public final boolean containsAll(Collection<?> c) {
3067	+	if (c != this) {
3068	+	for (Iterator<?> it = c.iterator(); it.hasNext();) {
3069	+	Object e = it.next();
3070	+	if (e == null || !contains(e))
3071	+	return false;
3072	+	}
3073	+	}
3074	+	return true;
3075	+	}
3076	+
3077	+	public final boolean removeAll(Collection<?> c) {
3078	+	boolean modified = false;
3079	+	for (Iterator<?> it = iterator(); it.hasNext();) {
3080	+	if (c.contains(it.next())) {
3081	+	it.remove();
3082	+	modified = true;
3083	+	}
3084	+	}
3085	+	return modified;
3086	+	}
3087	+
3088	+	public final boolean retainAll(Collection<?> c) {
3089	+	boolean modified = false;
3090	+	for (Iterator<?> it = iterator(); it.hasNext();) {
3091	+	if (!c.contains(it.next())) {
3092	+	it.remove();
3093	+	modified = true;
3094	+	}
3095	+	}
3096	+	return modified;
3097	+	}
3098	+
3099	+	}
3100	+
3101	+	static final class KeySet<K,V> extends MapView<K,V> implements Set<K> {
3102	+	KeySet(ConcurrentHashMapV8<K, V> map)   { super(map); }
3103		public final boolean contains(Object o) { return map.containsKey(o); }
3104		public final boolean remove(Object o)   { return map.remove(o) != null; }
3105		public final Iterator<K> iterator() {
3106		return new KeyIterator<K,V>(map);
3107		}
3108	+	public final boolean add(K e) {
3109	+	throw new UnsupportedOperationException();
3110	+	}
3111	+	public final boolean addAll(Collection<? extends K> c) {
3112	+	throw new UnsupportedOperationException();
3113	+	}
3114	+	public boolean equals(Object o) {
3115	+	Set<?> c;
3116	+	return ((o instanceof Set) &&
3117	+	((c = (Set<?>)o) == this ||
3118	+	(containsAll(c) && c.containsAll(this))));
3119	+	}
3120		}
3121
3122	<	static final class Values<K,V> extends AbstractCollection<V> {
3123	<	final ConcurrentHashMapV8<K, V> map;
3124	<	Values(ConcurrentHashMapV8<K, V> map)   { this.map = map; }
1537	<
1538	<	public final int size()                 { return map.size(); }
1539	<	public final boolean isEmpty()          { return map.isEmpty(); }
1540	<	public final void clear()               { map.clear(); }
3122	>	static final class Values<K,V> extends MapView<K,V>
3123	>	implements Collection<V> {
3124	>	Values(ConcurrentHashMapV8<K, V> map)   { super(map); }
3125		public final boolean contains(Object o) { return map.containsValue(o); }
3126	+	public final boolean remove(Object o) {
3127	+	if (o != null) {
3128	+	Iterator<V> it = new ValueIterator<K,V>(map);
3129	+	while (it.hasNext()) {
3130	+	if (o.equals(it.next())) {
3131	+	it.remove();
3132	+	return true;
3133	+	}
3134	+	}
3135	+	}
3136	+	return false;
3137	+	}
3138		public final Iterator<V> iterator() {
3139		return new ValueIterator<K,V>(map);
3140		}
3141	<	}
3142	<
3143	<	static final class EntrySet<K,V> extends AbstractSet<Map.Entry<K,V>> {
3144	<	final ConcurrentHashMapV8<K, V> map;
3145	<	EntrySet(ConcurrentHashMapV8<K, V> map) { this.map = map; }
1550	<
1551	<	public final int size()                 { return map.size(); }
1552	<	public final boolean isEmpty()          { return map.isEmpty(); }
1553	<	public final void clear()               { map.clear(); }
1554	<	public final Iterator<Map.Entry<K,V>> iterator() {
1555	<	return new EntryIterator<K,V>(map);
3141	>	public final boolean add(V e) {
3142	>	throw new UnsupportedOperationException();
3143	>	}
3144	>	public final boolean addAll(Collection<? extends V> c) {
3145	>	throw new UnsupportedOperationException();
3146		}
3147	+	}
3148
3149	+	static final class EntrySet<K,V> extends MapView<K,V>
3150	+	implements Set<Map.Entry<K,V>> {
3151	+	EntrySet(ConcurrentHashMapV8<K, V> map) { super(map); }
3152		public final boolean contains(Object o) {
3153		Object k, v, r; Map.Entry<?,?> e;
3154		return ((o instanceof Map.Entry) &&
#	Line 1563 | Line 3157 | public class ConcurrentHashMapV8<K, V>
3157		(v = e.getValue()) != null &&
3158		(v == r || v.equals(r)));
3159		}
1566	–
3160		public final boolean remove(Object o) {
3161		Object k, v; Map.Entry<?,?> e;
3162		return ((o instanceof Map.Entry) &&
#	Line 1571 | Line 3164 | public class ConcurrentHashMapV8<K, V>
3164		(v = e.getValue()) != null &&
3165		map.remove(k, v));
3166		}
3167	+	public final Iterator<Map.Entry<K,V>> iterator() {
3168	+	return new EntryIterator<K,V>(map);
3169	+	}
3170	+	public final boolean add(Entry<K,V> e) {
3171	+	throw new UnsupportedOperationException();
3172	+	}
3173	+	public final boolean addAll(Collection<? extends Entry<K,V>> c) {
3174	+	throw new UnsupportedOperationException();
3175	+	}
3176	+	public boolean equals(Object o) {
3177	+	Set<?> c;
3178	+	return ((o instanceof Set) &&
3179	+	((c = (Set<?>)o) == this ||
3180	+	(containsAll(c) && c.containsAll(this))));
3181	+	}
3182		}
3183
3184		/* ---------------- Serialization Support -------------- */
#	Line 1604 | Line 3212 | public class ConcurrentHashMapV8<K, V>
3212		segments[i] = new Segment<K,V>(LOAD_FACTOR);
3213		}
3214		s.defaultWriteObject();
3215	<	InternalIterator it = new InternalIterator(table);
3216	<	while (it.next != null) {
3215	>	InternalIterator<K,V> it = new InternalIterator<K,V>(this);
3216	>	Object v;
3217	>	while ((v = it.advance()) != null) {
3218		s.writeObject(it.nextKey);
3219	<	s.writeObject(it.nextVal);
1611	<	it.advance();
3219	>	s.writeObject(v);
3220		}
3221		s.writeObject(null);
3222		s.writeObject(null);
#	Line 1626 | Line 3234 | public class ConcurrentHashMapV8<K, V>
3234		this.segments = null; // unneeded
3235		// initialize transient final field
3236		UNSAFE.putObjectVolatile(this, counterOffset, new LongAdder());
1629	–	this.targetCapacity = DEFAULT_CAPACITY;
3237
3238		// Create all nodes, then place in table once size is known
3239		long size = 0L;
#	Line 1635 | Line 3242 | public class ConcurrentHashMapV8<K, V>
3242		K k = (K) s.readObject();
3243		V v = (V) s.readObject();
3244		if (k != null && v != null) {
3245	<	p = new Node(spread(k.hashCode()), k, v, p);
3245	>	int h = spread(k.hashCode());
3246	>	p = new Node(h, k, v, p);
3247		++size;
3248		}
3249		else
#	Line 1643 | Line 3251 | public class ConcurrentHashMapV8<K, V>
3251		}
3252		if (p != null) {
3253		boolean init = false;
3254	<	if (resizing == 0 &&
3255	<	UNSAFE.compareAndSwapInt(this, resizingOffset, 0, 1)) {
3254	>	int n;
3255	>	if (size >= (long)(MAXIMUM_CAPACITY >>> 1))
3256	>	n = MAXIMUM_CAPACITY;
3257	>	else {
3258	>	int sz = (int)size;
3259	>	n = tableSizeFor(sz + (sz >>> 1) + 1);
3260	>	}
3261	>	int sc = sizeCtl;
3262	>	boolean collide = false;
3263	>	if (n > sc &&
3264	>	UNSAFE.compareAndSwapInt(this, sizeCtlOffset, sc, -1)) {
3265		try {
3266		if (table == null) {
3267		init = true;
1651	–	int n;
1652	–	if (size >= (long)(MAXIMUM_CAPACITY >>> 1))
1653	–	n = MAXIMUM_CAPACITY;
1654	–	else {
1655	–	int sz = (int)size;
1656	–	n = tableSizeFor(sz + (sz >>> 1) + 1);
1657	–	}
1658	–	threshold = n - (n >>> 2) - THRESHOLD_OFFSET;
3268		Node[] tab = new Node[n];
3269		int mask = n - 1;
3270		while (p != null) {
3271		int j = p.hash & mask;
3272		Node next = p.next;
3273	<	p.next = tabAt(tab, j);
3273	>	Node q = p.next = tabAt(tab, j);
3274		setTabAt(tab, j, p);
3275	+	if (!collide && q != null && q.hash == p.hash)
3276	+	collide = true;
3277		p = next;
3278		}
3279		table = tab;
3280		counter.add(size);
3281	+	sc = n - (n >>> 2);
3282		}
3283		} finally {
3284	<	resizing = 0;
3284	>	sizeCtl = sc;
3285	>	}
3286	>	if (collide) { // rescan and convert to TreeBins
3287	>	Node[] tab = table;
3288	>	for (int i = 0; i < tab.length; ++i) {
3289	>	int c = 0;
3290	>	for (Node e = tabAt(tab, i); e != null; e = e.next) {
3291	>	if (++c > TREE_THRESHOLD &&
3292	>	(e.key instanceof Comparable)) {
3293	>	replaceWithTreeBin(tab, i, e.key);
3294	>	break;
3295	>	}
3296	>	}
3297	>	}
3298		}
3299		}
3300		if (!init) { // Can only happen if unsafely published.
3301		while (p != null) {
3302	<	internalPut(p.key, p.val, true);
3302	>	internalPut(p.key, p.val);
3303		p = p.next;
3304		}
3305		}
#	Line 1684 | Line 3309 | public class ConcurrentHashMapV8<K, V>
3309		// Unsafe mechanics
3310		private static final sun.misc.Unsafe UNSAFE;
3311		private static final long counterOffset;
3312	<	private static final long resizingOffset;
3312	>	private static final long sizeCtlOffset;
3313		private static final long ABASE;
3314		private static final int ASHIFT;
3315
#	Line 1695 | Line 3320 | public class ConcurrentHashMapV8<K, V>
3320		Class<?> k = ConcurrentHashMapV8.class;
3321		counterOffset = UNSAFE.objectFieldOffset
3322		(k.getDeclaredField("counter"));
3323	<	resizingOffset = UNSAFE.objectFieldOffset
3324	<	(k.getDeclaredField("resizing"));
3323	>	sizeCtlOffset = UNSAFE.objectFieldOffset
3324	>	(k.getDeclaredField("sizeCtl"));
3325		Class<?> sc = Node[].class;
3326		ABASE = UNSAFE.arrayBaseOffset(sc);
3327		ss = UNSAFE.arrayIndexScale(sc);

Diff Legend

–	Removed lines
+	Added lines
<	Changed lines
>	Changed lines