ViewVC Help

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

root/jsr166/jsr166/src/jsr166y/LinkedTransferQueue.java

Comparing jsr166/src/jsr166y/LinkedTransferQueue.java (file contents):
Revision 1.46 by jsr166, Thu Oct 22 08:19:44 2009 UTC vs.
Revision 1.87 by jsr166, Fri Jun 3 14:17:10 2011 UTC

#	Line 1 | Line 1
1		/*
2		* Written by Doug Lea with assistance from members of JCP JSR-166
3		* Expert Group and released to the public domain, as explained at
4	<	* http://creativecommons.org/licenses/publicdomain
4	>	* http://creativecommons.org/publicdomain/zero/1.0/
5		*/
6
7		package jsr166y;
8
9	–	import java.util.concurrent.*;
10	–
9		import java.util.AbstractQueue;
10		import java.util.Collection;
13	–	import java.util.ConcurrentModificationException;
11		import java.util.Iterator;
12		import java.util.NoSuchElementException;
13		import java.util.Queue;
14	+	import java.util.concurrent.TimeUnit;
15		import java.util.concurrent.locks.LockSupport;
16	+
17		/**
18		* An unbounded {@link TransferQueue} based on linked nodes.
19		* This queue orders elements FIFO (first-in-first-out) with respect
#	Line 23 | Line 22 | import java.util.concurrent.locks.LockSu
22		* producer.  The <em>tail</em> of the queue is that element that has
23		* been on the queue the shortest time for some producer.
24		*
25	<	* <p>Beware that, unlike in most collections, the {@code size}
26	<	* method is <em>NOT</em> a constant-time operation. Because of the
25	>	* <p>Beware that, unlike in most collections, the {@code size} method
26	>	* is <em>NOT</em> a constant-time operation. Because of the
27		* asynchronous nature of these queues, determining the current number
28	<	* of elements requires a traversal of the elements.
28	>	* of elements requires a traversal of the elements, and so may report
29	>	* inaccurate results if this collection is modified during traversal.
30	>	* Additionally, the bulk operations {@code addAll},
31	>	* {@code removeAll}, {@code retainAll}, {@code containsAll},
32	>	* {@code equals}, and {@code toArray} are <em>not</em> guaranteed
33	>	* to be performed atomically. For example, an iterator operating
34	>	* concurrently with an {@code addAll} operation might view only some
35	>	* of the added elements.
36		*
37		* <p>This class and its iterator implement all of the
38		* <em>optional</em> methods of the {@link Collection} and {@link
#	Line 105 | Line 111 | public class LinkedTransferQueue<E> exte
111		* successful atomic operation per enq/deq pair. But it also
112		* enables lower cost variants of queue maintenance mechanics. (A
113		* variation of this idea applies even for non-dual queues that
114	<	* support deletion of embedded elements, such as
114	>	* support deletion of interior elements, such as
115		* j.u.c.ConcurrentLinkedQueue.)
116		*
117	<	* Once a node is matched, its item can never again change.  We
118	<	* may thus arrange that the linked list of them contains a prefix
119	<	* of zero or more matched nodes, followed by a suffix of zero or
120	<	* more unmatched nodes. (Note that we allow both the prefix and
121	<	* suffix to be zero length, which in turn means that we do not
122	<	* use a dummy header.)  If we were not concerned with either time
123	<	* or space efficiency, we could correctly perform enqueue and
124	<	* dequeue operations by traversing from a pointer to the initial
125	<	* node; CASing the item of the first unmatched node on match and
126	<	* CASing the next field of the trailing node on appends.  While
127	<	* this would be a terrible idea in itself, it does have the
128	<	* benefit of not requiring ANY atomic updates on head/tail
129	<	* fields.
117	>	* Once a node is matched, its match status can never again
118	>	* change.  We may thus arrange that the linked list of them
119	>	* contain a prefix of zero or more matched nodes, followed by a
120	>	* suffix of zero or more unmatched nodes. (Note that we allow
121	>	* both the prefix and suffix to be zero length, which in turn
122	>	* means that we do not use a dummy header.)  If we were not
123	>	* concerned with either time or space efficiency, we could
124	>	* correctly perform enqueue and dequeue operations by traversing
125	>	* from a pointer to the initial node; CASing the item of the
126	>	* first unmatched node on match and CASing the next field of the
127	>	* trailing node on appends. (Plus some special-casing when
128	>	* initially empty).  While this would be a terrible idea in
129	>	* itself, it does have the benefit of not requiring ANY atomic
130	>	* updates on head/tail fields.
131		*
132		* We introduce here an approach that lies between the extremes of
133	<	* never versus always updating queue (head and tail) pointers
134	<	* that reflects the tradeoff of sometimes requiring extra traversal
135	<	* steps to locate the first and/or last unmatched nodes, versus
136	<	* the reduced overhead and contention of fewer updates to queue
137	<	* pointers. For example, a possible snapshot of a queue is:
133	>	* never versus always updating queue (head and tail) pointers.
134	>	* This offers a tradeoff between sometimes requiring extra
135	>	* traversal steps to locate the first and/or last unmatched
136	>	* nodes, versus the reduced overhead and contention of fewer
137	>	* updates to queue pointers. For example, a possible snapshot of
138	>	* a queue is:
139		*
140		*  head           tail
141		*    |              |
#	Line 139 | Line 147 | public class LinkedTransferQueue<E> exte
147		* similarly for "tail") is an empirical matter. We have found
148		* that using very small constants in the range of 1-3 work best
149		* over a range of platforms. Larger values introduce increasing
150	<	* costs of cache misses and risks of long traversal chains.
150	>	* costs of cache misses and risks of long traversal chains, while
151	>	* smaller values increase CAS contention and overhead.
152		*
153		* Dual queues with slack differ from plain M&S dual queues by
154		* virtue of only sometimes updating head or tail pointers when
#	Line 158 | Line 167 | public class LinkedTransferQueue<E> exte
167		* targets.  Even when using very small slack values, this
168		* approach works well for dual queues because it allows all
169		* operations up to the point of matching or appending an item
170	<	* (hence potentially releasing another thread) to be read-only,
171	<	* thus not introducing any further contention. As described
172	<	* below, we implement this by performing slack maintenance
173	<	* retries only after these points.
170	>	* (hence potentially allowing progress by another thread) to be
171	>	* read-only, thus not introducing any further contention. As
172	>	* described below, we implement this by performing slack
173	>	* maintenance retries only after these points.
174		*
175		* As an accompaniment to such techniques, traversal overhead can
176		* be further reduced without increasing contention of head
177	<	* pointer updates.  During traversals, threads may sometimes
178	<	* shortcut the "next" link path from the current "head" node to
179	<	* be closer to the currently known first unmatched node. Again,
180	<	* this may be triggered with using thresholds or randomization.
177	>	* pointer updates: Threads may sometimes shortcut the "next" link
178	>	* path from the current "head" node to be closer to the currently
179	>	* known first unmatched node, and similarly for tail. Again, this
180	>	* may be triggered with using thresholds or randomization.
181		*
182		* These ideas must be further extended to avoid unbounded amounts
183		* of costly-to-reclaim garbage caused by the sequential "next"
#	Line 196 | Line 205 | public class LinkedTransferQueue<E> exte
205		* mechanics because an update may leave head at a detached node.
206		* And while direct writes are possible for tail updates, they
207		* increase the risk of long retraversals, and hence long garbage
208	<	* chains which can be much more costly than is worthwhile
208	>	* chains, which can be much more costly than is worthwhile
209		* considering that the cost difference of performing a CAS vs
210		* write is smaller when they are not triggered on each operation
211		* (especially considering that writes and CASes equally require
212		* additional GC bookkeeping ("write barriers") that are sometimes
213		* more costly than the writes themselves because of contention).
214		*
206	–	* Removal of internal nodes (due to timed out or interrupted
207	–	* waits, or calls to remove or Iterator.remove) uses a scheme
208	–	* roughly similar to that in Scherer, Lea, and Scott
209	–	* SynchronousQueue. Given a predecessor, we can unsplice any node
210	–	* except the (actual) tail of the queue. To avoid build-up of
211	–	* cancelled trailing nodes, upon a request to remove a trailing
212	–	* node, it is placed in field "cleanMe" to be unspliced later.
213	–	*
215		* *** Overview of implementation ***
216		*
217	<	* We use a threshold-based approach to updates, with a target
218	<	* slack of two.  The slack value is hard-wired: a path greater
217	>	* We use a threshold-based approach to updates, with a slack
218	>	* threshold of two -- that is, we update head/tail when the
219	>	* current pointer appears to be two or more steps away from the
220	>	* first/last node. The slack value is hard-wired: a path greater
221		* than one is naturally implemented by checking equality of
222		* traversal pointers except when the list has only one element,
223	<	* in which case we keep max slack at one. Avoiding tracking
224	<	* explicit counts across situations slightly simplifies an
223	>	* in which case we keep slack threshold at one. Avoiding tracking
224	>	* explicit counts across method calls slightly simplifies an
225		* already-messy implementation. Using randomization would
226		* probably work better if there were a low-quality dirt-cheap
227		* per-thread one available, but even ThreadLocalRandom is too
228		* heavy for these purposes.
229		*
230	<	* With such a small slack value, path short-circuiting is rarely
231	<	* worthwhile. However, it is used (in awaitMatch) immediately
232	<	* before a waiting thread starts to block, as a final bit of
233	<	* helping at a point when contention with others is extremely
234	<	* unlikely (since if other threads that could release it are
235	<	* operating, then the current thread wouldn't be blocking).
230	>	* With such a small slack threshold value, it is not worthwhile
231	>	* to augment this with path short-circuiting (i.e., unsplicing
232	>	* interior nodes) except in the case of cancellation/removal (see
233	>	* below).
234	>	*
235	>	* We allow both the head and tail fields to be null before any
236	>	* nodes are enqueued; initializing upon first append.  This
237	>	* simplifies some other logic, as well as providing more
238	>	* efficient explicit control paths instead of letting JVMs insert
239	>	* implicit NullPointerExceptions when they are null.  While not
240	>	* currently fully implemented, we also leave open the possibility
241	>	* of re-nulling these fields when empty (which is complicated to
242	>	* arrange, for little benefit.)
243		*
244		* All enqueue/dequeue operations are handled by the single method
245		* "xfer" with parameters indicating whether to act as some form
246		* of offer, put, poll, take, or transfer (each possibly with
247		* timeout). The relative complexity of using one monolithic
248		* method outweighs the code bulk and maintenance problems of
249	<	* using nine separate methods.
249	>	* using separate methods for each case.
250		*
251		* Operation consists of up to three phases. The first is
252		* implemented within method xfer, the second in tryAppend, and
#	Line 249 | Line 259 | public class LinkedTransferQueue<E> exte
259		*    case matching it and returning, also if necessary updating
260		*    head to one past the matched node (or the node itself if the
261		*    list has no other unmatched nodes). If the CAS misses, then
262	<	*    a retry loops until the slack is at most two. Traversals
263	<	*    also check if the initial head is now off-list, in which
264	<	*    case they start at the new head.
262	>	*    a loop retries advancing head by two steps until either
263	>	*    success or the slack is at most two. By requiring that each
264	>	*    attempt advances head by two (if applicable), we ensure that
265	>	*    the slack does not grow without bound. Traversals also check
266	>	*    if the initial head is now off-list, in which case they
267	>	*    start at the new head.
268		*
269		*    If no candidates are found and the call was untimed
270		*    poll/offer, (argument "how" is NOW) return.
271		*
272		* 2. Try to append a new node (method tryAppend)
273		*
274	<	*    Starting at current tail pointer, try to append a new node
275	<	*    to the list (or if head was null, establish the first
276	<	*    node). Nodes can be appended only if their predecessors are
277	<	*    either already matched or are of the same mode. If we detect
278	<	*    otherwise, then a new node with opposite mode must have been
279	<	*    appended during traversal, so must restart at phase 1. The
280	<	*    traversal and update steps are otherwise similar to phase 1:
281	<	*    Retrying upon CAS misses and checking for staleness.  In
282	<	*    particular, if a self-link is encountered, then we can
283	<	*    safely jump to a node on the list by continuing the
284	<	*    traversal at current head.
274	>	*    Starting at current tail pointer, find the actual last node
275	>	*    and try to append a new node (or if head was null, establish
276	>	*    the first node). Nodes can be appended only if their
277	>	*    predecessors are either already matched or are of the same
278	>	*    mode. If we detect otherwise, then a new node with opposite
279	>	*    mode must have been appended during traversal, so we must
280	>	*    restart at phase 1. The traversal and update steps are
281	>	*    otherwise similar to phase 1: Retrying upon CAS misses and
282	>	*    checking for staleness.  In particular, if a self-link is
283	>	*    encountered, then we can safely jump to a node on the list
284	>	*    by continuing the traversal at current head.
285		*
286		*    On successful append, if the call was ASYNC, return.
287		*
288		* 3. Await match or cancellation (method awaitMatch)
289		*
290		*    Wait for another thread to match node; instead cancelling if
291	<	*    current thread was interrupted or the wait timed out. On
291	>	*    the current thread was interrupted or the wait timed out. On
292		*    multiprocessors, we use front-of-queue spinning: If a node
293		*    appears to be the first unmatched node in the queue, it
294		*    spins a bit before blocking. In either case, before blocking
#	Line 288 | Line 301 | public class LinkedTransferQueue<E> exte
301		*    of less-contended queues.  During spins threads check their
302		*    interrupt status and generate a thread-local random number
303		*    to decide to occasionally perform a Thread.yield. While
304	<	*    yield has underdefined specs, we assume that might it help,
304	>	*    yield has underdefined specs, we assume that it might help,
305		*    and will not hurt in limiting impact of spinning on busy
306	<	*    systems.  We also use much smaller (1/4) spins for nodes
307	<	*    that are not known to be front but whose predecessors have
308	<	*    not blocked -- these "chained" spins avoid artifacts of
306	>	*    systems.  We also use smaller (1/2) spins for nodes that are
307	>	*    not known to be front but whose predecessors have not
308	>	*    blocked -- these "chained" spins avoid artifacts of
309		*    front-of-queue rules which otherwise lead to alternating
310		*    nodes spinning vs blocking. Further, front threads that
311		*    represent phase changes (from data to request node or vice
312		*    versa) compared to their predecessors receive additional
313	<	*    spins, reflecting the longer code path lengths necessary to
314	<	*    release them under contention.
313	>	*    chained spins, reflecting longer paths typically required to
314	>	*    unblock threads during phase changes.
315	>	*
316	>	*
317	>	* ** Unlinking removed interior nodes **
318	>	*
319	>	* In addition to minimizing garbage retention via self-linking
320	>	* described above, we also unlink removed interior nodes. These
321	>	* may arise due to timed out or interrupted waits, or calls to
322	>	* remove(x) or Iterator.remove.  Normally, given a node that was
323	>	* at one time known to be the predecessor of some node s that is
324	>	* to be removed, we can unsplice s by CASing the next field of
325	>	* its predecessor if it still points to s (otherwise s must
326	>	* already have been removed or is now offlist). But there are two
327	>	* situations in which we cannot guarantee to make node s
328	>	* unreachable in this way: (1) If s is the trailing node of list
329	>	* (i.e., with null next), then it is pinned as the target node
330	>	* for appends, so can only be removed later after other nodes are
331	>	* appended. (2) We cannot necessarily unlink s given a
332	>	* predecessor node that is matched (including the case of being
333	>	* cancelled): the predecessor may already be unspliced, in which
334	>	* case some previous reachable node may still point to s.
335	>	* (For further explanation see Herlihy & Shavit "The Art of
336	>	* Multiprocessor Programming" chapter 9).  Although, in both
337	>	* cases, we can rule out the need for further action if either s
338	>	* or its predecessor are (or can be made to be) at, or fall off
339	>	* from, the head of list.
340	>	*
341	>	* Without taking these into account, it would be possible for an
342	>	* unbounded number of supposedly removed nodes to remain
343	>	* reachable.  Situations leading to such buildup are uncommon but
344	>	* can occur in practice; for example when a series of short timed
345	>	* calls to poll repeatedly time out but never otherwise fall off
346	>	* the list because of an untimed call to take at the front of the
347	>	* queue.
348	>	*
349	>	* When these cases arise, rather than always retraversing the
350	>	* entire list to find an actual predecessor to unlink (which
351	>	* won't help for case (1) anyway), we record a conservative
352	>	* estimate of possible unsplice failures (in "sweepVotes").
353	>	* We trigger a full sweep when the estimate exceeds a threshold
354	>	* ("SWEEP_THRESHOLD") indicating the maximum number of estimated
355	>	* removal failures to tolerate before sweeping through, unlinking
356	>	* cancelled nodes that were not unlinked upon initial removal.
357	>	* We perform sweeps by the thread hitting threshold (rather than
358	>	* background threads or by spreading work to other threads)
359	>	* because in the main contexts in which removal occurs, the
360	>	* caller is already timed-out, cancelled, or performing a
361	>	* potentially O(n) operation (e.g. remove(x)), none of which are
362	>	* time-critical enough to warrant the overhead that alternatives
363	>	* would impose on other threads.
364	>	*
365	>	* Because the sweepVotes estimate is conservative, and because
366	>	* nodes become unlinked "naturally" as they fall off the head of
367	>	* the queue, and because we allow votes to accumulate even while
368	>	* sweeps are in progress, there are typically significantly fewer
369	>	* such nodes than estimated.  Choice of a threshold value
370	>	* balances the likelihood of wasted effort and contention, versus
371	>	* providing a worst-case bound on retention of interior nodes in
372	>	* quiescent queues. The value defined below was chosen
373	>	* empirically to balance these under various timeout scenarios.
374	>	*
375	>	* Note that we cannot self-link unlinked interior nodes during
376	>	* sweeps. However, the associated garbage chains terminate when
377	>	* some successor ultimately falls off the head of the list and is
378	>	* self-linked.
379		*/
380
381		/** True if on multiprocessor */
#	Line 306 | Line 383 | public class LinkedTransferQueue<E> exte
383		Runtime.getRuntime().availableProcessors() > 1;
384
385		/**
386	<	* The number of times to spin (with on average one randomly
387	<	* interspersed call to Thread.yield) on multiprocessor before
388	<	* blocking when a node is apparently the first waiter in the
389	<	* queue.  See above for explanation. Must be a power of two. The
390	<	* value is empirically derived -- it works pretty well across a
391	<	* variety of processors, numbers of CPUs, and OSes.
386	>	* The number of times to spin (with randomly interspersed calls
387	>	* to Thread.yield) on multiprocessor before blocking when a node
388	>	* is apparently the first waiter in the queue.  See above for
389	>	* explanation. Must be a power of two. The value is empirically
390	>	* derived -- it works pretty well across a variety of processors,
391	>	* numbers of CPUs, and OSes.
392		*/
393		private static final int FRONT_SPINS   = 1 << 7;
394
395		/**
396		* The number of times to spin before blocking when a node is
397	<	* preceded by another node that is apparently spinning.
397	>	* preceded by another node that is apparently spinning.  Also
398	>	* serves as an increment to FRONT_SPINS on phase changes, and as
399	>	* base average frequency for yielding during spins. Must be a
400	>	* power of two.
401		*/
402	<	private static final int CHAINED_SPINS = FRONT_SPINS >>> 2;
402	>	private static final int CHAINED_SPINS = FRONT_SPINS >>> 1;
403	>
404	>	/**
405	>	* The maximum number of estimated removal failures (sweepVotes)
406	>	* to tolerate before sweeping through the queue unlinking
407	>	* cancelled nodes that were not unlinked upon initial
408	>	* removal. See above for explanation. The value must be at least
409	>	* two to avoid useless sweeps when removing trailing nodes.
410	>	*/
411	>	static final int SWEEP_THRESHOLD = 32;
412
413		/**
414		* Queue nodes. Uses Object, not E, for items to allow forgetting
415		* them after use.  Relies heavily on Unsafe mechanics to minimize
416	<	* unnecessary ordering constraints: Writes that intrinsically
417	<	* precede or follow CASes use simple relaxed forms.  Other
329	<	* cleanups use releasing/lazy writes.
416	>	* unnecessary ordering constraints: Writes that are intrinsically
417	>	* ordered wrt other accesses or CASes use simple relaxed forms.
418		*/
419		static final class Node {
420		final boolean isData;   // false if this is a request node
#	Line 340 | Line 428 | public class LinkedTransferQueue<E> exte
428		}
429
430		final boolean casItem(Object cmp, Object val) {
431	+	// assert cmp == null || cmp.getClass() != Node.class;
432		return UNSAFE.compareAndSwapObject(this, itemOffset, cmp, val);
433		}
434
435		/**
436	<	* Creates a new node. Uses relaxed write because item can only
437	<	* be seen if followed by CAS.
436	>	* Constructs a new node.  Uses relaxed write because item can
437	>	* only be seen after publication via casNext.
438		*/
439		Node(Object item, boolean isData) {
440		UNSAFE.putObject(this, itemOffset, item); // relaxed write
#	Line 361 | Line 450 | public class LinkedTransferQueue<E> exte
450		}
451
452		/**
453	<	* Sets item to self (using a releasing/lazy write) and waiter
454	<	* to null, to avoid garbage retention after extracting or
455	<	* cancelling.
453	>	* Sets item to self and waiter to null, to avoid garbage
454	>	* retention after matching or cancelling. Uses relaxed writes
455	>	* because order is already constrained in the only calling
456	>	* contexts: item is forgotten only after volatile/atomic
457	>	* mechanics that extract items.  Similarly, clearing waiter
458	>	* follows either CAS or return from park (if ever parked;
459	>	* else we don't care).
460		*/
461		final void forgetContents() {
462	<	UNSAFE.putOrderedObject(this, itemOffset, this);
463	<	UNSAFE.putOrderedObject(this, waiterOffset, null);
462	>	UNSAFE.putObject(this, itemOffset, this);
463	>	UNSAFE.putObject(this, waiterOffset, null);
464		}
465
466		/**
#	Line 376 | Line 469 | public class LinkedTransferQueue<E> exte
469		*/
470		final boolean isMatched() {
471		Object x = item;
472	<	return x == this || (x != null) != isData;
472	>	return (x == this) || ((x == null) == isData);
473	>	}
474	>
475	>	/**
476	>	* Returns true if this is an unmatched request node.
477	>	*/
478	>	final boolean isUnmatchedRequest() {
479	>	return !isData && item == null;
480		}
481
482		/**
#	Line 394 | Line 494 | public class LinkedTransferQueue<E> exte
494		* Tries to artificially match a data node -- used by remove.
495		*/
496		final boolean tryMatchData() {
497	+	// assert isData;
498		Object x = item;
499		if (x != null && x != this && casItem(x, null)) {
500		LockSupport.unpark(waiter);
#	Line 402 | Line 503 | public class LinkedTransferQueue<E> exte
503		return false;
504		}
505
405	–	// Unsafe mechanics
406	–	private static final sun.misc.Unsafe UNSAFE = getUnsafe();
407	–	private static final long nextOffset =
408	–	objectFieldOffset(UNSAFE, "next", Node.class);
409	–	private static final long itemOffset =
410	–	objectFieldOffset(UNSAFE, "item", Node.class);
411	–	private static final long waiterOffset =
412	–	objectFieldOffset(UNSAFE, "waiter", Node.class);
413	–
506		private static final long serialVersionUID = -3375979862319811754L;
507	+
508	+	// Unsafe mechanics
509	+	private static final sun.misc.Unsafe UNSAFE;
510	+	private static final long itemOffset;
511	+	private static final long nextOffset;
512	+	private static final long waiterOffset;
513	+	static {
514	+	try {
515	+	UNSAFE = getUnsafe();
516	+	Class<?> k = Node.class;
517	+	itemOffset = UNSAFE.objectFieldOffset
518	+	(k.getDeclaredField("item"));
519	+	nextOffset = UNSAFE.objectFieldOffset
520	+	(k.getDeclaredField("next"));
521	+	waiterOffset = UNSAFE.objectFieldOffset
522	+	(k.getDeclaredField("waiter"));
523	+	} catch (Exception e) {
524	+	throw new Error(e);
525	+	}
526	+	}
527		}
528
529		/** head of the queue; null until first enqueue */
530	<	private transient volatile Node head;
419	<
420	<	/** predecessor of dangling unspliceable node */
421	<	private transient volatile Node cleanMe; // decl here to reduce contention
530	>	transient volatile Node head;
531
532		/** tail of the queue; null until first append */
533		private transient volatile Node tail;
534
535	+	/** The number of apparent failures to unsplice removed nodes */
536	+	private transient volatile int sweepVotes;
537	+
538		// CAS methods for fields
539		private boolean casTail(Node cmp, Node val) {
540		return UNSAFE.compareAndSwapObject(this, tailOffset, cmp, val);
#	Line 432 | Line 544 | public class LinkedTransferQueue<E> exte
544		return UNSAFE.compareAndSwapObject(this, headOffset, cmp, val);
545		}
546
547	<	private boolean casCleanMe(Node cmp, Node val) {
548	<	return UNSAFE.compareAndSwapObject(this, cleanMeOffset, cmp, val);
547	>	private boolean casSweepVotes(int cmp, int val) {
548	>	return UNSAFE.compareAndSwapInt(this, sweepVotesOffset, cmp, val);
549		}
550
551		/*
552	<	* Possible values for "how" argument in xfer method. Beware that
441	<	* the order of assigned numerical values matters.
552	>	* Possible values for "how" argument in xfer method.
553		*/
554	<	private static final int NOW     = 0; // for untimed poll, tryTransfer
555	<	private static final int ASYNC   = 1; // for offer, put, add
556	<	private static final int SYNC    = 2; // for transfer, take
557	<	private static final int TIMEOUT = 3; // for timed poll, tryTransfer
554	>	private static final int NOW   = 0; // for untimed poll, tryTransfer
555	>	private static final int ASYNC = 1; // for offer, put, add
556	>	private static final int SYNC  = 2; // for transfer, take
557	>	private static final int TIMED = 3; // for timed poll, tryTransfer
558	>
559	>	@SuppressWarnings("unchecked")
560	>	static <E> E cast(Object item) {
561	>	// assert item == null || item.getClass() != Node.class;
562	>	return (E) item;
563	>	}
564
565		/**
566		* Implements all queuing methods. See above for explanation.
567		*
568		* @param e the item or null for take
569		* @param haveData true if this is a put, else a take
570	<	* @param how NOW, ASYNC, SYNC, or TIMEOUT
571	<	* @param nanos timeout in nanosecs, used only if mode is TIMEOUT
570	>	* @param how NOW, ASYNC, SYNC, or TIMED
571	>	* @param nanos timeout in nanosecs, used only if mode is TIMED
572		* @return an item if matched, else e
573		* @throws NullPointerException if haveData mode but e is null
574		*/
575	<	private Object xfer(Object e, boolean haveData, int how, long nanos) {
575	>	private E xfer(E e, boolean haveData, int how, long nanos) {
576		if (haveData && (e == null))
577		throw new NullPointerException();
578		Node s = null;                        // the node to append, if needed
579
580	<	retry: for (;;) {                     // restart on append race
580	>	retry:
581	>	for (;;) {                            // restart on append race
582
583		for (Node h = head, p = h; p != null;) { // find & match first node
584		boolean isData = p.isData;
#	Line 469 | Line 587 | public class LinkedTransferQueue<E> exte
587		if (isData == haveData)   // can't match
588		break;
589		if (p.casItem(item, e)) { // match
590	<	Thread w = p.waiter;
591	<	while (p != h) {      // update head
592	<	Node n = p.next;  // by 2 unless singleton
475	<	if (n != null)
476	<	p = n;
477	<	if (head == h && casHead(h, p)) {
590	>	for (Node q = p; q != h;) {
591	>	Node n = q.next;  // update by 2 unless singleton
592	>	if (head == h && casHead(h, n == null ? q : n)) {
593		h.forgetNext();
594		break;
595		}                 // advance and retry
596		if ((h = head)   == null ||
597	<	(p = h.next) == null || !p.isMatched())
597	>	(q = h.next) == null || !q.isMatched())
598		break;        // unless slack < 2
599		}
600	<	LockSupport.unpark(w);
601	<	return item;
600	>	LockSupport.unpark(p.waiter);
601	>	return this.<E>cast(item);
602		}
603		}
604		Node n = p.next;
605	<	p = p != n ? n : (h = head);  // Use head if p offlist
605	>	p = (p != n) ? n : (h = head); // Use head if p offlist
606		}
607
608	<	if (how >= ASYNC) {               // No matches available
608	>	if (how != NOW) {                 // No matches available
609		if (s == null)
610		s = new Node(e, haveData);
611		Node pred = tryAppend(s, haveData);
612		if (pred == null)
613		continue retry;           // lost race vs opposite mode
614	<	if (how >= SYNC)
615	<	return awaitMatch(pred, s, e, how, nanos);
614	>	if (how != ASYNC)
615	>	return awaitMatch(s, pred, e, (how == TIMED), nanos);
616		}
617		return e; // not waiting
618		}
#	Line 506 | Line 621 | public class LinkedTransferQueue<E> exte
621		/**
622		* Tries to append node s as tail.
623		*
509	–	* @param haveData true if appending in data mode
624		* @param s the node to append
625	+	* @param haveData true if appending in data mode
626		* @return null on failure due to losing race with append in
627		* different mode, else s's predecessor, or s itself if no
628		* predecessor
629		*/
630		private Node tryAppend(Node s, boolean haveData) {
631	<	for (Node t = tail, p = t;;) { // move p to actual tail and append
631	>	for (Node t = tail, p = t;;) {        // move p to last node and append
632		Node n, u;                        // temps for reads of next & tail
633		if (p == null && (p = head) == null) {
634		if (casHead(null, s))
#	Line 521 | Line 636 | public class LinkedTransferQueue<E> exte
636		}
637		else if (p.cannotPrecede(haveData))
638		return null;                  // lost race vs opposite mode
639	<	else if ((n = p.next) != null)    // Not tail; keep traversing
639	>	else if ((n = p.next) != null)    // not last; keep traversing
640		p = p != t && t != (u = tail) ? (t = u) : // stale tail
641	<	p != n ? n : null;        // restart if off list
641	>	(p != n) ? n : null;      // restart if off list
642		else if (!p.casNext(null, s))
643		p = p.next;                   // re-read on CAS failure
644		else {
645	<	if (p != t) {                 // Update if slack now >= 2
645	>	if (p != t) {                 // update if slack now >= 2
646		while ((tail != t || !casTail(t, s)) &&
647		(t = tail)   != null &&
648		(s = t.next) != null && // advance and retry
#	Line 541 | Line 656 | public class LinkedTransferQueue<E> exte
656		/**
657		* Spins/yields/blocks until node s is matched or caller gives up.
658		*
544	–	* @param pred the predecessor of s or s or null if none
659		* @param s the waiting node
660	+	* @param pred the predecessor of s, or s itself if it has no
661	+	* predecessor, or null if unknown (the null case does not occur
662	+	* in any current calls but may in possible future extensions)
663		* @param e the comparison value for checking match
664	<	* @param how either SYNC or TIMEOUT
665	<	* @param nanos timeout value
664	>	* @param timed if true, wait only until timeout elapses
665	>	* @param nanos timeout in nanosecs, used only if timed is true
666		* @return matched item, or e if unmatched on interrupt or timeout
667		*/
668	<	private Object awaitMatch(Node pred, Node s, Object e,
669	<	int how, long nanos) {
553	<	long lastTime = (how == TIMEOUT) ? System.nanoTime() : 0L;
668	>	private E awaitMatch(Node s, Node pred, E e, boolean timed, long nanos) {
669	>	long lastTime = timed ? System.nanoTime() : 0L;
670		Thread w = Thread.currentThread();
671		int spins = -1; // initialized after first item and cancel checks
672		ThreadLocalRandom randomYields = null; // bound if needed
#	Line 558 | Line 674 | public class LinkedTransferQueue<E> exte
674		for (;;) {
675		Object item = s.item;
676		if (item != e) {                  // matched
677	+	// assert item != s;
678		s.forgetContents();           // avoid garbage
679	<	return item;
679	>	return this.<E>cast(item);
680		}
681	<	if ((w.isInterrupted() || (how == TIMEOUT && nanos <= 0)) &&
682	<	s.casItem(e, s)) {       // cancel
681	>	if ((w.isInterrupted() || (timed && nanos <= 0)) &&
682	>	s.casItem(e, s)) {        // cancel
683		unsplice(pred, s);
684		return e;
685		}
#	Line 571 | Line 688 | public class LinkedTransferQueue<E> exte
688		if ((spins = spinsFor(pred, s.isData)) > 0)
689		randomYields = ThreadLocalRandom.current();
690		}
691	<	else if (spins > 0) {             // spin, occasionally yield
575	<	if (randomYields.nextInt(FRONT_SPINS) == 0)
576	<	Thread.yield();
691	>	else if (spins > 0) {             // spin
692		--spins;
693	+	if (randomYields.nextInt(CHAINED_SPINS) == 0)
694	+	Thread.yield();           // occasionally yield
695		}
696		else if (s.waiter == null) {
697	<	shortenHeadPath();            // reduce slack before blocking
581	<	s.waiter = w;                 // request unpark
697	>	s.waiter = w;                 // request unpark then recheck
698		}
699	<	else if (how == TIMEOUT) {
699	>	else if (timed) {
700		long now = System.nanoTime();
701		if ((nanos -= now - lastTime) > 0)
702		LockSupport.parkNanos(this, nanos);
#	Line 588 | Line 704 | public class LinkedTransferQueue<E> exte
704		}
705		else {
706		LockSupport.park(this);
591	–	spins = -1;                   // spin if front upon wakeup
707		}
708		}
709		}
#	Line 599 | Line 714 | public class LinkedTransferQueue<E> exte
714		*/
715		private static int spinsFor(Node pred, boolean haveData) {
716		if (MP && pred != null) {
717	<	boolean predData = pred.isData;
718	<	if (predData != haveData)         // front and phase change
719	<	return FRONT_SPINS + (FRONT_SPINS >>> 1);
605	<	if (predData != (pred.item != null)) // probably at front
717	>	if (pred.isData != haveData)      // phase change
718	>	return FRONT_SPINS + CHAINED_SPINS;
719	>	if (pred.isMatched())             // probably at front
720		return FRONT_SPINS;
721		if (pred.waiter == null)          // pred apparently spinning
722		return CHAINED_SPINS;
#	Line 610 | Line 724 | public class LinkedTransferQueue<E> exte
724		return 0;
725		}
726
727	+	/* -------------- Traversal methods -------------- */
728	+
729		/**
730	<	* Tries (once) to unsplice nodes between head and first unmatched
731	<	* or trailing node; failing on contention.
732	<	*/
733	<	private void shortenHeadPath() {
734	<	Node h, hn, p, q;
735	<	if ((p = h = head) != null && h.isMatched() &&
736	<	(q = hn = h.next) != null) {
621	<	Node n;
622	<	while ((n = q.next) != q) {
623	<	if (n == null || !q.isMatched()) {
624	<	if (hn != q && h.next == hn)
625	<	h.casNext(hn, q);
626	<	break;
627	<	}
628	<	p = q;
629	<	q = n;
630	<	}
631	<	}
730	>	* Returns the successor of p, or the head node if p.next has been
731	>	* linked to self, which will only be true if traversing with a
732	>	* stale pointer that is now off the list.
733	>	*/
734	>	final Node succ(Node p) {
735	>	Node next = p.next;
736	>	return (p == next) ? head : next;
737		}
738
634	–	/* -------------- Traversal methods -------------- */
635	–
739		/**
740		* Returns the first unmatched node of the given mode, or null if
741		* none.  Used by methods isEmpty, hasWaitingConsumer.
742		*/
743	<	private Node firstOfMode(boolean data) {
744	<	for (Node p = head; p != null; ) {
743	>	private Node firstOfMode(boolean isData) {
744	>	for (Node p = head; p != null; p = succ(p)) {
745		if (!p.isMatched())
746	<	return p.isData == data? p : null;
644	<	Node n = p.next;
645	<	p = n != p ? n : head;
746	>	return (p.isData == isData) ? p : null;
747		}
748		return null;
749		}
750
751		/**
752		* Returns the item in the first unmatched node with isData; or
753	<	* null if none. Used by peek.
753	>	* null if none.  Used by peek.
754		*/
755	<	private Object firstDataItem() {
756	<	for (Node p = head; p != null; ) {
656	<	boolean isData = p.isData;
755	>	private E firstDataItem() {
756	>	for (Node p = head; p != null; p = succ(p)) {
757		Object item = p.item;
758	<	if (item != p && (item != null) == isData)
759	<	return isData ? item : null;
760	<	Node n = p.next;
761	<	p = n != p ? n : head;
758	>	if (p.isData) {
759	>	if (item != null && item != p)
760	>	return this.<E>cast(item);
761	>	}
762	>	else if (item == null)
763	>	return null;
764		}
765		return null;
766		}
#	Line 689 | Line 791 | public class LinkedTransferQueue<E> exte
791
792		final class Itr implements Iterator<E> {
793		private Node nextNode;   // next node to return item for
794	<	private Object nextItem; // the corresponding item
794	>	private E nextItem;      // the corresponding item
795		private Node lastRet;    // last returned node, to support remove
796	+	private Node lastPred;   // predecessor to unlink lastRet
797
798		/**
799		* Moves to next node after prev, or first node if prev null.
800		*/
801		private void advance(Node prev) {
802	<	lastRet = prev;
803	<	Node p;
804	<	if (prev == null || (p = prev.next) == prev)
805	<	p = head;
806	<	while (p != null) {
807	<	Object item = p.item;
808	<	if (p.isData) {
809	<	if (item != null && item != p) {
810	<	nextItem = item;
811	<	nextNode = p;
802	>	/*
803	>	* To track and avoid buildup of deleted nodes in the face
804	>	* of calls to both Queue.remove and Itr.remove, we must
805	>	* include variants of unsplice and sweep upon each
806	>	* advance: Upon Itr.remove, we may need to catch up links
807	>	* from lastPred, and upon other removes, we might need to
808	>	* skip ahead from stale nodes and unsplice deleted ones
809	>	* found while advancing.
810	>	*/
811	>
812	>	Node r, b; // reset lastPred upon possible deletion of lastRet
813	>	if ((r = lastRet) != null && !r.isMatched())
814	>	lastPred = r;    // next lastPred is old lastRet
815	>	else if ((b = lastPred) == null || b.isMatched())
816	>	lastPred = null; // at start of list
817	>	else {
818	>	Node s, n;       // help with removal of lastPred.next
819	>	while ((s = b.next) != null &&
820	>	s != b && s.isMatched() &&
821	>	(n = s.next) != null && n != s)
822	>	b.casNext(s, n);
823	>	}
824	>
825	>	this.lastRet = prev;
826	>
827	>	for (Node p = prev, s, n;;) {
828	>	s = (p == null) ? head : p.next;
829	>	if (s == null)
830	>	break;
831	>	else if (s == p) {
832	>	p = null;
833	>	continue;
834	>	}
835	>	Object item = s.item;
836	>	if (s.isData) {
837	>	if (item != null && item != s) {
838	>	nextItem = LinkedTransferQueue.<E>cast(item);
839	>	nextNode = s;
840		return;
841		}
842		}
843		else if (item == null)
844		break;
845	<	Node n = p.next;
846	<	p = n != p ? n : head;
845	>	// assert s.isMatched();
846	>	if (p == null)
847	>	p = s;
848	>	else if ((n = s.next) == null)
849	>	break;
850	>	else if (s == n)
851	>	p = null;
852	>	else
853	>	p.casNext(s, n);
854		}
855		nextNode = null;
856	+	nextItem = null;
857		}
858
859		Itr() {
#	Line 728 | Line 867 | public class LinkedTransferQueue<E> exte
867		public final E next() {
868		Node p = nextNode;
869		if (p == null) throw new NoSuchElementException();
870	<	Object e = nextItem;
870	>	E e = nextItem;
871		advance(p);
872	<	return (E) e;
872	>	return e;
873		}
874
875		public final void remove() {
876	<	Node p = lastRet;
877	<	if (p == null) throw new IllegalStateException();
878	<	lastRet = null;
879	<	findAndRemoveNode(p);
876	>	final Node lastRet = this.lastRet;
877	>	if (lastRet == null)
878	>	throw new IllegalStateException();
879	>	this.lastRet = null;
880	>	if (lastRet.tryMatchData())
881	>	unsplice(lastPred, lastRet);
882		}
883		}
884
#	Line 747 | Line 888 | public class LinkedTransferQueue<E> exte
888		* Unsplices (now or later) the given deleted/cancelled node with
889		* the given predecessor.
890		*
891	<	* @param pred predecessor of node to be unspliced
891	>	* @param pred a node that was at one time known to be the
892	>	* predecessor of s, or null or s itself if s is/was at head
893		* @param s the node to be unspliced
894		*/
895	<	private void unsplice(Node pred, Node s) {
896	<	s.forgetContents(); // clear unneeded fields
895	>	final void unsplice(Node pred, Node s) {
896	>	s.forgetContents(); // forget unneeded fields
897		/*
898	<	* At any given time, exactly one node on list cannot be
899	<	* deleted -- the last inserted node. To accommodate this, if
900	<	* we cannot delete s, we save its predecessor as "cleanMe",
901	<	* processing the previously saved version first. Because only
902	<	* one node in the list can have a null next, at least one of
761	<	* node s or the node previously saved can always be
762	<	* processed, so this always terminates.
898	>	* See above for rationale. Briefly: if pred still points to
899	>	* s, try to unlink s.  If s cannot be unlinked, because it is
900	>	* trailing node or pred might be unlinked, and neither pred
901	>	* nor s are head or offlist, add to sweepVotes, and if enough
902	>	* votes have accumulated, sweep.
903		*/
904	<	if (pred != null && pred != s) {
905	<	while (pred.next == s) {
906	<	Node oldpred = cleanMe == null? null : reclean();
907	<	Node n = s.next;
908	<	if (n != null) {
909	<	if (n != s)
910	<	pred.casNext(s, n);
911	<	break;
904	>	if (pred != null && pred != s && pred.next == s) {
905	>	Node n = s.next;
906	>	if (n == null ||
907	>	(n != s && pred.casNext(s, n) && pred.isMatched())) {
908	>	for (;;) {               // check if at, or could be, head
909	>	Node h = head;
910	>	if (h == pred || h == s || h == null)
911	>	return;          // at head or list empty
912	>	if (!h.isMatched())
913	>	break;
914	>	Node hn = h.next;
915	>	if (hn == null)
916	>	return;          // now empty
917	>	if (hn != h && casHead(h, hn))
918	>	h.forgetNext();  // advance head
919	>	}
920	>	if (pred.next != pred && s.next != s) { // recheck if offlist
921	>	for (;;) {           // sweep now if enough votes
922	>	int v = sweepVotes;
923	>	if (v < SWEEP_THRESHOLD) {
924	>	if (casSweepVotes(v, v + 1))
925	>	break;
926	>	}
927	>	else if (casSweepVotes(v, 0)) {
928	>	sweep();
929	>	break;
930	>	}
931	>	}
932		}
773	–	if (oldpred == pred ||      // Already saved
774	–	(oldpred == null && casCleanMe(null, pred)))
775	–	break;                  // Postpone cleaning
933		}
934		}
935		}
936
937		/**
938	<	* Tries to unsplice the deleted/cancelled node held in cleanMe
939	<	* that was previously uncleanable because it was at tail.
783	<	*
784	<	* @return current cleanMe node (or null)
938	>	* Unlinks matched (typically cancelled) nodes encountered in a
939	>	* traversal from head.
940		*/
941	<	private Node reclean() {
942	<	/*
943	<	* cleanMe is, or at one time was, predecessor of a cancelled
944	<	* node s that was the tail so could not be unspliced.  If it
945	<	* is no longer the tail, try to unsplice if necessary and
946	<	* make cleanMe slot available.  This differs from similar
792	<	* code in unsplice() because we must check that pred still
793	<	* points to a matched node that can be unspliced -- if not,
794	<	* we can (must) clear cleanMe without unsplicing.  This can
795	<	* loop only due to contention.
796	<	*/
797	<	Node pred;
798	<	while ((pred = cleanMe) != null) {
799	<	Node s = pred.next;
800	<	Node n;
801	<	if (s == null || s == pred || !s.isMatched())
802	<	casCleanMe(pred, null); // already gone
803	<	else if ((n = s.next) != null) {
804	<	if (n != s)
805	<	pred.casNext(s, n);
806	<	casCleanMe(pred, null);
807	<	}
808	<	else
941	>	private void sweep() {
942	>	for (Node p = head, s, n; p != null && (s = p.next) != null; ) {
943	>	if (!s.isMatched())
944	>	// Unmatched nodes are never self-linked
945	>	p = s;
946	>	else if ((n = s.next) == null) // trailing node is pinned
947		break;
948	<	}
949	<	return pred;
950	<	}
951	<
952	<	/**
815	<	* Main implementation of Iterator.remove(). Find
816	<	* and unsplice the given node.
817	<	*/
818	<	final void findAndRemoveNode(Node s) {
819	<	if (s.tryMatchData()) {
820	<	Node pred = null;
821	<	Node p = head;
822	<	while (p != null) {
823	<	if (p == s) {
824	<	unsplice(pred, p);
825	<	break;
826	<	}
827	<	if (!p.isData && !p.isMatched())
828	<	break;
829	<	pred = p;
830	<	if ((p = p.next) == pred) { // stale
831	<	pred = null;
832	<	p = head;
833	<	}
834	<	}
948	>	else if (s == n)    // stale
949	>	// No need to also check for p == s, since that implies s == n
950	>	p = head;
951	>	else
952	>	p.casNext(s, n);
953		}
954		}
955
#	Line 840 | Line 958 | public class LinkedTransferQueue<E> exte
958		*/
959		private boolean findAndRemove(Object e) {
960		if (e != null) {
961	<	Node pred = null;
844	<	Node p = head;
845	<	while (p != null) {
961	>	for (Node pred = null, p = head; p != null; ) {
962		Object item = p.item;
963		if (p.isData) {
964		if (item != null && item != p && e.equals(item) &&
#	Line 854 | Line 970 | public class LinkedTransferQueue<E> exte
970		else if (item == null)
971		break;
972		pred = p;
973	<	if ((p = p.next) == pred) {
973	>	if ((p = p.next) == pred) { // stale
974		pred = null;
975		p = head;
976		}
#	Line 900 | Line 1016 | public class LinkedTransferQueue<E> exte
1016		* return {@code false}.
1017		*
1018		* @return {@code true} (as specified by
1019	<	*  {@link BlockingQueue#offer(Object,long,TimeUnit) BlockingQueue.offer})
1019	>	*  {@link java.util.concurrent.BlockingQueue#offer(Object,long,TimeUnit)
1020	>	*  BlockingQueue.offer})
1021		* @throws NullPointerException if the specified element is null
1022		*/
1023		public boolean offer(E e, long timeout, TimeUnit unit) {
#	Line 912 | Line 1029 | public class LinkedTransferQueue<E> exte
1029		* Inserts the specified element at the tail of this queue.
1030		* As the queue is unbounded, this method will never return {@code false}.
1031		*
1032	<	* @return {@code true} (as specified by
916	<	*         {@link BlockingQueue#offer(Object) BlockingQueue.offer})
1032	>	* @return {@code true} (as specified by {@link Queue#offer})
1033		* @throws NullPointerException if the specified element is null
1034		*/
1035		public boolean offer(E e) {
#	Line 982 | Line 1098 | public class LinkedTransferQueue<E> exte
1098		*/
1099		public boolean tryTransfer(E e, long timeout, TimeUnit unit)
1100		throws InterruptedException {
1101	<	if (xfer(e, true, TIMEOUT, unit.toNanos(timeout)) == null)
1101	>	if (xfer(e, true, TIMED, unit.toNanos(timeout)) == null)
1102		return true;
1103		if (!Thread.interrupted())
1104		return false;
#	Line 990 | Line 1106 | public class LinkedTransferQueue<E> exte
1106		}
1107
1108		public E take() throws InterruptedException {
1109	<	Object e = xfer(null, false, SYNC, 0);
1109	>	E e = xfer(null, false, SYNC, 0);
1110		if (e != null)
1111	<	return (E)e;
1111	>	return e;
1112		Thread.interrupted();
1113		throw new InterruptedException();
1114		}
1115
1116		public E poll(long timeout, TimeUnit unit) throws InterruptedException {
1117	<	Object e = xfer(null, false, TIMEOUT, unit.toNanos(timeout));
1117	>	E e = xfer(null, false, TIMED, unit.toNanos(timeout));
1118		if (e != null || !Thread.interrupted())
1119	<	return (E)e;
1119	>	return e;
1120		throw new InterruptedException();
1121		}
1122
1123		public E poll() {
1124	<	return (E)xfer(null, false, NOW, 0);
1124	>	return xfer(null, false, NOW, 0);
1125		}
1126
1127		/**
#	Line 1045 | Line 1161 | public class LinkedTransferQueue<E> exte
1161		}
1162
1163		/**
1164	<	* Returns an iterator over the elements in this queue in proper
1165	<	* sequence, from head to tail.
1164	>	* Returns an iterator over the elements in this queue in proper sequence.
1165	>	* The elements will be returned in order from first (head) to last (tail).
1166		*
1167		* <p>The returned iterator is a "weakly consistent" iterator that
1168	<	* will never throw
1169	<	* {@link ConcurrentModificationException ConcurrentModificationException},
1170	<	* and guarantees to traverse elements as they existed upon
1171	<	* construction of the iterator, and may (but is not guaranteed
1172	<	* to) reflect any modifications subsequent to construction.
1168	>	* will never throw {@link java.util.ConcurrentModificationException
1169	>	* ConcurrentModificationException}, and guarantees to traverse
1170	>	* elements as they existed upon construction of the iterator, and
1171	>	* may (but is not guaranteed to) reflect any modifications
1172	>	* subsequent to construction.
1173		*
1174		* @return an iterator over the elements in this queue in proper sequence
1175		*/
#	Line 1062 | Line 1178 | public class LinkedTransferQueue<E> exte
1178		}
1179
1180		public E peek() {
1181	<	return (E) firstDataItem();
1181	>	return firstDataItem();
1182		}
1183
1184		/**
#	Line 1071 | Line 1187 | public class LinkedTransferQueue<E> exte
1187		* @return {@code true} if this queue contains no elements
1188		*/
1189		public boolean isEmpty() {
1190	<	return firstOfMode(true) == null;
1190	>	for (Node p = head; p != null; p = succ(p)) {
1191	>	if (!p.isMatched())
1192	>	return !p.isData;
1193	>	}
1194	>	return true;
1195		}
1196
1197		public boolean hasWaitingConsumer() {
#	Line 1114 | Line 1234 | public class LinkedTransferQueue<E> exte
1234		}
1235
1236		/**
1237	+	* Returns {@code true} if this queue contains the specified element.
1238	+	* More formally, returns {@code true} if and only if this queue contains
1239	+	* at least one element {@code e} such that {@code o.equals(e)}.
1240	+	*
1241	+	* @param o object to be checked for containment in this queue
1242	+	* @return {@code true} if this queue contains the specified element
1243	+	*/
1244	+	public boolean contains(Object o) {
1245	+	if (o == null) return false;
1246	+	for (Node p = head; p != null; p = succ(p)) {
1247	+	Object item = p.item;
1248	+	if (p.isData) {
1249	+	if (item != null && item != p && o.equals(item))
1250	+	return true;
1251	+	}
1252	+	else if (item == null)
1253	+	break;
1254	+	}
1255	+	return false;
1256	+	}
1257	+
1258	+	/**
1259		* Always returns {@code Integer.MAX_VALUE} because a
1260		* {@code LinkedTransferQueue} is not capacity constrained.
1261		*
1262		* @return {@code Integer.MAX_VALUE} (as specified by
1263	<	*         {@link BlockingQueue#remainingCapacity()})
1263	>	*         {@link java.util.concurrent.BlockingQueue#remainingCapacity()
1264	>	*         BlockingQueue.remainingCapacity})
1265		*/
1266		public int remainingCapacity() {
1267		return Integer.MAX_VALUE;
#	Line 1158 | Line 1301 | public class LinkedTransferQueue<E> exte
1301		}
1302		}
1303
1161	–
1304		// Unsafe mechanics
1305
1306	<	private static final sun.misc.Unsafe UNSAFE = getUnsafe();
1307	<	private static final long headOffset =
1308	<	objectFieldOffset(UNSAFE, "head", LinkedTransferQueue.class);
1309	<	private static final long tailOffset =
1310	<	objectFieldOffset(UNSAFE, "tail", LinkedTransferQueue.class);
1169	<	private static final long cleanMeOffset =
1170	<	objectFieldOffset(UNSAFE, "cleanMe", LinkedTransferQueue.class);
1171	<
1172	<	static long objectFieldOffset(sun.misc.Unsafe UNSAFE,
1173	<	String field, Class<?> klazz) {
1306	>	private static final sun.misc.Unsafe UNSAFE;
1307	>	private static final long headOffset;
1308	>	private static final long tailOffset;
1309	>	private static final long sweepVotesOffset;
1310	>	static {
1311		try {
1312	<	return UNSAFE.objectFieldOffset(klazz.getDeclaredField(field));
1313	<	} catch (NoSuchFieldException e) {
1314	<	// Convert Exception to corresponding Error
1315	<	NoSuchFieldError error = new NoSuchFieldError(field);
1316	<	error.initCause(e);
1317	<	throw error;
1312	>	UNSAFE = getUnsafe();
1313	>	Class<?> k = LinkedTransferQueue.class;
1314	>	headOffset = UNSAFE.objectFieldOffset
1315	>	(k.getDeclaredField("head"));
1316	>	tailOffset = UNSAFE.objectFieldOffset
1317	>	(k.getDeclaredField("tail"));
1318	>	sweepVotesOffset = UNSAFE.objectFieldOffset
1319	>	(k.getDeclaredField("sweepVotes"));
1320	>	} catch (Exception e) {
1321	>	throw new Error(e);
1322		}
1323		}
1324
1325	<	private static sun.misc.Unsafe getUnsafe() {
1325	>	/**
1326	>	* Returns a sun.misc.Unsafe.  Suitable for use in a 3rd party package.
1327	>	* Replace with a simple call to Unsafe.getUnsafe when integrating
1328	>	* into a jdk.
1329	>	*
1330	>	* @return a sun.misc.Unsafe
1331	>	*/
1332	>	static sun.misc.Unsafe getUnsafe() {
1333		try {
1334		return sun.misc.Unsafe.getUnsafe();
1335		} catch (SecurityException se) {

Diff Legend

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