[ViewVC] Diff of: jsr166/jsr166/src/jsr166y/LinkedTransferQueue.java

Comparing jsr166/src/jsr166y/LinkedTransferQueue.java (file contents):
Revision 1.7 by dl, Wed Sep 24 10:48:43 2008 UTC vs.
Revision 1.59 by jsr166, Thu Oct 29 00:29:16 2009 UTC

+ */
+package jsr166y;
-+
+import java.util.concurrent.*;
-–
+import java.util.concurrent.locks.*;
-–
+import java.util.concurrent.atomic.*;
-–
+import java.util.*;
-–
+import java.io.*;
-–
+import sun.misc.Unsafe;
-–
+import java.lang.reflect.*;
-+
+import java.util.AbstractQueue;
-+
+import java.util.Collection;
-+
+import java.util.ConcurrentModificationException;
-+
+import java.util.Iterator;
-+
+import java.util.NoSuchElementException;
-+
+import java.util.Queue;
-+
+import java.util.concurrent.locks.LockSupport;
+/**
-<
+ * An unbounded {@linkplain TransferQueue} based on linked nodes.
->
+ * An unbounded {@link TransferQueue} based on linked nodes.
+ * This queue orders elements FIFO (first-in-first-out) with respect
+ * to any given producer.  The <em>head</em> of the queue is that
+ * element that has been on the queue the longest time for some
+ * producer.  The <em>tail</em> of the queue is that element that has
+ * been on the queue the shortest time for some producer.
-<
+ * <p>Beware that, unlike in most collections, the <tt>size</tt>
->
+ * <p>Beware that, unlike in most collections, the {@code size}
+ * method is <em>NOT</em> a constant-time operation. Because of the
+ * asynchronous nature of these queues, determining the current number
+ * of elements requires a traversal of the elements.
+ * @since 1.7
+ * @author Doug Lea
+ * @param <E> the type of elements held in this collection
-–
+ *
+ */
+public class LinkedTransferQueue<E> extends AbstractQueue<E>
+    implements TransferQueue<E>, java.io.Serializable {
+    private static final long serialVersionUID = -3223113410248163686L;
+    /*
-<
+     * This is still a work in progress...
->
+     * *** Overview of Dual Queues with Slack ***
->
+     *
->
+     * Dual Queues, introduced by Scherer and Scott
->
+     * (http://www.cs.rice.edu/~wns1/papers/2004-DISC-DDS.pdf) are
->
+     * (linked) queues in which nodes may represent either data or
->
+     * requests.  When a thread tries to enqueue a data node, but
->
+     * encounters a request node, it instead "matches" and removes it;
->
+     * and vice versa for enqueuing requests. Blocking Dual Queues
->
+     * arrange that threads enqueuing unmatched requests block until
->
+     * other threads provide the match. Dual Synchronous Queues (see
->
+     * Scherer, Lea, & Scott
->
+     * http://www.cs.rochester.edu/u/scott/papers/2009_Scherer_CACM_SSQ.pdf)
->
+     * additionally arrange that threads enqueuing unmatched data also
->
+     * block.  Dual Transfer Queues support all of these modes, as
->
+     * dictated by callers.
->
+     *
->
+     * A FIFO dual queue may be implemented using a variation of the
->
+     * Michael & Scott (M&S) lock-free queue algorithm
->
+     * (http://www.cs.rochester.edu/u/scott/papers/1996_PODC_queues.pdf).
->
+     * It maintains two pointer fields, "head", pointing to a
->
+     * (matched) node that in turn points to the first actual
->
+     * (unmatched) queue node (or null if empty); and "tail" that
->
+     * points to the last node on the queue (or again null if
->
+     * empty). For example, here is a possible queue with four data
->
+     * elements:
->
+     *
->
+     *  head                tail
->
+     *    |                   |
->
+     *    v                   v
->
+     *    M -> U -> U -> U -> U
->
+     *
->
+     * The M&S queue algorithm is known to be prone to scalability and
->
+     * overhead limitations when maintaining (via CAS) these head and
->
+     * tail pointers. This has led to the development of
->
+     * contention-reducing variants such as elimination arrays (see
->
+     * Moir et al http://portal.acm.org/citation.cfm?id=1074013) and
->
+     * optimistic back pointers (see Ladan-Mozes & Shavit
->
+     * http://people.csail.mit.edu/edya/publications/OptimisticFIFOQueue-journal.pdf).
->
+     * However, the nature of dual queues enables a simpler tactic for
->
+     * improving M&S-style implementations when dual-ness is needed.
->
+     *
->
+     * In a dual queue, each node must atomically maintain its match
->
+     * status. While there are other possible variants, we implement
->
+     * this here as: for a data-mode node, matching entails CASing an
->
+     * "item" field from a non-null data value to null upon match, and
->
+     * vice-versa for request nodes, CASing from null to a data
->
+     * value. (Note that the linearization properties of this style of
->
+     * queue are easy to verify -- elements are made available by
->
+     * linking, and unavailable by matching.) Compared to plain M&S
->
+     * queues, this property of dual queues requires one additional
->
+     * successful atomic operation per enq/deq pair. But it also
->
+     * enables lower cost variants of queue maintenance mechanics. (A
->
+     * variation of this idea applies even for non-dual queues that
->
+     * support deletion of interior elements, such as
->
+     * j.u.c.ConcurrentLinkedQueue.)
->
+     *
->
+     * Once a node is matched, its match status can never again
->
+     * change.  We may thus arrange that the linked list of them
->
+     * contain a prefix of zero or more matched nodes, followed by a
->
+     * suffix of zero or more unmatched nodes. (Note that we allow
->
+     * both the prefix and suffix to be zero length, which in turn
->
+     * means that we do not use a dummy header.)  If we were not
->
+     * concerned with either time or space efficiency, we could
->
+     * correctly perform enqueue and dequeue operations by traversing
->
+     * from a pointer to the initial node; CASing the item of the
->
+     * first unmatched node on match and CASing the next field of the
->
+     * trailing node on appends. (Plus some special-casing when
->
+     * initially empty).  While this would be a terrible idea in
->
+     * itself, it does have the benefit of not requiring ANY atomic
->
+     * updates on head/tail fields.
->
+     *
->
+     * We introduce here an approach that lies between the extremes of
->
+     * never versus always updating queue (head and tail) pointers.
->
+     * This offers a tradeoff between sometimes requiring extra
->
+     * traversal steps to locate the first and/or last unmatched
->
+     * nodes, versus the reduced overhead and contention of fewer
->
+     * updates to queue pointers. For example, a possible snapshot of
->
+     * a queue is:
->
+     *
->
+     *  head           tail
->
+     *    |              |
->
+     *    v              v
->
+     *    M -> M -> U -> U -> U -> U
->
+     *
->
+     * The best value for this "slack" (the targeted maximum distance
->
+     * between the value of "head" and the first unmatched node, and
->
+     * similarly for "tail") is an empirical matter. We have found
->
+     * that using very small constants in the range of 1-3 work best
->
+     * over a range of platforms. Larger values introduce increasing
->
+     * costs of cache misses and risks of long traversal chains, while
->
+     * smaller values increase CAS contention and overhead.
->
+     *
->
+     * Dual queues with slack differ from plain M&S dual queues by
->
+     * virtue of only sometimes updating head or tail pointers when
->
+     * matching, appending, or even traversing nodes; in order to
->
+     * maintain a targeted slack.  The idea of "sometimes" may be
->
+     * operationalized in several ways. The simplest is to use a
->
+     * per-operation counter incremented on each traversal step, and
->
+     * to try (via CAS) to update the associated queue pointer
->
+     * whenever the count exceeds a threshold. Another, that requires
->
+     * more overhead, is to use random number generators to update
->
+     * with a given probability per traversal step.
->
+     *
->
+     * In any strategy along these lines, because CASes updating
->
+     * fields may fail, the actual slack may exceed targeted
->
+     * slack. However, they may be retried at any time to maintain
->
+     * targets.  Even when using very small slack values, this
->
+     * approach works well for dual queues because it allows all
->
+     * operations up to the point of matching or appending an item
->
+     * (hence potentially allowing progress by another thread) to be
->
+     * read-only, thus not introducing any further contention. As
->
+     * described below, we implement this by performing slack
->
+     * maintenance retries only after these points.
->
+     *
->
+     * As an accompaniment to such techniques, traversal overhead can
->
+     * be further reduced without increasing contention of head
->
+     * pointer updates: Threads may sometimes shortcut the "next" link
->
+     * path from the current "head" node to be closer to the currently
->
+     * known first unmatched node, and similarly for tail. Again, this
->
+     * may be triggered with using thresholds or randomization.
->
+     *
->
+     * These ideas must be further extended to avoid unbounded amounts
->
+     * of costly-to-reclaim garbage caused by the sequential "next"
->
+     * links of nodes starting at old forgotten head nodes: As first
->
+     * described in detail by Boehm
->
+     * (http://portal.acm.org/citation.cfm?doid=503272.503282) if a GC
->
+     * delays noticing that any arbitrarily old node has become
->
+     * garbage, all newer dead nodes will also be unreclaimed.
->
+     * (Similar issues arise in non-GC environments.)  To cope with
->
+     * this in our implementation, upon CASing to advance the head
->
+     * pointer, we set the "next" link of the previous head to point
->
+     * only to itself; thus limiting the length of connected dead lists.
->
+     * (We also take similar care to wipe out possibly garbage
->
+     * retaining values held in other Node fields.)  However, doing so
->
+     * adds some further complexity to traversal: If any "next"
->
+     * pointer links to itself, it indicates that the current thread
->
+     * has lagged behind a head-update, and so the traversal must
->
+     * continue from the "head".  Traversals trying to find the
->
+     * current tail starting from "tail" may also encounter
->
+     * self-links, in which case they also continue at "head".
->
+     *
->
+     * It is tempting in slack-based scheme to not even use CAS for
->
+     * updates (similarly to Ladan-Mozes & Shavit). However, this
->
+     * cannot be done for head updates under the above link-forgetting
->
+     * mechanics because an update may leave head at a detached node.
->
+     * And while direct writes are possible for tail updates, they
->
+     * increase the risk of long retraversals, and hence long garbage
->
+     * chains, which can be much more costly than is worthwhile
->
+     * considering that the cost difference of performing a CAS vs
->
+     * write is smaller when they are not triggered on each operation
->
+     * (especially considering that writes and CASes equally require
->
+     * additional GC bookkeeping ("write barriers") that are sometimes
->
+     * more costly than the writes themselves because of contention).
->
+     *
->
+     * Removal of interior nodes (due to timed out or interrupted
->
+     * waits, or calls to remove(x) or Iterator.remove) can use a
->
+     * scheme roughly similar to that described in Scherer, Lea, and
->
+     * Scott's SynchronousQueue. Given a predecessor, we can unsplice
->
+     * any node except the (actual) tail of the queue. To avoid
->
+     * build-up of cancelled trailing nodes, upon a request to remove
->
+     * a trailing node, it is placed in field "cleanMe" to be
->
+     * unspliced upon the next call to unsplice any other node.
->
+     * Situations needing such mechanics are not common but do occur
->
+     * in practice; for example when an unbounded series of short
->
+     * timed calls to poll repeatedly time out but never otherwise
->
+     * fall off the list because of an untimed call to take at the
->
+     * front of the queue. Note that maintaining field cleanMe does
->
+     * not otherwise much impact garbage retention even if never
->
+     * cleared by some other call because the held node will
->
+     * eventually either directly or indirectly lead to a self-link
->
+     * once off the list.
->
+     *
->
+     * *** Overview of implementation ***
-<
+     * This class extends the approach used in FIFO-mode
-<
+     * SynchronousQueues. See the internal documentation, as well as
-<
+     * the PPoPP 2006 paper "Scalable Synchronous Queues" by Scherer,
-<
+     * Lea & Scott
-<
+     * (http://www.cs.rice.edu/~wns1/papers/2006-PPoPP-SQ.pdf)
->
+     * We use a threshold-based approach to updates, with a slack
->
+     * threshold of two -- that is, we update head/tail when the
->
+     * current pointer appears to be two or more steps away from the
->
+     * first/last node. The slack value is hard-wired: a path greater
->
+     * than one is naturally implemented by checking equality of
->
+     * traversal pointers except when the list has only one element,
->
+     * in which case we keep slack threshold at one. Avoiding tracking
->
+     * explicit counts across method calls slightly simplifies an
->
+     * already-messy implementation. Using randomization would
->
+     * probably work better if there were a low-quality dirt-cheap
->
+     * per-thread one available, but even ThreadLocalRandom is too
->
+     * heavy for these purposes.
-<
+     * The main extension is to provide different Wait modes
-<
+     * for the main "xfer" method that puts or takes items.
-<
+     * These don't impact the basic dual-queue logic, but instead
-<
+     * control whether or how threads block upon insertion
-<
+     * of request or data nodes into the dual queue.
->
+     * With such a small slack threshold value, it is rarely
->
+     * worthwhile to augment this with path short-circuiting; i.e.,
->
+     * unsplicing nodes between head and the first unmatched node, or
->
+     * similarly for tail, rather than advancing head or tail
->
+     * proper. However, it is used (in awaitMatch) immediately before
->
+     * a waiting thread starts to block, as a final bit of helping at
->
+     * a point when contention with others is extremely unlikely
->
+     * (since if other threads that could release it are operating,
->
+     * then the current thread wouldn't be blocking).
->
+     *
->
+     * We allow both the head and tail fields to be null before any
->
+     * nodes are enqueued; initializing upon first append.  This
->
+     * simplifies some other logic, as well as providing more
->
+     * efficient explicit control paths instead of letting JVMs insert
->
+     * implicit NullPointerExceptions when they are null.  While not
->
+     * currently fully implemented, we also leave open the possibility
->
+     * of re-nulling these fields when empty (which is complicated to
->
+     * arrange, for little benefit.)
->
+     *
->
+     * All enqueue/dequeue operations are handled by the single method
->
+     * "xfer" with parameters indicating whether to act as some form
->
+     * of offer, put, poll, take, or transfer (each possibly with
->
+     * timeout). The relative complexity of using one monolithic
->
+     * method outweighs the code bulk and maintenance problems of
->
+     * using separate methods for each case.
->
+     *
->
+     * Operation consists of up to three phases. The first is
->
+     * implemented within method xfer, the second in tryAppend, and
->
+     * the third in method awaitMatch.
->
+     *
->
+     * 1. Try to match an existing node
->
+     *
->
+     *    Starting at head, skip already-matched nodes until finding
->
+     *    an unmatched node of opposite mode, if one exists, in which
->
+     *    case matching it and returning, also if necessary updating
->
+     *    head to one past the matched node (or the node itself if the
->
+     *    list has no other unmatched nodes). If the CAS misses, then
->
+     *    a loop retries advancing head by two steps until either
->
+     *    success or the slack is at most two. By requiring that each
->
+     *    attempt advances head by two (if applicable), we ensure that
->
+     *    the slack does not grow without bound. Traversals also check
->
+     *    if the initial head is now off-list, in which case they
->
+     *    start at the new head.
->
+     *
->
+     *    If no candidates are found and the call was untimed
->
+     *    poll/offer, (argument "how" is NOW) return.
->
+     *
->
+     * 2. Try to append a new node (method tryAppend)
->
+     *
->
+     *    Starting at current tail pointer, find the actual last node
->
+     *    and try to append a new node (or if head was null, establish
->
+     *    the first node). Nodes can be appended only if their
->
+     *    predecessors are either already matched or are of the same
->
+     *    mode. If we detect otherwise, then a new node with opposite
->
+     *    mode must have been appended during traversal, so we must
->
+     *    restart at phase 1. The traversal and update steps are
->
+     *    otherwise similar to phase 1: Retrying upon CAS misses and
->
+     *    checking for staleness.  In particular, if a self-link is
->
+     *    encountered, then we can safely jump to a node on the list
->
+     *    by continuing the traversal at current head.
->
+     *
->
+     *    On successful append, if the call was ASYNC, return.
->
+     *
->
+     * 3. Await match or cancellation (method awaitMatch)
->
+     *
->
+     *    Wait for another thread to match node; instead cancelling if
->
+     *    the current thread was interrupted or the wait timed out. On
->
+     *    multiprocessors, we use front-of-queue spinning: If a node
->
+     *    appears to be the first unmatched node in the queue, it
->
+     *    spins a bit before blocking. In either case, before blocking
->
+     *    it tries to unsplice any nodes between the current "head"
->
+     *    and the first unmatched node.
->
+     *
->
+     *    Front-of-queue spinning vastly improves performance of
->
+     *    heavily contended queues. And so long as it is relatively
->
+     *    brief and "quiet", spinning does not much impact performance
->
+     *    of less-contended queues.  During spins threads check their
->
+     *    interrupt status and generate a thread-local random number
->
+     *    to decide to occasionally perform a Thread.yield. While
->
+     *    yield has underdefined specs, we assume that might it help,
->
+     *    and will not hurt in limiting impact of spinning on busy
->
+     *    systems.  We also use smaller (1/2) spins for nodes that are
->
+     *    not known to be front but whose predecessors have not
->
+     *    blocked -- these "chained" spins avoid artifacts of
->
+     *    front-of-queue rules which otherwise lead to alternating
->
+     *    nodes spinning vs blocking. Further, front threads that
->
+     *    represent phase changes (from data to request node or vice
->
+     *    versa) compared to their predecessors receive additional
->
+     *    chained spins, reflecting longer paths typically required to
->
+     *    unblock threads during phase changes.
+     */
-<
+    // Wait modes for xfer method
-<
+    static final int NOWAIT  = 0;
-<
+    static final int TIMEOUT = 1;
-<
+    static final int WAIT    = 2;
-<
-<
+    /** The number of CPUs, for spin control */
-<
+    static final int NCPUS = Runtime.getRuntime().availableProcessors();
->
+    /** True if on multiprocessor */
->
+    private static final boolean MP =
->
+        Runtime.getRuntime().availableProcessors() > 1;
+    /**
-<
+     * The number of times to spin before blocking in timed waits.
-<
+     * The value is empirically derived -- it works well across a
-<
+     * variety of processors and OSes. Empirically, the best value
-<
+     * seems not to vary with number of CPUs (beyond 2) so is just
-<
+     * a constant.
->
+     * The number of times to spin (with randomly interspersed calls
->
+     * to Thread.yield) on multiprocessor before blocking when a node
->
+     * is apparently the first waiter in the queue.  See above for
->
+     * explanation. Must be a power of two. The value is empirically
->
+     * derived -- it works pretty well across a variety of processors,
->
+     * numbers of CPUs, and OSes.
+     */
-<
+    static final int maxTimedSpins = (NCPUS < 2)? 0 : 32;
->
+    private static final int FRONT_SPINS   = 1 << 7;
+    /**
-<
+     * The number of times to spin before blocking in untimed waits.
-<
+     * This is greater than timed value because untimed waits spin
-<
+     * faster since they don't need to check times on each spin.
->
+     * The number of times to spin before blocking when a node is
->
+     * preceded by another node that is apparently spinning.  Also
->
+     * serves as an increment to FRONT_SPINS on phase changes, and as
->
+     * base average frequency for yielding during spins. Must be a
->
+     * power of two.
+     */
-<
+    static final int maxUntimedSpins = maxTimedSpins * 16;
->
+    private static final int CHAINED_SPINS = FRONT_SPINS >>> 1;
+    /**
-<
+     * The number of nanoseconds for which it is faster to spin
-<
+     * rather than to use timed park. A rough estimate suffices.
->
+     * Queue nodes. Uses Object, not E, for items to allow forgetting
->
+     * them after use.  Relies heavily on Unsafe mechanics to minimize
->
+     * unnecessary ordering constraints: Writes that intrinsically
->
+     * precede or follow CASes use simple relaxed forms.  Other
->
+     * cleanups use releasing/lazy writes.
+     */
-<
+    static final long spinForTimeoutThreshold = 1000L;
-<
-<
+    /**
-<
+     * Node class for LinkedTransferQueue. Opportunistically subclasses from
-<
+     * AtomicReference to represent item. Uses Object, not E, to allow
-<
+     * setting item to "this" after use, to avoid garbage
-<
+     * retention. Similarly, setting the next field to this is used as
-<
+     * sentinel that node is off list.
-<
+     */
-<
+    static final class QNode extends AtomicReference<Object> {
-<
+        volatile QNode next;
-<
+        volatile Thread waiter;       // to control park/unpark
-<
+        final boolean isData;
-<
+        QNode(Object item, boolean isData) {
-<
+            super(item);
->
+    static final class Node<E> {
->
+        final boolean isData;   // false if this is a request node
->
+        volatile Object item;   // initially non-null if isData; CASed to match
->
+        volatile Node<E> next;
->
+        volatile Thread waiter; // null until waiting
->
->
+        // CAS methods for fields
->
+        final boolean casNext(Node<E> cmp, Node<E> val) {
->
+            return UNSAFE.compareAndSwapObject(this, nextOffset, cmp, val);
->
+        }
->
->
+        final boolean casItem(Object cmp, Object val) {
->
+            assert cmp == null || cmp.getClass() != Node.class;
->
+            return UNSAFE.compareAndSwapObject(this, itemOffset, cmp, val);
->
+        }
->
->
+        /**
->
+         * Creates a new node. Uses relaxed write because item can only
->
+         * be seen if followed by CAS.
->
+         */
->
+        Node(E item, boolean isData) {
->
+            UNSAFE.putObject(this, itemOffset, item); // relaxed write
+            this.isData = isData;
-<
+        static final AtomicReferenceFieldUpdater<QNode, QNode>
-<
+            nextUpdater = AtomicReferenceFieldUpdater.newUpdater
-<
+            (QNode.class, QNode.class, "next");
-<
-<
+        boolean casNext(QNode cmp, QNode val) {
-<
+            return nextUpdater.compareAndSet(this, cmp, val);
->
+        /**
->
+         * Links node to itself to avoid garbage retention.  Called
->
+         * only after CASing head field, so uses relaxed write.
->
+         */
->
+        final void forgetNext() {
->
+            UNSAFE.putObject(this, nextOffset, this);
->
+        }
->
->
+        /**
->
+         * Sets item to self (using a releasing/lazy write) and waiter
->
+         * to null, to avoid garbage retention after extracting or
->
+         * cancelling.
->
+         */
->
+        final void forgetContents() {
->
+            UNSAFE.putOrderedObject(this, itemOffset, this);
->
+            UNSAFE.putOrderedObject(this, waiterOffset, null);
->
+        }
->
->
+        /**
->
+         * Returns true if this node has been matched, including the
->
+         * case of artificial matches due to cancellation.
->
+         */
->
+        final boolean isMatched() {
->
+            Object x = item;
->
+            return (x == this) || ((x == null) == isData);
->
+        }
->
->
+        /**
->
+         * Returns true if this is an unmatched request node.
->
+         */
->
+        final boolean isUnmatchedRequest() {
->
+            return !isData && item == null;
->
+        }
->
->
+        /**
->
+         * Returns true if a node with the given mode cannot be
->
+         * appended to this node because this node is unmatched and
->
+         * has opposite data mode.
->
+         */
->
+        final boolean cannotPrecede(boolean haveData) {
->
+            boolean d = isData;
->
+            Object x;
->
+            return d != haveData && (x = item) != this && (x != null) == d;
->
+        }
->
->
+        /**
->
+         * Tries to artificially match a data node -- used by remove.
->
+         */
->
+        final boolean tryMatchData() {
->
+            assert isData;
->
+            Object x = item;
->
+            if (x != null && x != this && casItem(x, null)) {
->
+                LockSupport.unpark(waiter);
->
+                return true;
->
+            }
->
+            return false;
-–
+    }
-<
+    /**
-<
+     * Padded version of AtomicReference used for head, tail and
-<
+     * cleanMe, to alleviate contention across threads CASing one vs
-<
+     * the other.
-<
+     */
-<
+    static final class PaddedAtomicReference<T> extends AtomicReference<T> {
-<
+        // enough padding for 64bytes with 4byte refs
-<
+        Object p0, p1, p2, p3, p4, p5, p6, p7, p8, p9, pa, pb, pc, pd, pe;
-<
+        PaddedAtomicReference(T r) { super(r); }
->
+        // Unsafe mechanics
->
+        private static final sun.misc.Unsafe UNSAFE = getUnsafe();
->
+        private static final long nextOffset =
->
+            objectFieldOffset(UNSAFE, "next", Node.class);
->
+        private static final long itemOffset =
->
+            objectFieldOffset(UNSAFE, "item", Node.class);
->
+        private static final long waiterOffset =
->
+            objectFieldOffset(UNSAFE, "waiter", Node.class);
->
->
+        private static final long serialVersionUID = -3375979862319811754L;
-+
+    /** head of the queue; null until first enqueue */
-+
+    transient volatile Node<E> head;
-<
+    /** head of the queue */
-<
+    private transient final PaddedAtomicReference<QNode> head;
-<
+    /** tail of the queue */
-<
+    private transient final PaddedAtomicReference<QNode> tail;
->
+    /** predecessor of dangling unspliceable node */
->
+    private transient volatile Node<E> cleanMe; // decl here reduces contention
-<
+    /**
-<
+     * Reference to a cancelled node that might not yet have been
-<
+     * unlinked from queue because it was the last inserted node
-<
+     * when it cancelled.
-<
+     */
-<
+    private transient final PaddedAtomicReference<QNode> cleanMe;
->
+    /** tail of the queue; null until first append */
->
+    private transient volatile Node<E> tail;
-<
+    /**
-<
+     * Tries to cas nh as new head; if successful, unlink
-<
+     * old head's next node to avoid garbage retention.
->
+    // CAS methods for fields
->
+    private boolean casTail(Node<E> cmp, Node<E> val) {
->
+        return UNSAFE.compareAndSwapObject(this, tailOffset, cmp, val);
->
+    }
->
->
+    private boolean casHead(Node<E> cmp, Node<E> val) {
->
+        return UNSAFE.compareAndSwapObject(this, headOffset, cmp, val);
->
+    }
->
->
+    private boolean casCleanMe(Node<E> cmp, Node<E> val) {
->
+        return UNSAFE.compareAndSwapObject(this, cleanMeOffset, cmp, val);
->
+    }
->
->
+    /*
->
+     * Possible values for "how" argument in xfer method. Beware that
->
+     * the order of assigned numerical values matters.
+     */
-<
+    private boolean advanceHead(QNode h, QNode nh) {
-<
+        if (h == head.get() && head.compareAndSet(h, nh)) {
-<
+            h.next = h; // forget old next
-<
+            return true;
-<
+        }
-<
+        return false;
->
+    private static final int NOW     = 0; // for untimed poll, tryTransfer
->
+    private static final int ASYNC   = 1; // for offer, put, add
->
+    private static final int SYNC    = 2; // for transfer, take
->
+    private static final int TIMEOUT = 3; // for timed poll, tryTransfer
->
->
+    @SuppressWarnings("unchecked")
->
+    static <E> E cast(Object item) {
->
+        assert item == null || item.getClass() != Node.class;
->
+        return (E) item;
+    /**
-<
+     * Puts or takes an item. Used for most queue operations (except
-<
+     * poll() and tryTransfer())
-<
+     * @param e the item or if null, signifies that this is a take
-<
+     * @param mode the wait mode: NOWAIT, TIMEOUT, WAIT
->
+     * Implements all queuing methods. See above for explanation.
->
+     *
->
+     * @param e the item or null for take
->
+     * @param haveData true if this is a put, else a take
->
+     * @param how NOW, ASYNC, SYNC, or TIMEOUT
+     * @param nanos timeout in nanosecs, used only if mode is TIMEOUT
-<
+     * @return an item, or null on failure
->
+     * @return an item if matched, else e
->
+     * @throws NullPointerException if haveData mode but e is null
+     */
-<
+    private Object xfer(Object e, int mode, long nanos) {
-<
+        boolean isData = (e != null);
-<
+        QNode s = null;
-<
+        final PaddedAtomicReference<QNode> head = this.head;
-<
+        final PaddedAtomicReference<QNode> tail = this.tail;
->
+    private E xfer(E e, boolean haveData, int how, long nanos) {
->
+        if (haveData && (e == null))
->
+            throw new NullPointerException();
->
+        Node<E> s = null;                     // the node to append, if needed
-<
+        for (;;) {
-<
+            QNode t = tail.get();
-<
+            QNode h = head.get();
->
+        retry: for (;;) {                     // restart on append race
-<
+            if (t != null && (t == h || t.isData == isData)) {
-<
+                if (s == null)
-<
+                    s = new QNode(e, isData);
-<
+                QNode last = t.next;
-<
+                if (last != null) {
-<
+                    if (t == tail.get())
-<
+                        tail.compareAndSet(t, last);
-<
+                }
-<
+                else if (t.casNext(null, s)) {
-<
+                    tail.compareAndSet(t, s);
-<
+                    return awaitFulfill(t, s, e, mode, nanos);
->
+            for (Node<E> h = head, p = h; p != null;) {
->
+                // find & match first node
->
+                boolean isData = p.isData;
->
+                Object item = p.item;
->
+                if (item != p && (item != null) == isData) { // unmatched
->
+                    if (isData == haveData)   // can't match
->
+                        break;
->
+                    if (p.casItem(item, e)) { // match
->
+                        for (Node<E> q = p; q != h;) {
->
+                            Node<E> n = q.next; // update head by 2
->
+                            if (n != null)    // unless singleton
->
+                                q = n;
->
+                            if (head == h && casHead(h, q)) {
->
+                                h.forgetNext();
->
+                                break;
->
+                            }                 // advance and retry
->
+                            if ((h = head)   == null ||
->
+                                (q = h.next) == null || !q.isMatched())
->
+                                break;        // unless slack < 2
->
+                        }
->
+                        LockSupport.unpark(p.waiter);
->
+                        return this.<E>cast(item);
->
+                    }
-+
+                Node<E> n = p.next;
-+
+                p = (p != n) ? n : (h = head); // Use head if p offlist
-<
+            else if (h != null) {
-<
+                QNode first = h.next;
-<
+                if (t == tail.get() && first != null &&
-<
+                    advanceHead(h, first)) {
-<
+                    Object x = first.get();
-<
+                    if (x != first && first.compareAndSet(x, e)) {
-<
+                        LockSupport.unpark(first.waiter);
-<
+                        return isData? e : x;
-<
+                    }
-<
+                }
->
+            if (how >= ASYNC) {               // No matches available
->
+                if (s == null)
->
+                    s = new Node<E>(e, haveData);
->
+                Node<E> pred = tryAppend(s, haveData);
->
+                if (pred == null)
->
+                    continue retry;           // lost race vs opposite mode
->
+                if (how >= SYNC)
->
+                    return awaitMatch(s, pred, e, how, nanos);
-+
+            return e; // not waiting
-–
+    /**
-<
+     * Version of xfer for poll() and tryTransfer, which
-<
+     * simplifies control paths both here and in xfer
->
+     * Tries to append node s as tail.
->
+     *
->
+     * @param s the node to append
->
+     * @param haveData true if appending in data mode
->
+     * @return null on failure due to losing race with append in
->
+     * different mode, else s's predecessor, or s itself if no
->
+     * predecessor
+     */
-<
+    private Object fulfill(Object e) {
-<
+        boolean isData = (e != null);
-<
+        final PaddedAtomicReference<QNode> head = this.head;
-<
+        final PaddedAtomicReference<QNode> tail = this.tail;
-<
-<
+        for (;;) {
-<
+            QNode t = tail.get();
-<
+            QNode h = head.get();
-<
-<
+            if (t != null && (t == h || t.isData == isData)) {
-<
+                QNode last = t.next;
-<
+                if (t == tail.get()) {
-<
+                    if (last != null)
-<
+                        tail.compareAndSet(t, last);
-<
+                    else
-<
+                        return null;
-<
+                }
->
+    private Node<E> tryAppend(Node<E> s, boolean haveData) {
->
+        for (Node<E> t = tail, p = t;;) { // move p to last node and append
->
+            Node<E> n, u;                     // temps for reads of next & tail
->
+            if (p == null && (p = head) == null) {
->
+                if (casHead(null, s))
->
+                    return s;                 // initialize
-<
+            else if (h != null) {
-<
+                QNode first = h.next;
-<
+                if (t == tail.get() &&
-<
+                    first != null &&
-<
+                    advanceHead(h, first)) {
-<
+                    Object x = first.get();
-<
+                    if (x != first && first.compareAndSet(x, e)) {
-<
+                        LockSupport.unpark(first.waiter);
-<
+                        return isData? e : x;
-<
+                    }
->
+            else if (p.cannotPrecede(haveData))
->
+                return null;                  // lost race vs opposite mode
->
+            else if ((n = p.next) != null)    // not last; keep traversing
->
+                p = p != t && t != (u = tail) ? (t = u) : // stale tail
->
+                    (p != n) ? n : null;      // restart if off list
->
+            else if (!p.casNext(null, s))
->
+                p = p.next;                   // re-read on CAS failure
->
+            else {
->
+                if (p != t) {                 // update if slack now >= 2
->
+                    while ((tail != t || !casTail(t, s)) &&
->
+                           (t = tail)   != null &&
->
+                           (s = t.next) != null && // advance and retry
->
+                           (s = s.next) != null && s != t);
-+
+                return p;
+    /**
-<
+     * Spins/blocks until node s is fulfilled or caller gives up,
-<
+     * depending on wait mode.
->
+     * Spins/yields/blocks until node s is matched or caller gives up.
-–
+     * @param pred the predecessor of waiting node
+     * @param s the waiting node
-+
+     * @param pred the predecessor of s, or s itself if it has no
-+
+     * predecessor, or null if unknown (the null case does not occur
-+
+     * in any current calls but may in possible future extensions)
+     * @param e the comparison value for checking match
-<
+     * @param mode mode
->
+     * @param how either SYNC or TIMEOUT
+     * @param nanos timeout value
-<
+     * @return matched item, or s if cancelled
->
+     * @return matched item, or e if unmatched on interrupt or timeout
+     */
-<
+    private Object awaitFulfill(QNode pred, QNode s, Object e,
-<
+                                int mode, long nanos) {
-<
+        if (mode == NOWAIT)
-<
+            return null;
-<
-<
+        long lastTime = (mode == TIMEOUT)? System.nanoTime() : 0;
->
+    private E awaitMatch(Node<E> s, Node<E> pred, E e, int how, long nanos) {
->
+        long lastTime = (how == TIMEOUT) ? System.nanoTime() : 0L;
+        Thread w = Thread.currentThread();
-<
+        int spins = -1; // set to desired spin count below
->
+        int spins = -1; // initialized after first item and cancel checks
->
+        ThreadLocalRandom randomYields = null; // bound if needed
->
+        for (;;) {
-<
+            if (w.isInterrupted())
-<
+                s.compareAndSet(e, s);
-<
+            Object x = s.get();
-<
+            if (x != e) {                 // Node was matched or cancelled
-<
+                advanceHead(pred, s);     // unlink if head
-<
+                if (x == s)               // was cancelled
-<
+                    return clean(pred, s);
-<
+                else if (x != null) {
-<
+                    s.set(s);             // avoid garbage retention
-<
+                    return x;
-<
+                }
-<
+                else
-<
+                    return e;
->
+            Object item = s.item;
->
+            if (item != e) {                  // matched
->
+                assert item != s;
->
+                s.forgetContents();           // avoid garbage
->
+                return this.<E>cast(item);
->
+            }
->
+            if ((w.isInterrupted() || (how == TIMEOUT && nanos <= 0)) &&
->
+                    s.casItem(e, s)) {       // cancel
->
+                unsplice(pred, s);
->
+                return e;
-<
+            if (mode == TIMEOUT) {
->
+            if (spins < 0) {                  // establish spins at/near front
->
+                if ((spins = spinsFor(pred, s.isData)) > 0)
->
+                    randomYields = ThreadLocalRandom.current();
->
+            }
->
+            else if (spins > 0) {             // spin
->
+                if (--spins == 0)
->
+                    shortenHeadPath();        // reduce slack before blocking
->
+                else if (randomYields.nextInt(CHAINED_SPINS) == 0)
->
+                    Thread.yield();           // occasionally yield
->
+            }
->
+            else if (s.waiter == null) {
->
+                s.waiter = w;                 // request unpark then recheck
->
+            }
->
+            else if (how == TIMEOUT) {
+                long now = System.nanoTime();
-<
+                nanos -= now - lastTime;
->
+                if ((nanos -= now - lastTime) > 0)
->
+                    LockSupport.parkNanos(this, nanos);
+                lastTime = now;
-–
+                if (nanos <= 0) {
-–
+                    s.compareAndSet(e, s); // try to cancel
-–
+                    continue;
-–
+                }
-<
+            if (spins < 0) {
-<
+                QNode h = head.get(); // only spin if at head
-<
+                spins = ((h != null && h.next == s) ?
-<
+                         (mode == TIMEOUT?
-<
+                          maxTimedSpins : maxUntimedSpins) : 0);
-<
+            }
-<
+            if (spins > 0)
-<
+                --spins;
-<
+            else if (s.waiter == null)
-<
+                s.waiter = w;
-<
+            else if (mode != TIMEOUT) {
-<
+                //                LockSupport.park(this);
-<
+                LockSupport.park(); // allows run on java5
->
+            else {
->
+                LockSupport.park(this);
+                s.waiter = null;
-<
+                spins = -1;
->
+                spins = -1;                   // spin if front upon wakeup
-<
+            else if (nanos > spinForTimeoutThreshold) {
-<
+                //                LockSupport.parkNanos(this, nanos);
-<
+                LockSupport.parkNanos(nanos);
-<
+                s.waiter = null;
-<
+                spins = -1;
->
+        }
->
+    }
->
->
+    /**
->
+     * Returns spin/yield value for a node with given predecessor and
->
+     * data mode. See above for explanation.
->
+     */
->
+    private static int spinsFor(Node<?> pred, boolean haveData) {
->
+        if (MP && pred != null) {
->
+            if (pred.isData != haveData)      // phase change
->
+                return FRONT_SPINS + CHAINED_SPINS;
->
+            if (pred.isMatched())             // probably at front
->
+                return FRONT_SPINS;
->
+            if (pred.waiter == null)          // pred apparently spinning
->
+                return CHAINED_SPINS;
->
+        }
->
+        return 0;
->
+    }
->
->
+    /**
->
+     * Tries (once) to unsplice nodes between head and first unmatched
->
+     * or trailing node; failing on contention.
->
+     */
->
+    private void shortenHeadPath() {
->
+        Node<E> h, hn, p, q;
->
+        if ((p = h = head) != null && h.isMatched() &&
->
+            (q = hn = h.next) != null) {
->
+            Node<E> n;
->
+            while ((n = q.next) != q) {
->
+                if (n == null || !q.isMatched()) {
->
+                    if (hn != q && h.next == hn)
->
+                        h.casNext(hn, q);
->
+                    break;
->
+                }
->
+                p = q;
->
+                q = n;
-+
+    /* -------------- Traversal methods -------------- */
-+
+    /**
-<
+     * Gets rid of cancelled node s with original predecessor pred.
-<
+     * @return null (to simplify use by callers)
->
+     * Returns the first unmatched node of the given mode, or null if
->
+     * none.  Used by methods isEmpty, hasWaitingConsumer.
+     */
-<
+    private Object clean(QNode pred, QNode s) {
-<
+        Thread w = s.waiter;
-<
+        if (w != null) {             // Wake up thread
-<
+            s.waiter = null;
-<
+            if (w != Thread.currentThread())
-<
+                LockSupport.unpark(w);
->
+    private Node<E> firstOfMode(boolean data) {
->
+        for (Node<E> p = head; p != null; ) {
->
+            if (!p.isMatched())
->
+                return (p.isData == data) ? p : null;
->
+            Node<E> n = p.next;
->
+            p = (n != p) ? n : head;
-+
+        return null;
-+
+    }
-<
+        for (;;) {
-<
+            if (pred.next != s) // already cleaned
-<
+                return null;
-<
+            QNode h = head.get();
-<
+            QNode hn = h.next;   // Absorb cancelled first node as head
-<
+            if (hn != null && hn.next == hn) {
-<
+                advanceHead(h, hn);
-<
+                continue;
-<
+            }
-<
+            QNode t = tail.get();      // Ensure consistent read for tail
-<
+            if (t == h)
-<
+                return null;
-<
+            QNode tn = t.next;
-<
+            if (t != tail.get())
-<
+                continue;
-<
+            if (tn != null) {          // Help advance tail
-<
+                tail.compareAndSet(t, tn);
-<
+                continue;
-<
+            }
-<
+            if (s != t) {             // If not tail, try to unsplice
-<
+                QNode sn = s.next;
-<
+                if (sn == s || pred.casNext(s, sn))
-<
+                    return null;
-<
+            }
-<
+            QNode dp = cleanMe.get();
-<
+            if (dp != null) {    // Try unlinking previous cancelled node
-<
+                QNode d = dp.next;
-<
+                QNode dn;
-<
+                if (d == null ||               // d is gone or
-<
+                    d == dp ||                 // d is off list or
-<
+                    d.get() != d ||            // d not cancelled or
-<
+                    (d != t &&                 // d not tail and
-<
+                     (dn = d.next) != null &&  //   has successor
-<
+                     dn != d &&                //   that is on list
-<
+                     dp.casNext(d, dn)))       // d unspliced
-<
+                    cleanMe.compareAndSet(dp, null);
-<
+                if (dp == pred)
-<
+                    return null;      // s is already saved node
->
+    /**
->
+     * Returns the item in the first unmatched node with isData; or
->
+     * null if none.  Used by peek.
->
+     */
->
+    private E firstDataItem() {
->
+        for (Node<E> p = head; p != null; ) {
->
+            boolean isData = p.isData;
->
+            Object item = p.item;
->
+            if (item != p && (item != null) == isData)
->
+                return isData ? this.<E>cast(item) : null;
->
+            Node<E> n = p.next;
->
+            p = (n != p) ? n : head;
->
+        }
->
+        return null;
->
+    }
->
->
+    /**
->
+     * Traverses and counts unmatched nodes of the given mode.
->
+     * Used by methods size and getWaitingConsumerCount.
->
+     */
->
+    private int countOfMode(boolean data) {
->
+        int count = 0;
->
+        for (Node<E> p = head; p != null; ) {
->
+            if (!p.isMatched()) {
->
+                if (p.isData != data)
->
+                    return 0;
->
+                if (++count == Integer.MAX_VALUE) // saturated
->
+                    break;
->
+            }
->
+            Node<E> n = p.next;
->
+            if (n != p)
->
+                p = n;
->
+            else {
->
+                count = 0;
->
+                p = head;
->
+            }
->
+        }
->
+        return count;
->
+    }
->
->
+    final class Itr implements Iterator<E> {
->
+        private Node<E> nextNode;   // next node to return item for
->
+        private E nextItem;         // the corresponding item
->
+        private Node<E> lastRet;    // last returned node, to support remove
->
->
+        /**
->
+         * Moves to next node after prev, or first node if prev null.
->
+         */
->
+        private void advance(Node<E> prev) {
->
+            lastRet = prev;
->
+            Node<E> p;
->
+            if (prev == null || (p = prev.next) == prev)
->
+                p = head;
->
+            while (p != null) {
->
+                Object item = p.item;
->
+                if (p.isData) {
->
+                    if (item != null && item != p) {
->
+                        nextItem = LinkedTransferQueue.this.<E>cast(item);
->
+                        nextNode = p;
->
+                        return;
->
+                    }
->
+                }
->
+                else if (item == null)
->
+                    break;
->
+                Node<E> n = p.next;
->
+                p = (n != p) ? n : head;
->
+            }
->
+            nextNode = null;
->
+        }
->
->
+        Itr() {
->
+            advance(null);
->
+        }
->
->
+        public final boolean hasNext() {
->
+            return nextNode != null;
->
+        }
->
->
+        public final E next() {
->
+            Node<E> p = nextNode;
->
+            if (p == null) throw new NoSuchElementException();
->
+            E e = nextItem;
->
+            advance(p);
->
+            return e;
->
+        }
->
->
+        public final void remove() {
->
+            Node<E> p = lastRet;
->
+            if (p == null) throw new IllegalStateException();
->
+            lastRet = null;
->
+            findAndRemoveDataNode(p);
->
+        }
->
+    }
->
->
+    /* -------------- Removal methods -------------- */
->
->
+    /**
->
+     * Unsplices (now or later) the given deleted/cancelled node with
->
+     * the given predecessor.
->
+     *
->
+     * @param pred predecessor of node to be unspliced
->
+     * @param s the node to be unspliced
->
+     */
->
+    private void unsplice(Node<E> pred, Node<E> s) {
->
+        s.forgetContents(); // clear unneeded fields
->
+        /*
->
+         * At any given time, exactly one node on list cannot be
->
+         * unlinked -- the last inserted node. To accommodate this, if
->
+         * we cannot unlink s, we save its predecessor as "cleanMe",
->
+         * processing the previously saved version first. Because only
->
+         * one node in the list can have a null next, at least one of
->
+         * node s or the node previously saved can always be
->
+         * processed, so this always terminates.
->
+         */
->
+        if (pred != null && pred != s) {
->
+            while (pred.next == s) {
->
+                Node<E> oldpred = (cleanMe == null) ? null : reclean();
->
+                Node<E> n = s.next;
->
+                if (n != null) {
->
+                    if (n != s)
->
+                        pred.casNext(s, n);
->
+                    break;
->
+                }
->
+                if (oldpred == pred ||      // Already saved
->
+                    ((oldpred == null || oldpred.next == s) &&
->
+                     casCleanMe(oldpred, pred))) {
->
+                    break;
->
+                }
->
+            }
->
+        }
->
+    }
->
->
+    /**
->
+     * Tries to unsplice the deleted/cancelled node held in cleanMe
->
+     * that was previously uncleanable because it was at tail.
->
+     *
->
+     * @return current cleanMe node (or null)
->
+     */
->
+    private Node<E> reclean() {
->
+        /*
->
+         * cleanMe is, or at one time was, predecessor of a cancelled
->
+         * node s that was the tail so could not be unspliced.  If it
->
+         * is no longer the tail, try to unsplice if necessary and
->
+         * make cleanMe slot available.  This differs from similar
->
+         * code in unsplice() because we must check that pred still
->
+         * points to a matched node that can be unspliced -- if not,
->
+         * we can (must) clear cleanMe without unsplicing.  This can
->
+         * loop only due to contention.
->
+         */
->
+        Node<E> pred;
->
+        while ((pred = cleanMe) != null) {
->
+            Node<E> s = pred.next;
->
+            Node<E> n;
->
+            if (s == null || s == pred || !s.isMatched())
->
+                casCleanMe(pred, null); // already gone
->
+            else if ((n = s.next) != null) {
->
+                if (n != s)
->
+                    pred.casNext(s, n);
->
+                casCleanMe(pred, null);
->
+            }
->
+            else
->
+                break;
->
+        }
->
+        return pred;
->
+    }
->
->
+    /**
->
+     * Main implementation of Iterator.remove(). Find
->
+     * and unsplice the given data node.
->
+     */
->
+    final void findAndRemoveDataNode(Node<E> s) {
->
+        assert s.isData;
->
+        if (s.tryMatchData()) {
->
+            for (Node<E> pred = null, p = head; p != null; ) {
->
+                if (p == s) {
->
+                    unsplice(pred, p);
->
+                    break;
->
+                }
->
+                if (p.isUnmatchedRequest())
->
+                    break;
->
+                pred = p;
->
+                if ((p = p.next) == pred) { // stale
->
+                    pred = null;
->
+                    p = head;
->
+                }
-–
+            else if (cleanMe.compareAndSet(null, pred))
-–
+                return null;          // Postpone cleaning s
+    /**
-<
+     * Creates an initially empty <tt>LinkedTransferQueue</tt>.
->
+     * Main implementation of remove(Object)
->
+     */
->
+    private boolean findAndRemove(Object e) {
->
+        if (e != null) {
->
+            for (Node<E> pred = null, p = head; p != null; ) {
->
+                Object item = p.item;
->
+                if (p.isData) {
->
+                    if (item != null && item != p && e.equals(item) &&
->
+                        p.tryMatchData()) {
->
+                        unsplice(pred, p);
->
+                        return true;
->
+                    }
->
+                }
->
+                else if (item == null)
->
+                    break;
->
+                pred = p;
->
+                if ((p = p.next) == pred) { // stale
->
+                    pred = null;
->
+                    p = head;
->
+                }
->
+            }
->
+        }
->
+        return false;
->
+    }
->
->
->
+    /**
->
+     * Creates an initially empty {@code LinkedTransferQueue}.
+     */
+    public LinkedTransferQueue() {
-–
+        QNode dummy = new QNode(null, false);
-–
+        head = new PaddedAtomicReference<QNode>(dummy);
-–
+        tail = new PaddedAtomicReference<QNode>(dummy);
-–
+        cleanMe = new PaddedAtomicReference<QNode>(null);
+    /**
-<
+     * Creates a <tt>LinkedTransferQueue</tt>
->
+     * Creates a {@code LinkedTransferQueue}
+     * initially containing the elements of the given collection,
+     * added in traversal order of the collection's iterator.
-+
+     *
+     * @param c the collection of elements to initially contain
+     * @throws NullPointerException if the specified collection or any
+     *         of its elements are null
+        addAll(c);
-<
+    public void put(E e) throws InterruptedException {
-<
+        if (e == null) throw new NullPointerException();
-<
+        if (Thread.interrupted()) throw new InterruptedException();
-<
+        xfer(e, NOWAIT, 0);
->
+    /**
->
+     * Inserts the specified element at the tail of this queue.
->
+     * As the queue is unbounded, this method will never block.
->
+     *
->
+     * @throws NullPointerException if the specified element is null
->
+     */
->
+    public void put(E e) {
->
+        xfer(e, true, ASYNC, 0);
-<
+    public boolean offer(E e, long timeout, TimeUnit unit)
-<
+        throws InterruptedException {
-<
+        if (e == null) throw new NullPointerException();
-<
+        if (Thread.interrupted()) throw new InterruptedException();
-<
+        xfer(e, NOWAIT, 0);
->
+    /**
->
+     * Inserts the specified element at the tail of this queue.
->
+     * As the queue is unbounded, this method will never block or
->
+     * return {@code false}.
->
+     *
->
+     * @return {@code true} (as specified by
->
+     *  {@link BlockingQueue#offer(Object,long,TimeUnit) BlockingQueue.offer})
->
+     * @throws NullPointerException if the specified element is null
->
+     */
->
+    public boolean offer(E e, long timeout, TimeUnit unit) {
->
+        xfer(e, true, ASYNC, 0);
+        return true;
-+
+    /**
-+
+     * Inserts the specified element at the tail of this queue.
-+
+     * As the queue is unbounded, this method will never return {@code false}.
-+
+     *
-+
+     * @return {@code true} (as specified by
-+
+     *         {@link BlockingQueue#offer(Object) BlockingQueue.offer})
-+
+     * @throws NullPointerException if the specified element is null
-+
+     */
+    public boolean offer(E e) {
-<
+        if (e == null) throw new NullPointerException();
-<
+        xfer(e, NOWAIT, 0);
->
+        xfer(e, true, ASYNC, 0);
+        return true;
-+
+    /**
-+
+     * Inserts the specified element at the tail of this queue.
-+
+     * As the queue is unbounded, this method will never throw
-+
+     * {@link IllegalStateException} or return {@code false}.
-+
+     *
-+
+     * @return {@code true} (as specified by {@link Collection#add})
-+
+     * @throws NullPointerException if the specified element is null
-+
+     */
-+
+    public boolean add(E e) {
-+
+        xfer(e, true, ASYNC, 0);
-+
+        return true;
-+
+    }
-+
-+
+    /**
-+
+     * Transfers the element to a waiting consumer immediately, if possible.
-+
+     *
-+
+     * <p>More precisely, transfers the specified element immediately
-+
+     * if there exists a consumer already waiting to receive it (in
-+
+     * {@link #take} or timed {@link #poll(long,TimeUnit) poll}),
-+
+     * otherwise returning {@code false} without enqueuing the element.
-+
+     *
-+
+     * @throws NullPointerException if the specified element is null
-+
+     */
-+
+    public boolean tryTransfer(E e) {
-+
+        return xfer(e, true, NOW, 0) == null;
-+
+    }
-+
-+
+    /**
-+
+     * Transfers the element to a consumer, waiting if necessary to do so.
-+
+     *
-+
+     * <p>More precisely, transfers the specified element immediately
-+
+     * if there exists a consumer already waiting to receive it (in
-+
+     * {@link #take} or timed {@link #poll(long,TimeUnit) poll}),
-+
+     * else inserts the specified element at the tail of this queue
-+
+     * and waits until the element is received by a consumer.
-+
+     *
-+
+     * @throws NullPointerException if the specified element is null
-+
+     */
+    public void transfer(E e) throws InterruptedException {
-<
+        if (e == null) throw new NullPointerException();
-<
+        if (xfer(e, WAIT, 0) == null) {
-<
+            Thread.interrupted();
->
+        if (xfer(e, true, SYNC, 0) != null) {
->
+            Thread.interrupted(); // failure possible only due to interrupt
+            throw new InterruptedException();
-+
+    /**
-+
+     * Transfers the element to a consumer if it is possible to do so
-+
+     * before the timeout elapses.
-+
+     *
-+
+     * <p>More precisely, transfers the specified element immediately
-+
+     * if there exists a consumer already waiting to receive it (in
-+
+     * {@link #take} or timed {@link #poll(long,TimeUnit) poll}),
-+
+     * else inserts the specified element at the tail of this queue
-+
+     * and waits until the element is received by a consumer,
-+
+     * returning {@code false} if the specified wait time elapses
-+
+     * before the element can be transferred.
-+
+     *
-+
+     * @throws NullPointerException if the specified element is null
-+
+     */
+    public boolean tryTransfer(E e, long timeout, TimeUnit unit)
+        throws InterruptedException {
-<
+        if (e == null) throw new NullPointerException();
-<
+        if (xfer(e, TIMEOUT, unit.toNanos(timeout)) != null)
->
+        if (xfer(e, true, TIMEOUT, unit.toNanos(timeout)) == null)
+            return true;
+        if (!Thread.interrupted())
+            return false;
+        throw new InterruptedException();
-–
+    public boolean tryTransfer(E e) {
-–
+        if (e == null) throw new NullPointerException();
-–
+        return fulfill(e) != null;
-–
+    }
-–
+    public E take() throws InterruptedException {
-<
+        Object e = xfer(null, WAIT, 0);
->
+        E e = xfer(null, false, SYNC, 0);
+        if (e != null)
-<
+            return (E)e;
->
+            return e;
+        Thread.interrupted();
+        throw new InterruptedException();
+    public E poll(long timeout, TimeUnit unit) throws InterruptedException {
-<
+        Object e = xfer(null, TIMEOUT, unit.toNanos(timeout));
->
+        E e = xfer(null, false, TIMEOUT, unit.toNanos(timeout));
+        if (e != null || !Thread.interrupted())
-<
+            return (E)e;
->
+            return e;
+        throw new InterruptedException();
+    public E poll() {
-<
+        return (E)fulfill(null);
->
+        return xfer(null, false, NOW, 0);
-+
+    /**
-+
+     * @throws NullPointerException     {@inheritDoc}
-+
+     * @throws IllegalArgumentException {@inheritDoc}
-+
+     */
+    public int drainTo(Collection<? super E> c) {
+        if (c == null)
+            throw new NullPointerException();
+        return n;
-+
+    /**
-+
+     * @throws NullPointerException     {@inheritDoc}
-+
+     * @throws IllegalArgumentException {@inheritDoc}
-+
+     */
+    public int drainTo(Collection<? super E> c, int maxElements) {
+        if (c == null)
+            throw new NullPointerException();
+        return n;
-–
+    // Traversal-based methods
-–
+    /**
-<
+     * Return head after performing any outstanding helping steps
->
+     * Returns an iterator over the elements in this queue in proper
->
+     * sequence, from head to tail.
->
+     *
->
+     * <p>The returned iterator is a "weakly consistent" iterator that
->
+     * will never throw
->
+     * {@link ConcurrentModificationException ConcurrentModificationException},
->
+     * and guarantees to traverse elements as they existed upon
->
+     * construction of the iterator, and may (but is not guaranteed
->
+     * to) reflect any modifications subsequent to construction.
->
+     *
->
+     * @return an iterator over the elements in this queue in proper sequence
+     */
-–
+    private QNode traversalHead() {
-–
+        for (;;) {
-–
+            QNode t = tail.get();
-–
+            QNode h = head.get();
-–
+            if (h != null && t != null) {
-–
+                QNode last = t.next;
-–
+                QNode first = h.next;
-–
+                if (t == tail.get()) {
-–
+                    if (last != null)
-–
+                        tail.compareAndSet(t, last);
-–
+                    else if (first != null) {
-–
+                        Object x = first.get();
-–
+                        if (x == first)
-–
+                            advanceHead(h, first);
-–
+                        else
-–
+                            return h;
-–
+                    }
-–
+                    else
-–
+                        return h;
-–
+                }
-–
+            }
-–
+        }
-–
+    }
-–
-–
+    public Iterator<E> iterator() {
+        return new Itr();
-–
+    /**
-–
+     * Iterators. Basic strategy is to traverse list, treating
-–
+     * non-data (i.e., request) nodes as terminating list.
-–
+     * Once a valid data node is found, the item is cached
-–
+     * so that the next call to next() will return it even
-–
+     * if subsequently removed.
-–
+     */
-–
+    class Itr implements Iterator<E> {
-–
+        QNode nextNode;    // Next node to return next
-–
+        QNode currentNode; // last returned node, for remove()
-–
+        QNode prevNode;    // predecessor of last returned node
-–
+        E nextItem;        // Cache of next item, once commited to in next
-–
-–
+        Itr() {
-–
+            nextNode = traversalHead();
-–
+            advance();
-–
+        }
-–
-–
+        E advance() {
-–
+            prevNode = currentNode;
-–
+            currentNode = nextNode;
-–
+            E x = nextItem;
-–
-–
+            QNode p = nextNode.next;
-–
+            for (;;) {
-–
+                if (p == null || !p.isData) {
-–
+                    nextNode = null;
-–
+                    nextItem = null;
-–
+                    return x;
-–
+                }
-–
+                Object item = p.get();
-–
+                if (item != p && item != null) {
-–
+                    nextNode = p;
-–
+                    nextItem = (E)item;
-–
+                    return x;
-–
+                }
-–
+                prevNode = p;
-–
+                p = p.next;
-–
+            }
-–
+        }
-–
-–
+        public boolean hasNext() {
-–
+            return nextNode != null;
-–
+        }
-–
-–
+        public E next() {
-–
+            if (nextNode == null) throw new NoSuchElementException();
-–
+            return advance();
-–
+        }
-–
-–
+        public void remove() {
-–
+            QNode p = currentNode;
-–
+            QNode prev = prevNode;
-–
+            if (prev == null || p == null)
-–
+                throw new IllegalStateException();
-–
+            Object x = p.get();
-–
+            if (x != null && x != p && p.compareAndSet(x, p))
-–
+                clean(prev, p);
-–
+        }
-–
+    }
-–
+    public E peek() {
-<
+        for (;;) {
-<
+            QNode h = traversalHead();
-<
+            QNode p = h.next;
-<
+            if (p == null)
-<
+                return null;
-<
+            Object x = p.get();
-<
+            if (p != x) {
-<
+                if (!p.isData)
-<
+                    return null;
-<
+                if (x != null)
-<
+                    return (E)x;
-<
+            }
-<
+        }
->
+        return firstDataItem();
-+
+    /**
-+
+     * Returns {@code true} if this queue contains no elements.
-+
+     *
-+
+     * @return {@code true} if this queue contains no elements
-+
+     */
+    public boolean isEmpty() {
-<
+        for (;;) {
-<
+            QNode h = traversalHead();
-<
+            QNode p = h.next;
-<
+            if (p == null)
-<
+                return true;
-<
+            Object x = p.get();
-<
+            if (p != x) {
-<
+                if (!p.isData)
-<
+                    return true;
-<
+                if (x != null)
-<
+                    return false;
-<
+            }
-<
+        }
->
+        return firstOfMode(true) == null;
+    public boolean hasWaitingConsumer() {
-<
+        for (;;) {
-<
+            QNode h = traversalHead();
-<
+            QNode p = h.next;
-<
+            if (p == null)
-<
+                return false;
-<
+            Object x = p.get();
-<
+            if (p != x)
-<
+                return !p.isData;
-<
+        }
->
+        return firstOfMode(false) != null;
+    /**
+     * Returns the number of elements in this queue.  If this queue
-<
+     * contains more than <tt>Integer.MAX_VALUE</tt> elements, returns
-<
+     * <tt>Integer.MAX_VALUE</tt>.
->
+     * contains more than {@code Integer.MAX_VALUE} elements, returns
->
+     * {@code Integer.MAX_VALUE}.
+     * <p>Beware that, unlike in most collections, this method is
+     * <em>NOT</em> a constant-time operation. Because of the
+     * @return the number of elements in this queue
+     */
+    public int size() {
-<
+        int count = 0;
-<
+        QNode h = traversalHead();
-<
+        for (QNode p = h.next; p != null && p.isData; p = p.next) {
-<
+            Object x = p.get();
-<
+            if (x != null && x != p) {
-<
+                if (++count == Integer.MAX_VALUE) // saturated
-<
+                    break;
-<
+            }
-<
+        }
-<
+        return count;
->
+        return countOfMode(true);
+    public int getWaitingConsumerCount() {
-<
+        int count = 0;
-<
+        QNode h = traversalHead();
-<
+        for (QNode p = h.next; p != null && !p.isData; p = p.next) {
-<
+            if (p.get() == null) {
-<
+                if (++count == Integer.MAX_VALUE)
-<
+                    break;
-<
+            }
-<
+        }
-<
+        return count;
->
+        return countOfMode(false);
->
+    }
->
->
+    /**
->
+     * Removes a single instance of the specified element from this queue,
->
+     * if it is present.  More formally, removes an element {@code e} such
->
+     * that {@code o.equals(e)}, if this queue contains one or more such
->
+     * elements.
->
+     * Returns {@code true} if this queue contained the specified element
->
+     * (or equivalently, if this queue changed as a result of the call).
->
+     *
->
+     * @param o element to be removed from this queue, if present
->
+     * @return {@code true} if this queue changed as a result of the call
->
+     */
->
+    public boolean remove(Object o) {
->
+        return findAndRemove(o);
-+
+    /**
-+
+     * Always returns {@code Integer.MAX_VALUE} because a
-+
+     * {@code LinkedTransferQueue} is not capacity constrained.
-+
+     *
-+
+     * @return {@code Integer.MAX_VALUE} (as specified by
-+
+     *         {@link BlockingQueue#remainingCapacity()})
-+
+     */
+    public int remainingCapacity() {
+        return Integer.MAX_VALUE;
+    /**
-<
+     * Save the state to a stream (that is, serialize it).
->
+     * Saves the state to a stream (that is, serializes it).
-<
+     * @serialData All of the elements (each an <tt>E</tt>) in
->
+     * @serialData All of the elements (each an {@code E}) in
+     * the proper order, followed by a null
+     * @param s the stream
+     */
+    private void writeObject(java.io.ObjectOutputStream s)
+        throws java.io.IOException {
+        s.defaultWriteObject();
-<
+        for (Iterator<E> it = iterator(); it.hasNext(); )
-<
+            s.writeObject(it.next());
->
+        for (E e : this)
->
+            s.writeObject(e);
+        // Use trailing null as sentinel
+        s.writeObject(null);
+    /**
-<
+     * Reconstitute the Queue instance from a stream (that is,
-<
+     * deserialize it).
->
+     * Reconstitutes the Queue instance from a stream (that is,
->
+     * deserializes it).
->
+     *
+     * @param s the stream
+     */
+    private void readObject(java.io.ObjectInputStream s)
+        throws java.io.IOException, ClassNotFoundException {
+        s.defaultReadObject();
-–
+        resetHeadAndTail();
+        for (;;) {
-<
+            E item = (E)s.readObject();
->
+            @SuppressWarnings("unchecked") E item = (E) s.readObject();
+            if (item == null)
+                break;
+            else
-+
+    // Unsafe mechanics
-<
+    // Support for resetting head/tail while deserializing
->
+    private static final sun.misc.Unsafe UNSAFE = getUnsafe();
->
+    private static final long headOffset =
->
+        objectFieldOffset(UNSAFE, "head", LinkedTransferQueue.class);
->
+    private static final long tailOffset =
->
+        objectFieldOffset(UNSAFE, "tail", LinkedTransferQueue.class);
->
+    private static final long cleanMeOffset =
->
+        objectFieldOffset(UNSAFE, "cleanMe", LinkedTransferQueue.class);
-<
+    // Temporary Unsafe mechanics for preliminary release
-<
+    private static final Unsafe _unsafe;
-<
+    private static final long headOffset;
-<
+    private static final long tailOffset;
-<
+    private static final long cleanMeOffset;
-<
+    static {
->
+    static long objectFieldOffset(sun.misc.Unsafe UNSAFE,
->
+                                  String field, Class<?> klazz) {
+        try {
-<
+            if (LinkedTransferQueue.class.getClassLoader() != null) {
-<
+                Field f = Unsafe.class.getDeclaredField("theUnsafe");
-<
+                f.setAccessible(true);
-<
+                _unsafe = (Unsafe)f.get(null);
-<
+            }
-<
+            else
-<
+                _unsafe = Unsafe.getUnsafe();
-<
+            headOffset = _unsafe.objectFieldOffset
-<
+                (LinkedTransferQueue.class.getDeclaredField("head"));
-<
+            tailOffset = _unsafe.objectFieldOffset
-<
+                (LinkedTransferQueue.class.getDeclaredField("tail"));
-<
+            cleanMeOffset = _unsafe.objectFieldOffset
-<
+                (LinkedTransferQueue.class.getDeclaredField("cleanMe"));
-<
+        } catch (Exception e) {
-<
+            throw new RuntimeException("Could not initialize intrinsics", e);
->
+            return UNSAFE.objectFieldOffset(klazz.getDeclaredField(field));
->
+        } catch (NoSuchFieldException e) {
->
+            // Convert Exception to corresponding Error
->
+            NoSuchFieldError error = new NoSuchFieldError(field);
->
+            error.initCause(e);
->
+            throw error;
-<
+    private void resetHeadAndTail() {
-<
+        QNode dummy = new QNode(null, false);
-<
+        _unsafe.putObjectVolatile(this, headOffset, dummy);
-<
+        _unsafe.putObjectVolatile(this, tailOffset, dummy);
-<
+        _unsafe.putObjectVolatile(this, cleanMeOffset,
-<
+                                  new PaddedAtomicReference<QNode>(null));
-<
->
+    /**
->
+     * Returns a sun.misc.Unsafe.  Suitable for use in a 3rd party package.
->
+     * Replace with a simple call to Unsafe.getUnsafe when integrating
->
+     * into a jdk.
->
+     *
->
+     * @return a sun.misc.Unsafe
->
+     */
->
+    static sun.misc.Unsafe getUnsafe() {
->
+        try {
->
+            return sun.misc.Unsafe.getUnsafe();
->
+        } catch (SecurityException se) {
->
+            try {
->
+                return java.security.AccessController.doPrivileged
->
+                    (new java.security
->
+                     .PrivilegedExceptionAction<sun.misc.Unsafe>() {
->
+                        public sun.misc.Unsafe run() throws Exception {
->
+                            java.lang.reflect.Field f = sun.misc
->
+                                .Unsafe.class.getDeclaredField("theUnsafe");
->
+                            f.setAccessible(true);
->
+                            return (sun.misc.Unsafe) f.get(null);
->
+                        }});
->
+            } catch (java.security.PrivilegedActionException e) {
->
+                throw new RuntimeException("Could not initialize intrinsics",
->
+                                           e.getCause());
->
+            }
->
+        }

Diff Legend

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

Comparing jsr166/src/jsr166y/LinkedTransferQueue.java (file contents): Revision 1.7 by dl, Wed Sep 24 10:48:43 2008 UTC vs. Revision 1.59 by jsr166, Thu Oct 29 00:29:16 2009 UTC

Diff Legend

Comparing jsr166/src/jsr166y/LinkedTransferQueue.java (file contents):
Revision 1.7 by dl, Wed Sep 24 10:48:43 2008 UTC vs.
Revision 1.59 by jsr166, Thu Oct 29 00:29:16 2009 UTC