ViewVC Help
View File | Revision Log | Show Annotations | Download File | Root Listing
root/jsr166/jsr166/src/jsr166e/CountedCompleter.java
(Generate patch)

Comparing jsr166/src/jsr166e/CountedCompleter.java (file contents):
Revision 1.29 by jsr166, Tue Feb 5 19:54:06 2013 UTC vs.
Revision 1.30 by dl, Wed Jun 19 14:55:40 2013 UTC

# Line 8 | Line 8 | package jsr166e;
8  
9   /**
10   * A {@link ForkJoinTask} with a completion action performed when
11 < * triggered and there are no remaining pending
12 < * actions. CountedCompleters are in general more robust in the
11 > * triggered and there are no remaining pending actions.
12 > * CountedCompleters are in general more robust in the
13   * presence of subtask stalls and blockage than are other forms of
14   * ForkJoinTasks, but are less intuitive to program.  Uses of
15   * CountedCompleter are similar to those of other completion based
16   * components (such as {@link java.nio.channels.CompletionHandler})
17   * except that multiple <em>pending</em> completions may be necessary
18 < * to trigger the {@link #onCompletion} action, not just one. Unless
19 < * initialized otherwise, the {@link #getPendingCount pending count}
20 < * starts at zero, but may be (atomically) changed using methods
21 < * {@link #setPendingCount}, {@link #addToPendingCount}, and {@link
22 < * #compareAndSetPendingCount}. Upon invocation of {@link
18 > * to trigger the completion action {@link #onCompletion(CountedCompleter)},
19 > * not just one.
20 > * Unless initialized otherwise, the {@linkplain #getPendingCount pending
21 > * count} starts at zero, but may be (atomically) changed using
22 > * methods {@link #setPendingCount}, {@link #addToPendingCount}, and
23 > * {@link #compareAndSetPendingCount}. Upon invocation of {@link
24   * #tryComplete}, if the pending action count is nonzero, it is
25   * decremented; otherwise, the completion action is performed, and if
26   * this completer itself has a completer, the process is continued
# Line 40 | Line 41 | package jsr166e;
41   * <p>A concrete CountedCompleter class must define method {@link
42   * #compute}, that should in most cases (as illustrated below), invoke
43   * {@code tryComplete()} once before returning. The class may also
44 < * optionally override method {@link #onCompletion} to perform an
45 < * action upon normal completion, and method {@link
46 < * #onExceptionalCompletion} to perform an action upon any exception.
44 > * optionally override method {@link #onCompletion(CountedCompleter)}
45 > * to perform an action upon normal completion, and method
46 > * {@link #onExceptionalCompletion(Throwable, CountedCompleter)} to
47 > * perform an action upon any exception.
48   *
49   * <p>CountedCompleters most often do not bear results, in which case
50   * they are normally declared as {@code CountedCompleter<Void>}, and
# Line 63 | Line 65 | package jsr166e;
65   * only as an internal helper for other computations, so its own task
66   * status (as reported in methods such as {@link ForkJoinTask#isDone})
67   * is arbitrary; this status changes only upon explicit invocations of
68 < * {@link #complete}, {@link ForkJoinTask#cancel}, {@link
69 < * ForkJoinTask#completeExceptionally} or upon exceptional completion
70 < * of method {@code compute}. Upon any exceptional completion, the
71 < * exception may be relayed to a task's completer (and its completer,
72 < * and so on), if one exists and it has not otherwise already
73 < * completed. Similarly, cancelling an internal CountedCompleter has
74 < * only a local effect on that completer, so is not often useful.
68 > * {@link #complete}, {@link ForkJoinTask#cancel},
69 > * {@link ForkJoinTask#completeExceptionally(Throwable)} or upon
70 > * exceptional completion of method {@code compute}. Upon any
71 > * exceptional completion, the exception may be relayed to a task's
72 > * completer (and its completer, and so on), if one exists and it has
73 > * not otherwise already completed. Similarly, cancelling an internal
74 > * CountedCompleter has only a local effect on that completer, so is
75 > * not often useful.
76   *
77   * <p><b>Sample Usages.</b>
78   *
# Line 96 | Line 99 | package jsr166e;
99   * improve load balancing. In the recursive case, the second of each
100   * pair of subtasks to finish triggers completion of its parent
101   * (because no result combination is performed, the default no-op
102 < * implementation of method {@code onCompletion} is not overridden). A
103 < * static utility method sets up the base task and invokes it
102 > * implementation of method {@code onCompletion} is not overridden).
103 > * A static utility method sets up the base task and invokes it
104   * (here, implicitly using the {@link ForkJoinPool#commonPool()}).
105   *
106   * <pre> {@code
# Line 152 | Line 155 | package jsr166e;
155   *   }
156   * }</pre>
157   *
158 < * As a further improvement, notice that the left task need not even
159 < * exist.  Instead of creating a new one, we can iterate using the
160 < * original task, and add a pending count for each fork. Additionally,
161 < * because no task in this tree implements an {@link #onCompletion}
162 < * method, {@code tryComplete()} can be replaced with {@link
160 < * #propagateCompletion}.
158 > * As a further improvement, notice that the left task need not even exist.
159 > * Instead of creating a new one, we can iterate using the original task,
160 > * and add a pending count for each fork.  Additionally, because no task
161 > * in this tree implements an {@link #onCompletion(CountedCompleter)} method,
162 > * {@code tryComplete()} can be replaced with {@link #propagateCompletion}.
163   *
164   * <pre> {@code
165   * class ForEach<E> ...
# Line 235 | Line 237 | package jsr166e;
237   *
238   * <p><b>Recording subtasks.</b> CountedCompleter tasks that combine
239   * results of multiple subtasks usually need to access these results
240 < * in method {@link #onCompletion}. As illustrated in the following
240 > * in method {@link #onCompletion(CountedCompleter)}. As illustrated in the following
241   * class (that performs a simplified form of map-reduce where mappings
242   * and reductions are all of type {@code E}), one way to do this in
243   * divide and conquer designs is to have each subtask record its
# Line 336 | Line 338 | package jsr166e;
338   *     while (h - l >= 2) {
339   *       int mid = (l + h) >>> 1;
340   *       addToPendingCount(1);
341 < *       (forks = new MapReducer(this, array, mapper, reducer, mid, h, forks)).fork;
341 > *       (forks = new MapReducer(this, array, mapper, reducer, mid, h, forks)).fork();
342   *       h = mid;
343   *     }
344   *     if (h > l)
# Line 357 | Line 359 | package jsr166e;
359   *
360   * <p><b>Triggers.</b> Some CountedCompleters are themselves never
361   * forked, but instead serve as bits of plumbing in other designs;
362 < * including those in which the completion of one of more async tasks
362 > * including those in which the completion of one or more async tasks
363   * triggers another async task. For example:
364   *
365   * <pre> {@code
# Line 437 | Line 439 | public abstract class CountedCompleter<T
439      }
440  
441      /**
442 <     * Performs an action when method {@link #completeExceptionally}
443 <     * is invoked or method {@link #compute} throws an exception, and
444 <     * this task has not otherwise already completed normally. On
445 <     * entry to this method, this task {@link
446 <     * ForkJoinTask#isCompletedAbnormally}.  The return value of this
447 <     * method controls further propagation: If {@code true} and this
448 <     * task has a completer, then this completer is also completed
449 <     * exceptionally.  The default implementation of this method does
450 <     * nothing except return {@code true}.
442 >     * Performs an action when method {@link
443 >     * #completeExceptionally(Throwable)} is invoked or method {@link
444 >     * #compute} throws an exception, and this task has not already
445 >     * otherwise completed normally. On entry to this method, this task
446 >     * {@link ForkJoinTask#isCompletedAbnormally}.  The return value
447 >     * of this method controls further propagation: If {@code true}
448 >     * and this task has a completer that has not completed, then that
449 >     * completer is also completed exceptionally, with the same
450 >     * exception as this completer.  The default implementation of
451 >     * this method does nothing except return {@code true}.
452       *
453       * @param ex the exception
454       * @param caller the task invoking this method (which may
455       * be this task itself)
456 <     * @return true if this exception should be propagated to this
456 >     * @return {@code true} if this exception should be propagated to this
457       * task's completer, if one exists
458       */
459      public boolean onExceptionalCompletion(Throwable ex, CountedCompleter<?> caller) {
# Line 491 | Line 494 | public abstract class CountedCompleter<T
494       * @param delta the value to add
495       */
496      public final void addToPendingCount(int delta) {
497 <        int c; // note: can replace with intrinsic in jdk8
497 >        int c;
498          do {} while (!U.compareAndSwapInt(this, PENDING, c = pending, c+delta));
499      }
500  
# Line 501 | Line 504 | public abstract class CountedCompleter<T
504       *
505       * @param expected the expected value
506       * @param count the new value
507 <     * @return true if successful
507 >     * @return {@code true} if successful
508       */
509      public final boolean compareAndSetPendingCount(int expected, int count) {
510          return U.compareAndSwapInt(this, PENDING, expected, count);
# Line 535 | Line 538 | public abstract class CountedCompleter<T
538  
539      /**
540       * If the pending count is nonzero, decrements the count;
541 <     * otherwise invokes {@link #onCompletion} and then similarly
542 <     * tries to complete this task's completer, if one exists,
543 <     * else marks this task as complete.
541 >     * otherwise invokes {@link #onCompletion(CountedCompleter)}
542 >     * and then similarly tries to complete this task's completer,
543 >     * if one exists, else marks this task as complete.
544       */
545      public final void tryComplete() {
546          CountedCompleter<?> a = this, s = a;
# Line 556 | Line 559 | public abstract class CountedCompleter<T
559  
560      /**
561       * Equivalent to {@link #tryComplete} but does not invoke {@link
562 <     * #onCompletion} along the completion path: If the pending count
563 <     * is nonzero, decrements the count; otherwise, similarly tries to
564 <     * complete this task's completer, if one exists, else marks this
565 <     * task as complete. This method may be useful in cases where
566 <     * {@code onCompletion} should not, or need not, be invoked for
567 <     * each completer in a computation.
562 >     * #onCompletion(CountedCompleter)} along the completion path:
563 >     * If the pending count is nonzero, decrements the count;
564 >     * otherwise, similarly tries to complete this task's completer, if
565 >     * one exists, else marks this task as complete. This method may be
566 >     * useful in cases where {@code onCompletion} should not, or need
567 >     * not, be invoked for each completer in a computation.
568       */
569      public final void propagateCompletion() {
570          CountedCompleter<?> a = this, s = a;
# Line 578 | Line 581 | public abstract class CountedCompleter<T
581      }
582  
583      /**
584 <     * Regardless of pending count, invokes {@link #onCompletion},
585 <     * marks this task as complete and further triggers {@link
586 <     * #tryComplete} on this task's completer, if one exists.  The
587 <     * given rawResult is used as an argument to {@link #setRawResult}
588 <     * before invoking {@link #onCompletion} or marking this task as
589 <     * complete; its value is meaningful only for classes overriding
590 <     * {@code setRawResult}.
584 >     * Regardless of pending count, invokes
585 >     * {@link #onCompletion(CountedCompleter)}, marks this task as
586 >     * complete and further triggers {@link #tryComplete} on this
587 >     * task's completer, if one exists.  The given rawResult is
588 >     * used as an argument to {@link #setRawResult} before invoking
589 >     * {@link #onCompletion(CountedCompleter)} or marking this task
590 >     * as complete; its value is meaningful only for classes
591 >     * overriding {@code setRawResult}.  This method does not modify
592 >     * the pending count.
593       *
594       * <p>This method may be useful when forcing completion as soon as
595       * any one (versus all) of several subtask results are obtained.
# Line 624 | Line 629 | public abstract class CountedCompleter<T
629      /**
630       * If this task does not have a completer, invokes {@link
631       * ForkJoinTask#quietlyComplete} and returns {@code null}.  Or, if
632 <     * this task's pending count is non-zero, decrements its pending
633 <     * count and returns {@code null}.  Otherwise, returns the
632 >     * the completer's pending count is non-zero, decrements that
633 >     * pending count and returns {@code null}.  Otherwise, returns the
634       * completer.  This method can be used as part of a completion
635       * traversal loop for homogeneous task hierarchies:
636       *
# Line 667 | Line 672 | public abstract class CountedCompleter<T
672      void internalPropagateException(Throwable ex) {
673          CountedCompleter<?> a = this, s = a;
674          while (a.onExceptionalCompletion(ex, s) &&
675 <               (a = (s = a).completer) != null && a.status >= 0)
676 <            a.recordExceptionalCompletion(ex);
675 >               (a = (s = a).completer) != null && a.status >= 0 &&
676 >               a.recordExceptionalCompletion(ex) == EXCEPTIONAL)
677 >            ;
678      }
679  
680      /**

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines