ViewVC Help
View File | Revision Log | Show Annotations | Download File | Root Listing
root/jsr166/jsr166/src/jsr166y/CountedCompleter.java
(Generate patch)

Comparing jsr166/src/jsr166y/CountedCompleter.java (file contents):
Revision 1.17 by jsr166, Sat Nov 24 03:49:24 2012 UTC vs.
Revision 1.26 by jsr166, Mon Nov 26 12:08:49 2012 UTC

# Line 15 | Line 15 | package jsr166y;
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}, not just one.
19 > * Unless initialized otherwise, the {@linkplain #getPendingCount pending
20 > * count} starts at zero, but may be (atomically) changed using
21 > * methods {@link #setPendingCount}, {@link #addToPendingCount}, and
22 > * {@link #compareAndSetPendingCount}. Upon invocation of {@link
23   * #tryComplete}, if the pending action count is nonzero, it is
24   * decremented; otherwise, the completion action is performed, and if
25   * this completer itself has a completer, the process is continued
# Line 126 | Line 126 | package jsr166y;
126   *       op.apply(array[lo]);
127   *     tryComplete();
128   *   }
129 < * } }</pre>
129 > * }}</pre>
130   *
131   * This design can be improved by noticing that in the recursive case,
132   * the task has nothing to do after forking its right task, so can
133   * directly invoke its left task before returning. (This is an analog
134   * of tail recursion removal.)  Also, because the task returns upon
135   * executing its left task (rather than falling through to invoke
136 < * tryComplete) the pending count is set to one:
136 > * {@code tryComplete}) the pending count is set to one:
137   *
138   * <pre> {@code
139   * class ForEach<E> ...
# Line 183 | Line 183 | package jsr166y;
183   *
184   * <p><b>Searching.</b> A tree of CountedCompleters can search for a
185   * value or property in different parts of a data structure, and
186 < * report a result in an {@link java.util.concurrent.AtomicReference}
187 < * as soon as one is found. The others can poll the result to avoid
188 < * unnecessary work. (You could additionally {@link #cancel} other
189 < * tasks, but it is usually simpler and more efficient to just let
190 < * them notice that the result is set and if so skip further
191 < * processing.)  Illustrating again with an array using full
186 > * report a result in an {@link
187 > * java.util.concurrent.atomic.AtomicReference AtomicReference} as
188 > * soon as one is found. The others can poll the result to avoid
189 > * unnecessary work. (You could additionally {@linkplain #cancel
190 > * cancel} other tasks, but it is usually simpler and more efficient
191 > * to just let them notice that the result is set and if so skip
192 > * further processing.)  Illustrating again with an array using full
193   * partitioning (again, in practice, leaf tasks will almost always
194   * process more than one element):
195   *
# Line 277 | Line 278 | package jsr166y;
278   *   }
279   *   public void onCompletion(CountedCompleter<?> caller) {
280   *     if (caller != this) {
281 < *      MapReducer<E> child = (MapReducer<E>)caller;
282 < *      MapReducer<E> sib = child.sibling;
283 < *      if (sib == null || sib.result == null)
284 < *        result = child.result;
285 < *      else
286 < *        result = reducer.apply(child.result, sib.result);
281 > *       MapReducer<E> child = (MapReducer<E>)caller;
282 > *       MapReducer<E> sib = child.sibling;
283 > *       if (sib == null || sib.result == null)
284 > *         result = child.result;
285 > *       else
286 > *         result = reducer.apply(child.result, sib.result);
287   *     }
288   *   }
289   *   public E getRawResult() { return result; }
# Line 291 | Line 292 | package jsr166y;
292   *     return new MapReducer<E>(null, array, mapper, reducer,
293   *                              0, array.length).invoke();
294   *   }
295 < * } }</pre>
295 > * }}</pre>
296   *
297   * Here, method {@code onCompletion} takes a form common to many
298   * completion designs that combine results. This callback-style method
# Line 303 | Line 304 | package jsr166y;
304   * distinguishes cases.  Most often, when the caller is {@code this},
305   * no action is necessary. Otherwise the caller argument can be used
306   * (usually via a cast) to supply a value (and/or links to other
307 < * values) to be combined.  Asuuming proper use of pending counts, the
307 > * values) to be combined.  Assuming proper use of pending counts, the
308   * actions inside {@code onCompletion} occur (once) upon completion of
309   * a task and its subtasks. No additional synchronization is required
310   * within this method to ensure thread safety of accesses to fields of
# Line 388 | Line 389 | public abstract class CountedCompleter<T
389       * Creates a new CountedCompleter with the given completer
390       * and initial pending count.
391       *
392 <     * @param completer this tasks completer, or {@code null} if none
392 >     * @param completer this task's completer, or {@code null} if none
393       * @param initialPendingCount the initial pending count
394       */
395      protected CountedCompleter(CountedCompleter<?> completer,
# Line 401 | Line 402 | public abstract class CountedCompleter<T
402       * Creates a new CountedCompleter with the given completer
403       * and an initial pending count of zero.
404       *
405 <     * @param completer this tasks completer, or {@code null} if none
405 >     * @param completer this task's completer, or {@code null} if none
406       */
407      protected CountedCompleter(CountedCompleter<?> completer) {
408          this.completer = completer;
# Line 422 | Line 423 | public abstract class CountedCompleter<T
423  
424      /**
425       * Performs an action when method {@link #tryComplete} is invoked
426 <     * and there are no pending counts, or when the unconditional
426 >     * and the pending count is zero, or when the unconditional
427       * method {@link #complete} is invoked.  By default, this method
428       * does nothing. You can distinguish cases by checking the
429       * identity of the given caller argument. If not equal to {@code
# Line 450 | Line 451 | public abstract class CountedCompleter<T
451       * @param caller the task invoking this method (which may
452       * be this task itself).
453       * @return true if this exception should be propagated to this
454 <     * tasks completer, if one exists.
454 >     * task's completer, if one exists.
455       */
456      public boolean onExceptionalCompletion(Throwable ex, CountedCompleter<?> caller) {
457          return true;
# Line 626 | Line 627 | public abstract class CountedCompleter<T
627       * this task's pending count is non-zero, decrements its pending
628       * count and returns {@code null}.  Otherwise, returns the
629       * completer.  This method can be used as part of a completion
630 <     * traversal loop for homogenous task hierarchies:
630 >     * traversal loop for homogeneous task hierarchies:
631       *
632       * <pre> {@code
633       * for (CountedCompleter<?> c = firstComplete();
# Line 661 | Line 662 | public abstract class CountedCompleter<T
662      }
663  
664      /**
665 <     * Support for FJT exception propagation
665 >     * Supports ForkJoinTask exception propagation.
666       */
667      void internalPropagateException(Throwable ex) {
668          CountedCompleter<?> a = this, s = a;
# Line 671 | Line 672 | public abstract class CountedCompleter<T
672      }
673  
674      /**
675 <     * Implements execution conventions for CountedCompleters
675 >     * Implements execution conventions for CountedCompleters.
676       */
677      protected final boolean exec() {
678          compute();

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines