Class AggregateFutureState

output/java_guava/1.4.17/AggregateFutureState.java:37–161

A helper which does some thread-safe operations for aggregate futures, which must be implemented differently in GWT. Namely: Lazily initializes a set of seen exceptions Decrements a counter atomically

@GwtCompatible(emulated = true)
abstract class AggregateFutureState {
  // Lazily initialized the first time we see an exception; not released until all the input futures
  // & this future completes. Released when the future releases the reference to the running state
  private volatile Set<Throwable> seenExceptions = null;
  private volatile int remaining;
  private static final AtomicHelper ATOMIC_HELPER;
  private static final Logger log = Logger.getLogger(AggregateFutureState.class.getName());

  static {
    AtomicHelper helper;
    try {
      helper = new SafeAtomicHelper(
                   newUpdater(AggregateFutureState.class, (Class) Set.class, "seenExceptions"),
                   newUpdater(AggregateFutureState.class, "remaining"));
    } catch (Throwable reflectionFailure) {
      // Some Android 5.0.x Samsung devices have bugs in JDK reflection APIs that cause
      // getDeclaredField to throw a NoSuchFieldException when the field is definitely there.
      // For these users fallback to a suboptimal implementation, based on synchronized. This will
      // be a definite performance hit to those users.
      log.log(Level.SEVERE, "SafeAtomicHelper is broken!", reflectionFailure);
      helper = new SynchronizedAtomicHelper();
    }
    ATOMIC_HELPER = helper;
  }

  AggregateFutureState(int remainingFutures) {
    this.remaining = remainingFutures;
  }

  final Set<Throwable> getOrInitSeenExceptions() {
    /*
     * The initialization of seenExceptions has to be more complicated than we'd like. The simple
     * approach would be for each caller CAS it from null to a Set populated with its exception. But
     * there's another race: If the first thread fails with an exception and a second thread
     * immediately fails with the same exception:
     *
     * Thread1: calls setException(), which returns true, context switch before it can CAS
     * seenExceptions to its exception
     *
     * Thread2: calls setException(), which returns false, CASes seenExceptions to its exception,
     * and wrongly believes that its exception is new (leading it to logging it when it shouldn't)
     *
     * Our solution is for threads to CAS seenExceptions from null to a Set population with _the
     * initial exception_, no matter which thread does the work. This ensures that seenExceptions
     * always contains not just the current thread's exception but also the initial thread's.
     */
    Set<Throwable> seenExceptionsLocal = seenExceptions;
    if (seenExceptionsLocal == null) {
      seenExceptionsLocal = newConcurrentHashSet();
      /*
       * Other handleException() callers may see this as soon as we publish it. We need to populate
       * it with the initial failure before we do, or else they may think that the initial failure
       * has never been seen before.
       */
      addInitialException(seenExceptionsLocal);
      ATOMIC_HELPER.compareAndSetSeenExceptions(this, null, seenExceptionsLocal);
      /*