* either case, no one needs to read `futures` for cancellation purposes later. (And
     * cancellation purposes are the main reason to access `futures`, as discussed in its docs.)
     */
    this.futures = null;
  }

  enum ReleaseResourcesReason {
    OUTPUT_FUTURE_DONE,
    ALL_INPUT_FUTURES_PROCESSED,
  }

  /**

     * oldFuture is the previous task's newFuture.
     * outputFuture is the future we return to the caller, a nonCancellationPropagating taskFuture.
     *
     * newFuture is guaranteed to only complete once all tasks previously submitted to this instance
     * have completed - namely after oldFuture is done, and taskFuture has either completed or been

 * Executor, and sequentially such that no two will ever be running at the same time.
 *
 * <p>Tasks submitted to {@link #execute(Runnable)} are executed in FIFO order.
 *
 * <p>The execution of tasks is done by one thread as long as there are tasks left in the queue.
 * When a task is {@linkplain Thread#interrupt interrupted}, execution of subsequent tasks
 * continues. See {@link QueueWorker#workOnQueue} for details.
 *

 * Implementation detail for the internal structure of {@link Range} instances. Represents a unique
 * way of "cutting" a "number line" (actually of instances of type {@code C}, not necessarily
 * "numbers") into two sections; this can be done below a certain value, above a certain value,
 * below all values or above all values. With this object defined in this way, an interval can
 * always be represented by a pair of {@code Cut} instances.
 *

 *
 * <p>In order to allow proper garbage collection of unused locks, the edges of the lock graph are
 * weak references.
 *
 * <p><strong>Performance</strong>
 *
 * <p>The extra bookkeeping done by cycle detecting locks comes at some cost to performance.
 * Benchmarks (as of December 2011) show that:
 *
 * <ul>
 *   <li>for an unnested {@code lock()} and {@code unlock()}, a cycle detecting lock takes 38ns as

      }
      // The future may complete during or before the call to getPendingToString, so we use null
      // as a signal that we should try checking if the future is done again.
      if (!isNullOrEmpty(pendingDescription)) {
        builder.append("PENDING, info=[").append(pendingDescription).append("]");
      } else if (isDone()) {
        addDoneString(builder);
      } else {

    assertEquals(EMPTY_STATS, cache.stats());

    Object one = new Object();
    assertNull(map.put(one, one));
    assertSame(one, map.get(one));
    assertTrue(map.containsKey(one));
    assertTrue(map.containsValue(one));
    Object two = new Object();
    assertSame(one, map.replace(one, two));
    assertTrue(map.containsKey(one));
    assertFalse(map.containsValue(one));
    Object three = new Object();

  }

  /**
   * Invokes {@code future.}{@link Future#get() get()} uninterruptibly.
   *
   * <p>Similar methods:
   *
   * <ul>
   *   <li>To retrieve a result from a {@code Future} that is already done, use {@link
   *       Futures#getDone Futures.getDone}.
   *   <li>To treat {@link InterruptedException} uniformly with other exceptions, use {@link
   *       Futures#getChecked(Future, Class) Futures.getChecked}.

   *
   * <p>Duplicate elements are considered equal. For example, the list [1, 1] will have only one
   * permutation, instead of two. This is why the elements have to implement {@link Comparable}.
   *
   * <p>An empty iterable has only one permutation, which is an empty list.
   *
   * <p>This method is equivalent to {@code Collections2.orderedPermutations(list,
   * Ordering.natural())}.

    assertEqualCharNodes(Iterables.limit(result, 2), "ab");
    assertThat(graph.requestedNodes).containsExactly('a', 'a', 'b');

    // Iterate again to see if calculation is done again
    assertEqualCharNodes(Iterables.limit(result, 2), "ab");
    assertThat(graph.requestedNodes).containsExactly('a', 'a', 'a', 'b', 'b');
  }

  @Test