super.begin();
    }
  }

  /**
   * Because Thread.interrupt() can invoke arbitrary code, it can be slow (e.g. perform IO). To
   * protect ourselves from that we want to make sure that tasks don't spin too much waiting for the
   * interrupting thread to complete the protocol.
   */
  /*
   * This test hangs (or maybe is just *very* slow) under Android.
   *

 * which approach to use, <b>don't worry</b> too much about it; just pick something that seems
 * reasonable and it will be fine.
 *
 * <ul>
 *   <li>If checking whether the <i>caller</i> has violated your method or constructor's contract
 *       (such as by passing an invalid argument), use the utilities of the {@link Preconditions}
 *       class instead.

   * invokeAny} which are pending serial execution, even the subset of the tasks that have not yet
   * started execution. It is unclear from the {@code ExecutorService} specification if these should
   * be included, and it's much easier to implement the interpretation that they not be. Finally, a
   * call to {@code shutdown} or {@code shutdownNow} may result in concurrent calls to {@code

   * to reduce cache contention on most processors. Padding is
   * overkill for most Atomics because they are usually irregularly
   * scattered in memory and thus don't interfere much with each
   * other. But Atomic objects residing in arrays will tend to be
   * placed adjacent to each other, and so will most often share
   * cache lines (with a huge negative performance impact) without
   * this precaution.

      // timed out. e.g. if we weren't event able to grab the lock within the timeout we would never
      // even check the guard. I don't think we care too much about this use case but it could lead
      // to a confusing error message.
      throw new TimeoutException("Timed out waiting for " + this + " to reach the RUNNING state.");
    }
  }

  @Override

   *
   * <p><i>Performance notes:</i> while the cartesian product of lists of size {@code m, n, p} is a
   * list of size {@code m x n x p}, its actual memory consumption is much smaller. When the
   * cartesian product is constructed, the input lists are merely copied. Only as the resulting list
   * is iterated are the individual lists created, and these are not retained after iteration.
   *

   * <p>It should be noted that using a {@code Striped<L>} with relatively few stripes, and {@code
   * bulkGet(keys)} with a relative large number of keys can cause an excessive number of shared
   * stripes (much like the birthday paradox, where much fewer than anticipated birthdays are needed
   * for a pair of them to match). Please consider carefully the implications of the number of

 *       does it offer a way to batch multiple events from a single producer together.
 *   <li>It doesn't support backpressure and other features needed for resilience.
 *   <li>It doesn't provide much control of threading.
 *   <li>It doesn't offer much monitoring.
 *   <li>It doesn't propagate exceptions, so apps don't have a way to react to them.
 *   <li>It doesn't interoperate well with RxJava, coroutines, and other more commonly used

      fail();
    } catch (IndexOutOfBoundsException expected) {
    }
  }

  /*
   * Whenever an implementation uses `instanceof` on a parameter instance, the test has to know that
   * (so much for "black box") and try instances that both do and don't pass the check. The "don't"
   * half of that is more awkward to arrange...
   */
  private static <T> Iterable<T> iterable(final Collection<T> collection) {

    assertThat((long) arraysDim1 * arraysDim2).isNotEqualTo((long) (arraysDim1 * arraysDim2));

    long[][] arrays = new long[arraysDim1][];
    // it's shared to avoid using too much memory in tests
    long[] sharedArray = new long[arraysDim2];
    Arrays.fill(arrays, sharedArray);

    try {
      Longs.concat(arrays);
      fail();
    } catch (IllegalArgumentException expected) {