count++;
    }
    return count;
  }

  /**
   * Returns an iterator that cycles indefinitely over the elements of {@code iterable}.
   *
   * <p>The returned iterator supports {@code remove()} if the provided iterator does. After {@code
   * remove()} is called, subsequent cycles omit the removed element, which is no longer in {@code

  }

  /**
   * Returns a fluent iterable whose {@code Iterator} cycles indefinitely over the elements of this
   * fluent iterable.
   *
   * <p>That iterator supports {@code remove()} if {@code iterable.iterator()} does. After {@code
   * remove()} is called, subsequent cycles omit the removed element, which is no longer in this

 * </ul>
 *
 * <p>The locks created by a factory instance will detect lock acquisition cycles with locks created
 * by other {@code CycleDetectingLockFactory} instances (except those with {@code Policy.DISABLED}).
 * A lock's behavior when a cycle is detected, however, is defined by the {@code Policy} of the
 * factory that created it. This allows detection of cycles across components while delegating
 * control over lock behavior to individual components.
 *

   *
   * <pre>{@code
   * |--------------|
   * v              |
   * a -> b -> c -> d
   * |         ^
   * |---------|
   * }</pre>
   */
  private static final SuccessorsFunction<Character> TWO_CYCLES_GRAPH =
      createDirectedGraph("ab", "ac", "bc", "cd", "da");

  /**
   * A tree-shaped graph that looks as follows (all edges are directed facing downwards):
   *
   * <pre>{@code
   *        h

  /**
   * The number of other exception types in the cache of known-good exceptions and the number of
   * other {@code ClassValue} entries for the exception type to be tested. This lets us evaluate
   * whether our solution scales to use with multiple exception types and to whether it is affected
   * by other {@code ClassValue} users. Some of the benchmarked implementations don't use one or
   * both of these mechanisms, so they will be unaffected.
   */

    // "HashBiMap.hashTableKToV" and "BiEntry.nextInKToVBucket" references.
    // Under that assumption, the remaining references can be safely marked as @Weak.
    // Using @Weak is necessary to avoid retain-cycles between BiEntry instances on iOS,
    // which would cause memory leaks when non-empty HashBiMap with cyclic BiEntry
    // instances is deallocated.
    @CheckForNull BiEntry<K, V> nextInKToVBucket;

   *
   * <p>That iterator supports {@code remove()} if {@code iterable.iterator()} does. After {@code
   * remove()} is called, subsequent cycles omit the removed element, which is no longer in {@code
   * iterable}. The iterator's {@code hasNext()} method returns {@code true} until {@code iterable}
   * is empty.
   *

 *       {@code $} in its name is a nested class</a>.
 * </ul>
 *
 * <h2>{@code ClassPath} and symlinks</h2>
 *
 * <p>In the case of directory classloaders, symlinks are supported but cycles are not traversed.
 * This guarantees discovery of each unique loadable resource. However, not all possible
 * aliases for resources on cyclic paths will be listed.
 *
 * @author Ben Yu
 * @since 14.0
 */

    }
  }

  @AndroidIncompatible // Path (for symlink creation)

  public void testScanDirectory_symlinkCycle() throws IOException {
    if (isWindows()) {
      return; // TODO: b/136041958 - Can we detect cycles under Windows?
    }
    ClassLoader loader = ClassPathTest.class.getClassLoader();
    // directory with a cycle,
    // /root
    //    /left
    //       /[sibling -> right]
    //    /right

    //     moved at that point. Otherwise, we can rotate the cycle a[1], a[1 + d], a[1 + 2d], etc,
    //     then a[2] etc, and so on until we have rotated all elements. There are gcd(d, n) cycles
    //     in all.
    // (3) "Successive". We can consider that we are exchanging a block of size d (a[0..d-1]) with a
    //     block of size n-d (a[d..n-1]), where in general these blocks have different sizes. If we