*       maxPermits is equal to warmupPeriod.
   *   <li>When _used_, the time it takes, as explained in the introductory class note, is equal to
   *       the integral of our function, between X permits and X-K permits, assuming we want to
   *       spend K saved permits.
   * </ol>
   *

      if (iterations >= maxIterations) {
        @SuppressWarnings("nullness") // safe because we pass sort() a range that contains real Ts
        T[] castBuffer = (T[]) buffer;
        // We've already taken O(k log k), let's make sure we don't take longer than O(k log k).
        Arrays.sort(castBuffer, left, right + 1, comparator);
        break;
      }
    }
    bufferSize = k;

   * <p>This is identical to {@link #copy(Readable, Appendable)} but optimized for these specific
   * types. CharBuffer has poor performance when being written into or read out of so round tripping
   * all the bytes through the buffer takes a long time. With these specialized types we can just
   * use a char array.
   *
   * @param from the object to read from
   * @param to the object to write to
   * @return the number of characters copied

 *
 * <p>The time taken to compute multiple quantiles on the same dataset using {@link Scale#indexes
 * indexes} is generally less than the total time taken to compute each of them separately, and
 * sometimes much less. For example, on a large enough dataset, computing the 90th and 99th
 * percentiles together takes about 55% as long as computing them separately.
 *

          // sequences - "Western European" text
          return 0x90;
        } else if (userFriendly.matches("(?i)(?:Branch.*Prediction.*Hostile)")) {
          // Defeat branch predictor for: c < 0x80 ; branch taken 50% of the time.
          return 0x100;
        } else if (userFriendly.matches("(?i)(?:Greek|Cyrillic|European|ISO.?8859)")) {
          // Mostly 2-byte UTF-8 sequences - "European" text
          return 0x800;

     * and the arithmetic mean is always higher than the geometric mean.
     *
     * b) this iteration converges to floor(sqrt(x)). In fact, the number of correct digits doubles
     * with each iteration, so this algorithm takes O(log(digits)) iterations.
     *
     * We start out with a double-precision approximation, which may be higher or lower than the
     * true value. Therefore, we perform at least one Newton iteration to get a guess that's

   * // returns {"foo", "quux"}
   * }</pre>
   *
   * <p>This {@code Collector} uses O(k) memory and takes expected time O(n) (worst-case O(n log
   * k)), as opposed to e.g. {@code Stream.sorted(comparator).limit(k)}, which currently takes O(n
   * log n) time and O(n) space.
   *
   * @throws IllegalArgumentException if {@code k < 0}
   * @since 33.2.0 (available since 22.0 in guava-jre)

 * The GWT supersource for this class contains a field for each type.
 *
 * <p>For details about this hack, see {@code GwtSerializationDependencies}, which takes the same
 * approach but with a subclass rather than a superclass.
 *
 * <p>TODO(cpovirk): Consider applying this subclass approach to our other types.
 */
@GwtCompatible(emulated = true)

 * LinkedHashMultimap}. The GWT supersource for this class contains a field for each type.
 *
 * <p>For details about this hack, see {@code GwtSerializationDependencies}, which takes the same
 * approach but with a subclass rather than a superclass.
 *
 * <p>TODO(cpovirk): Consider applying this subclass approach to our other types.
 */
@GwtCompatible(emulated = true)

 * ArrayListMultimap}. The GWT supersource for this class contains a field for each type.
 *
 * <p>For details about this hack, see {@code GwtSerializationDependencies}, which takes the same
 * approach but with a subclass rather than a superclass.
 *
 * <p>TODO(cpovirk): Consider applying this subclass approach to our other types.
 */
@GwtCompatible(emulated = true)