* is polled. This amortizes the sorting cost over the iteration and requires less memory than
   * sorting all elements at once.
   *
   * <p>Retrieving a single element takes approximately O(log(M)) time, where M is the number of
   * iterators. (Retrieving all elements takes approximately O(N*log(M)) time, where N is the total
   * number of elements.)
   */
  private static final class MergingIterator<T extends @Nullable Object>

  }

  @GwtIncompatible // Shorts.fromByteArray, Shorts.toByteArray
  public void testByteArrayRoundTrips() {
    Random r = new Random(5);
    byte[] b = new byte[Shorts.BYTES];

    // total overkill, but, it takes 0.1 sec so why not...
    for (int i = 0; i < 10000; i++) {
      short num = (short) r.nextInt();
      assertThat(Shorts.fromByteArray(Shorts.toByteArray(num))).isEqualTo(num);

      r.nextBytes(b);

    `Headers` class offers full access to the HTTP headers.

 *  **Okio dependency added.** OkHttp now depends on
    [Okio](https://github.com/square/okio), an I/O library that makes it easier
    to access, store and process data. Using this library internally makes OkHttp
    faster while consuming less memory. You can write a `RequestBody` as an Okio
    `BufferedSink` and a `ResponseBody` as an Okio `BufferedSource`. Standard

 *       invalid URLs.</li>
 * </ul>
 *
 * <p>
 * <b>Usage:</b>
 * </p>
 * <p>
 * The {@code transform} method is the main entry point for transforming an HTML
 * response. It takes a {@link ResponseData} object as input and returns a
 * {@link ResultData} object containing the extracted data and child URLs.
 * </p>
 */
public class HtmlTransformer extends AbstractTransformer {

     *
     * <p>If this returns false, then no query can take more than O(log n).
     *
     * <p>Note that for a RegularImmutableSet with elements with truly random hash codes, contains
     * operations take expected O(1) time but with high probability take O(log n) for at least some
     * element. (https://en.wikipedia.org/wiki/Linear_probing#Analysis)
     *

     *            The Type-2 message which this represents a response to.
     * @param targetName
     *            SPN of the target system, optional
     * @param passwordHash
     *            The NT password hash, takes precedence over password (which is no longer required unless legacy LM
     *            authentication is needed)
     * @param password
     *            The password to use when constructing the response.

	if took > maxDurationSecondsForLog {
		// Log if we took a lot of time to load.
		logger.Info("IAM expired STS purge took %.2fs", took)
	}

	// Store the newly populated map in the iam cache. This takes care of
	// removing stale entries from the existing map.
	cache.iamSTSAccountsMap = stsAccountsFromStore

	stsAccPoliciesFromStore.Range(func(k string, v MappedPolicy) bool {
		cache.iamSTSPolicyMap.Store(k, v)

 *
 * <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.
 *

 *
 * <p>To implement a multimap, a subclass must define the method {@link #createCollection()}, which
 * creates an empty collection of values for a key.
 *
 * <p>The multimap constructor takes a map that has a single entry for each distinct key. When you
 * insert a key-value pair with a key that isn't already in the multimap, {@code

 * Benchmarks (as of December 2011) show that:
 *
 * <ul>
 *   <li>for an unnested {@code lock()} and {@code unlock()}, a cycle detecting lock takes 38ns as
 *       opposed to the 24ns taken by a plain lock.
 *   <li>for nested locking, the cost increases with the depth of the nesting:
 *       <ul>
 *         <li>2 levels: average of 64ns per lock()/unlock()