could use new flow control capacity before acknowledging it, causing strict HTTP/2 servers to
    fail the call.


## Version 3.12.4

_2019-09-04_

 *  Fix: Don't crash looking up an absent class on certain buggy Android 4.x devices.


## Version 3.12.3

_2019-05-07_

 *  Fix: Permit multipart file names to contain non-ASCII characters.

   * is empty.
   *
   * <p><b>Warning:</b> Typical uses of the resulting iterator may produce an infinite loop. You
   * should use an explicit {@code break} or be certain that you will eventually remove all the
   * elements.
   */
  public static <T extends @Nullable Object> Iterator<T> cycle(Iterable<T> iterable) {
    checkNotNull(iterable);
    return new Iterator<T>() {

   * is empty.
   *
   * <p><b>Warning:</b> Typical uses of the resulting iterator may produce an infinite loop. You
   * should use an explicit {@code break} or be certain that you will eventually remove all the
   * elements.
   */
  public static <T extends @Nullable Object> Iterator<T> cycle(Iterable<T> iterable) {
    checkNotNull(iterable);
    return new Iterator<T>() {

   *
   * <p>This method has no effect (returns {@code this}) when called in GWT: it's unclear whether a
   * precomputed matcher is faster, but it certainly would consume more memory (which doesn't seem
   * like a worthwhile tradeoff in a browser).
   */
  public CharMatcher precomputed() {
    return Platform.precomputeCharMatcher(this);
  }

 * each cache modification, on occasional cache accesses, or on calls to {@link Cache#cleanUp}; such
 * entries may be counted in {@link Cache#size}, but will never be visible to read or write
 * operations.
 *
 * <p>Certain cache configurations will result in the accrual of periodic maintenance tasks which
 * will be performed during write operations, or during occasional read operations in the absence of

            unicodeAllowed = true,
          )
      }

    /**
     * Re-encodes the components of this URL so that it satisfies (obsolete) RFC 2396, which is
     * particularly strict for certain components.
     */
    internal fun reencodeForUri() =
      apply {
        host = host?.replace(Regex("[\"<>^`{|}]"), "")

        for (i in 0 until encodedPathSegments.size) {

      K k6,
      V v6,
      K k7,
      V v7,
      K k8,
      V v8,
      K k9,
      V v9) {
    /*
     * This explicit type parameter works around what seems to be a javac bug in certain
     * configurations: b/339186525#comment6
     */
    return ImmutableSortedMap.fromEntries(
        entryOf(k1, v1),
        entryOf(k2, v2),
        entryOf(k3, v3),
        entryOf(k4, v4),

   *
   * <p>This method has no effect (returns {@code this}) when called in GWT: it's unclear whether a
   * precomputed matcher is faster, but it certainly would consume more memory (which doesn't seem
   * like a worthwhile tradeoff in a browser).
   */
  public CharMatcher precomputed() {
    return Platform.precomputeCharMatcher(this);
  }

  }

  /**
   * We had a bug where we'd perform infinite retries of route that fail with connection shutdown
   * errors. The problem was that the logic that decided whether to reuse a route didn't track
   * certain HTTP/2 errors. https://github.com/square/okhttp/issues/5547
   */
  @Test
  fun noRecoveryFromTwoRefusedStreams() {
    server.enqueue(
      MockResponse
        .Builder()

 * each cache modification, on occasional cache accesses, or on calls to {@link Cache#cleanUp}; such
 * entries may be counted in {@link Cache#size}, but will never be visible to read or write
 * operations.
 *
 * <p>Certain cache configurations will result in the accrual of periodic maintenance tasks which
 * will be performed during write operations, or during occasional read operations in the absence of