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

 *  Fix: Recover gracefully when a coalesced connection immediately goes unhealthy.

## Version 3.14.2

_2019-05-19_

 *  Fix: Lock in a route when recovering from an HTTP/2 connection error. We had a bug where two

        get() {
            if (this !is FirNamedReference) return null

            // In most cases, we can get the callee name from the callee's candidate symbols. However, there is at least one case where we
            // cannot do so:
            // ```
            // fun x(c: Char) {}
            // fun call(x: kotlin.Int) {
            //   operator fun Int.invoke(a: Int) {}

// or a string representing a sub-field or item. The string will follow one of these four formats:
// 'f:<name>', where <name> is the name of a field in a struct, or key in a map
// 'v:<value>', where <value> is the exact json formatted value of a list item
// 'i:<index>', where <index> is position of a item in a list
// 'k:<keys>', where <keys> is a map of  a list item's key fields to their unique values

| [`NSQ`](#NSQ)                     |                             |                                 |

## Prerequisites

- Install and configure MinIO Server from [here](https://min.io/docs/minio/linux/index.html#procedure).
- Install and configure MinIO Client from [here](https://min.io/docs/minio/linux/reference/minio-mc.html#quickstart).

```
$ mc admin config get myminio | grep notify

 *     .build();
 * System.out.println(url);
 * ```
 *
 * This prints:
 *
 * ```
 * http://who-let-the-dogs.out/_Who%3F_?_Who?_#_Who?_
 * ```
 *
 * When parsing URLs that lack percent encoding where it is required, this class will percent encode
 * the offending characters.
 *
 * ### IDNA Mapping and Punycode encoding
 *
 * Hostnames have different requirements and use a different encoding scheme. It consists of IDNA

    assertThat(server.takeRequest().sequenceNumber).isEqualTo(0)

    // Note that although we have two routes available, we only use one. The retry is permitted
    // because there are routes available, but it chooses the existing connection since it isn't
    // yet considered unhealthy.
    assertThat(server.takeRequest().sequenceNumber).isEqualTo(1)
  }

  @ParameterizedTest

    filesystem.deleteRecursively(cacheDir)
    assertThat(cache["a"]).isNull()
  }

  /**
   * We had a long-lived bug where [DiskLruCache.trimToSize] could infinite loop if entries
   * being edited required deletion for the operation to complete.
   */
  @ParameterizedTest
  @ArgumentsSource(FileSystemParamProvider::class)

  @get:JvmName("certificateChainCleaner")
  val certificateChainCleaner: CertificateChainCleaner?

  /**
   * Default call timeout (in milliseconds). By default there is no timeout for complete calls, but
   * there is for the connect, write, and read actions within a call.
   *
   * For WebSockets and duplex calls the timeout only applies to the initial setup.
   */
  @get:JvmName("callTimeoutMillis")

   * This inserts a bunch of synthetic 250 ms delays into both client and server and confirms that
   * the same delays make it back into the events.
   *
   * We've had bugs where we report an event when we request data rather than when the data actually
   * arrives. https://github.com/square/okhttp/issues/5578
   */
  private fun timeToFirstByte() {
    val applicationInterceptorDelay = 250L

    visibility = ["//visibility:public"],
)

# By default, XLA GPU is compiled into tensorflow when building with
# --config=cuda even when `with_xla_support` is false. The config setting
# here allows us to override the behavior if needed.
config_setting(
    name = "no_xla_deps_in_cuda",
    define_values = {"no_xla_deps_in_cuda": "true"},
    visibility = ["//visibility:public"],
)