class CustomDeviceOpHandler;
class Device;

// LINT.IfChange
// Note: Keep in sync with exported copy of enum in eager/c_api.h.
enum ContextDevicePlacementPolicy {
  // Running operations with input tensors on the wrong device will fail.
  DEVICE_PLACEMENT_EXPLICIT = 0,
  // Copy the tensor to the right device but log a warning.
  DEVICE_PLACEMENT_WARN = 1,
  // Silently copy the tensor, which has a performance cost since the operation

Requirements
------------

OkHttp works on Android 5.0+ (API level 21+) and Java 8+.

OkHttp depends on [Okio][okio] for high-performance I/O and the [Kotlin standard library][kotlin]. Both are small libraries with strong backward-compatibility.

We highly recommend you keep OkHttp up-to-date. As with auto-updating web browsers, staying current
with HTTPS clients is an important defense against potential security problems. [We

    // TODO: move this to invariant checks once the appropriate hook exists?
    super.expectContents(expected);
    for (Entry<K, V> entry : expected) {
      assertEquals(
          "Wrong value for key " + entry.getKey(), entry.getValue(), getMap().get(entry.getKey()));
    }
  }

  protected final void expectReplacement(Entry<K, V> newEntry) {

      return; // there's nothing to test
    }
    @Nullable Object[] params = buildParamList(invokable, paramIndex);
    try {
      @SuppressWarnings("unchecked") // We'll get a runtime exception if the type is wrong.
      Invokable<Object, ?> unsafe = (Invokable<Object, ?>) invokable;
      unsafe.invoke(instance, params);
      Assert.fail(
          "No exception thrown for parameter at index "
              + paramIndex

   * transitionService in the wrong order due to a race. Due to the fact that it is a race this test
   * isn't guaranteed to expose the issue, but it is at least likely to become flaky if the race
   * sneaks back in, and in this case flaky means something is definitely wrong.
   *
   * <p>Before the bug was fixed this test would fail at least 30% of the time.
   */

## Нет времени?<a name="in-a-hurry"></a>

<abbr title="too long; didn't read (основная мысль)"><strong>TL;DR:</strong></abbr>

Допустим, вы используете сторонюю библиотеку, которая требует вызова с ключевым словом `await`:

```Python
results = await some_library()
```

Detalhes sobre a sintaxe `async def` para *funções de operação de rota* e alguns conceitos de código assíncrono, concorrência e paralelismo.

## Com pressa?

<abbr title="muito longo; não li"><strong>TL;DR:</strong></abbr>

Se você estiver utilizando bibliotecas de terceiros que dizem para você chamar as funções com `await`, como:

```Python
results = await some_library()
```

   recover from another I/O failure.
 * Fix problems where spdy/3.1 headers may not have been compressed properly.
 * Fix problems with spdy/3.1 and http/2 where the wrong window size was being
   used.
 * Fix 1.5.0 regression where conditional cache responses could corrupt the
   connection pool.


## Version 1.5.0

_2014-03-07_

      this.path = checkNotNull(path);
      this.options = options.clone();
      this.followLinks = followLinks(this.options);
      // TODO(cgdecker): validate the provided options... for example, just WRITE seems wrong
    }

    private static boolean followLinks(OpenOption[] options) {
      for (OpenOption option : options) {
        if (option == NOFOLLOW_LINKS) {
          return false;
        }
      }

      this.path = checkNotNull(path);
      this.options = options.clone();
      this.followLinks = followLinks(this.options);
      // TODO(cgdecker): validate the provided options... for example, just WRITE seems wrong
    }

    private static boolean followLinks(OpenOption[] options) {
      for (OpenOption option : options) {
        if (option == NOFOLLOW_LINKS) {
          return false;
        }
      }