* Iterator}. Except as noted, each method has a corresponding {@link Iterable}-based method in the
 * {@link Iterables} class.
 *
 * <p><i>Performance notes:</i> Unless otherwise noted, all of the iterators produced in this class
 * are <i>lazy</i>, which means that they only advance the backing iteration when absolutely
 * necessary.
 *
 * <p>See the Guava User Guide section on <a href=

  font-style: italic;
}
</style>

<h2 id="introduction">Introduction</h2>

<p>
The Go memory model specifies the conditions under which
reads of a variable in one goroutine can be guaranteed to
observe values produced by writes to the same variable in a different goroutine.
</p>


<h3 id="advice">Advice</h3>

<p>
Programs that modify data being simultaneously accessed by multiple goroutines
must serialize such access.
</p>

 * #expireAfterAccess expireAfterAccess}, {@link #weakKeys weakKeys}, {@link #weakValues
 * weakValues}, or {@link #softValues softValues} perform periodic maintenance.
 *
 * <p>The caches produced by {@code CacheBuilder} are serializable, and the deserialized caches
 * retain all the configuration properties of the original cache. Note that the serialized form does
 * <i>not</i> include cache contents, but only configuration.

 * #expireAfterAccess expireAfterAccess}, {@link #weakKeys weakKeys}, {@link #weakValues
 * weakValues}, or {@link #softValues softValues} perform periodic maintenance.
 *
 * <p>The caches produced by {@code CacheBuilder} are serializable, and the deserialized caches
 * retain all the configuration properties of the original cache. Note that the serialized form does
 * <i>not</i> include cache contents, but only configuration.

  public static final MediaType SHOCKWAVE_FLASH =
      createConstant(APPLICATION_TYPE, "x-shockwave-flash");

  /**
   * {@code skp} files produced by the 3D Modeling software <a
   * href="https://www.sketchup.com/">SketchUp</a>
   *
   * @since 13.0
   */
  public static final MediaType SKETCHUP = createConstant(APPLICATION_TYPE, "vnd.sketchup.skp");

  // unique_ptrs to ensure this happens.
  std::function<void(BackwardFunction*)> backward_function_deleter;
};

// Map from tensor_id to internally-defined operation-id of the operation which
// produced this tensor. A value of -1 means that the tensor was directly
// watched and not the result of any operation in the tape.
using TensorTape = std::unordered_map<int64_t, int64_t>;

// Map from operation-id to tape entry.

   * the same exception (and the function is not invoked).
   *
   * <p>More precisely, the returned {@code Future} takes its result from a {@code Future} produced
   * by applying the given {@code AsyncFunction} to the result of the original {@code Future}.
   * Example usage:
   *
   * <pre>{@code
   * ListenableFuture<RowKey> rowKeyFuture = indexService.lookUp(query);

   *
   * <p>The outer list is unmodifiable, but reflects the latest state of the source list. The inner
   * lists are sublist views of the original list, produced on demand using {@link List#subList(int,
   * int)}, and are subject to all the usual caveats about modification as explained in that API.
   *
   * @param list the list to return consecutive sublists of

 * for comparisons. This class is not being deprecated, but we gently encourage you to migrate to
 * streams.
 *
 * <p><i>Performance notes:</i> Unless otherwise noted, all of the iterables produced in this class
 * are <i>lazy</i>, which means that their iterators only advance the backing iteration when
 * absolutely necessary.
 *
 * <p>See the Guava User Guide article on <a href=

  }

 private:
  TFE_Context* context_;
  TFE_CustomDevice device_;
  void* info_;
  string name_;
};

// An adapter which wraps the shape/data produced by C custom devices and uses
// it to implement custom device methods.
class CAPICustomDeviceTensorHandle
    : public tensorflow::CustomDeviceTensorHandle {
 public: