throw new UnsupportedOperationException();
  }

  /**
   * {@inheritDoc}
   *
   * <p>This implementation of {@code getAllPresent} lacks any insight into the internal cache data
   * structure, and is thus forced to return the query keys instead of the cached keys. This is only
   * possible with an unsafe cast which requires {@code keys} to actually be of type {@code K}.
   *
   * @since 11.0
   */
  /*

   * write.
   *
   * 2. visibility of the writes to an afterDone() call triggered by cancel():
   *
   * Since these fields are non-final that means that TimeoutFuture is not being 'safely published',
   * thus a motivated caller may be able to expose the reference to another thread that would then
   * call cancel() and be unable to cancel the delegate.

the Program, the distribution of the whole must be on the terms of this
License, whose permissions for other licensees extend to the entire
whole, and thus to each and every part regardless of who wrote it.

Thus, it is not the intent of this section to claim rights or contest
your rights to work written entirely by you; rather, the intent is to
exercise the right to control the distribution of derivative or

 * file) to initialize the <code>Config</code>. The <code>System</code>
 * properties will then populate the <code>Config</code> as well potentially
 * overwriting properties from the file. Thus properties provided on the
 * commandline with the <code>-Dproperty.name=value</code> VM parameter
 * will override properties from the configuration file.
 * <p>
 * There are several ways to set jCIFS properties. See

   * necessary. No observable state associated with this cache is modified until loading completes.
   * Unlike {@link #get}, this method does not throw a checked exception, and thus should only be
   * used in situations where checked exceptions are not thrown by the cache loader.
   *
   * <p>If another call to {@link #get} or {@link #getUnchecked} is currently loading the value for

 * of the <i>next</i> request. I.e., if an expensive task arrives at an idle RateLimiter, it will be
 * granted immediately, but it is the <i>next</i> request that will experience extra throttling,
 * thus paying for the cost of the expensive task.
 *
 * @author Dimitris Andreou
 * @since 13.0
 */
// TODO(user): switch to nano precision. A natural unit of cost is "bytes", and a micro precision

-   Different Python versions
-   Linux, MacOS, and Windows machines (these pool definitions are internal)
-   x86 and arm64
-   CPU-only, or with NVIDIA CUDA support (Linux only), or with TPUs

## How to Test Your Changes to TensorFlow

You may check how your changes will affect TensorFlow by:

1. Creating a PR and observing the presubmit test results
2. Running the CI scripts locally, as explained below

   * returns the elements in the order they are returned by the collection's iterator. The returned
   * array is "safe" in that no references to it are maintained by the collection. The caller is
   * thus free to modify the returned array.
   *
   * <p>This method assumes that the collection size doesn't change while the method is running.
   *
   * <p>TODO(kevinb): support concurrently modified collections?
   *

   * of this method requires a corresponding call to {@link #maximumWeight(long)} prior to calling
   * {@link #build}. Weights are measured and recorded when entries are inserted into the cache, and
   * are thus effectively static during the lifetime of a cache entry.
   *
   * <p>When the weight of an entry is zero it will not be considered for size-based eviction
   * (though it still may be evicted by other means).
   *

### Replication of object version and metadata