*   ...))`
     *
     * So it makes sense for the parameter (and thus the return type) to be @CheckForNull.
     *
     * Two other points:
     *
     * 1. We'll want to use something like @PolyNull once we can make that work for the various
     * platforms we target.
     *

*   Fixes a floating point exception in `SparseDenseCwiseDiv`
    ([CVE-2021-37636](https://cve.mitre.org/cgi-bin/cvename.cgi?name=CVE-2021-37636))
*   Fixes a null pointer dereference in `CompressElement`
    ([CVE-2021-37637](https://cve.mitre.org/cgi-bin/cvename.cgi?name=CVE-2021-37637))
*   Fixes a null pointer dereference in `RaggedTensorToTensor`

    /*
     * We keep an array of elements and counts.  Periodically -- when we need more room in the
     * array, or when we're building, or the like -- we sort, deduplicate, and combine the counts.
     * Negative counts indicate a setCount operation with ~counts[i].
     */

    private final Comparator<? super E> comparator;

    @VisibleForTesting E[] elements;
    private int[] counts;

    /*

	HostCNIBinDir      = "/host/opt/cni/bin"
	ServiceAccountPath = "/var/run/secrets/kubernetes.io/serviceaccount"
	// Well-known subpath we will mount any needed host-mounts under,
	// to preclude shadowing or breaking any pod-internal mounts
	HostMountsPath = "/host"

	for i, testCase := range testCases {
		// initialize HTTP NewRecorder, this records any mutations to response writer inside the handler.
		rec := httptest.NewRecorder()
		// construct HTTP request for Get Object end point.
		req, err := newTestSignedRequestV4(http.MethodHead, getHeadObjectURL("", testCase.bucketName, testCase.objectName),
			0, nil, testCase.accessKey, testCase.secretKey, nil)
		if err != nil {

  /**
   * Returns a {@code Collector} that accumulates elements into a multiset created via the specified
   * {@code Supplier}, whose elements are the result of applying {@code elementFunction} to the
   * inputs, with counts equal to the result of applying {@code countFunction} to the inputs.
   * Elements are added in encounter order.
   *
   * <p>If the mapped elements contain duplicates (according to {@link Object#equals}), the element

   *     }
   * };
   * Multimap<String, String> transformed =
   *     Multimaps.transformValues(multimap, square);
   *   System.out.println(transformed);
   * }</pre>
   *
   * ... prints {@code {a=[4, 16], b=[9, 9], c=[36]}}.
   *
   * <p>Changes in the underlying multimap are reflected in this view. Conversely, this view
   * supports removal operations, and these are reflected in the underlying multimap.

The key updated docs are:

* Python Types Intro:
    * [Type Hints with Metadata Annotations](https://fastapi.tiangolo.com/python-types/#type-hints-with-metadata-annotations).
* Tutorial:

	// apiCalls should be placed first so alignment is guaranteed for atomic operations.
	apiCalls     [storageMetricLast]uint64
	apiLatencies [storageMetricLast]*lockedLastMinuteLatency
	diskID       atomic.Pointer[string]
	storage      *xlStorage
	health       *diskHealthTracker
	healthCheck  bool

	metricsCache *cachevalue.Cache[DiskMetrics]
	diskCtx      context.Context
	diskCancel   context.CancelFunc
}

	// For hardrives it is possible to set this to a lower value to avoid any
	// spike in latency. But currently we are simply keeping it optimal for SSDs.

	// bigFileThreshold is the point where we add readahead to put operations.
	bigFileThreshold = 128 * humanize.MiByte

	// XL metadata file carries per object metadata.
	xlStorageFormatFile = "xl.meta"