"five", 5,
            "six", 6,
            "seven", 7,
            "eight", 8),
        "one",
        1,
        "two",
        2,
        "three",
        3,
        "four",
        4,
        "five",
        5,
        "six",
        6,
        "seven",
        7,
        "eight",
        8);
    assertMapEquals(
        ImmutableMap.of(
            "one", 1,

* Extension points that plugins might expose

h3. The plugin manager may have to deal with particular actions when a plugin is

* installation
* activation
* update
* deactivation
* un-installation

When a plugin is installed we may want to check the license and make users agree. We may also want to unpack the given plugin in a particular location.

     */
    long deadline = System.nanoTime() + unit.toNanos(timeout);
    int added = 0;
    while (added < numElements) {
      // we could rely solely on #poll, but #drainTo might be more efficient when there are multiple
      // elements already available (e.g. LinkedBlockingQueue#drainTo locks only once)
      added += q.drainTo(buffer, numElements - added);

     */
    long deadline = System.nanoTime() + unit.toNanos(timeout);
    int added = 0;
    while (added < numElements) {
      // we could rely solely on #poll, but #drainTo might be more efficient when there are multiple
      // elements already available (e.g. LinkedBlockingQueue#drainTo locks only once)
      added += q.drainTo(buffer, numElements - added);

Keep in mind that this means that the whole contents will be stored in memory. This will work well for small files.

But there are several cases in which you might benefit from using `UploadFile`.

## File Parameters with `UploadFile`

Define a file parameter with a type of `UploadFile`:

//// tab | Python 3.9+

```Python hl_lines="14"

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
  // will be blocked till the copy completes. This is the default policy.

      successor(pred(), succ());
      if (left == null) {
        return right;
      } else if (right == null) {
        return left;
      } else if (left.height >= right.height) {
        AvlNode<E> newTop = pred();
        // newTop is the maximum node in my left subtree
        newTop.left = left.removeMax(newTop);
        newTop.right = right;
        newTop.distinctElements = distinctElements - 1;

  static int compareOrThrow(Comparable left, Comparable right) {
    return left.compareTo(right);
  }

  /** Needed to serialize sorted collections of Ranges. */
  private static class RangeLexOrdering extends Ordering<Range<?>> implements Serializable {
    static final Ordering<?> INSTANCE = new RangeLexOrdering();

    @Override
    public int compare(Range<?> left, Range<?> right) {
      return ComparisonChain.start()

!==s||p.length===d+1)return e;n=e.placement.split('-')[0],i=T(n);var a=e.offsets.popper,l=e.offsets.reference,f=Z,m='left'===n&&f(a.right)>f(l.left)||'right'===n&&f(a.left)<f(l.right)||'top'===n&&f(a.bottom)>f(l.top)||'bottom'===n&&f(a.top)<f(l.bottom),h=f(a.left)<f(o.left),c=f(a.right)>f(o.right),g=f(a.top)<f(o.top),u=f(a.bottom)>f(o.bottom),b='left'===n&&h||'right'===n&&c||'top'===n&&g||'bottom'===n&&u,w=-1!==['top','bottom'].indexOf(n),y=!!t.flipVariations&&(w&&'start'===r&&h||w&&'end'===r&&c...

    INSTANCE;

    @Override
    public int compare(long[] left, long[] right) {
      int minLength = Math.min(left.length, right.length);
      for (int i = 0; i < minLength; i++) {
        int result = Long.compare(left[i], right[i]);
        if (result != 0) {
          return result;
        }
      }
      return left.length - right.length;
    }

    @Override
    public String toString() {