But you don't have to worry about them either, incoming dicts are converted automatically and your output is converted automatically to JSON too.

## Bodies of arbitrary `dict`s { #bodies-of-arbitrary-dicts }

You can also declare a body as a `dict` with keys of some type and values of some other type.

  /**
   * Returns a {@code Collector} that returns the {@code k} smallest (relative to the specified
   * {@code Comparator}) input elements, in ascending order, as an unmodifiable {@code List}. Ties
   * are broken arbitrarily.
   *
   * <p>For example:
   *
   * {@snippet :
   * Stream.of("foo", "quux", "banana", "elephant")
   *     .collect(least(2, comparingInt(String::length)))
   * // returns {"foo", "quux"}
   * }

Apart from all the fancy words used here, the **Dependency Injection** system is quite simple.

Just functions that look the same as the *path operation functions*.

But still, it is very powerful, and allows you to declare arbitrarily deeply nested dependency "graphs" (trees).

/// tip

All this might not seem as useful with these simple examples.

But you will see how useful it is in the chapters about **security**.

 *
 * <p>To compute quartiles, use {@link #quartiles()} instead of {@link #percentiles()}. To compute
 * arbitrary q-quantiles, use {@link #scale scale(q)}.
 *
 * <p>These examples all take a copy of your dataset. If you have a double array, you are okay with
 * it being arbitrarily reordered, and you want to avoid that copy, you can use {@code
 * computeInPlace} instead of {@code compute}.
 *

                return 33;
            }
        };
        byte[] buf = new byte[128];
        buf[33] = 4; // wordCount
        buf[34] = 0x55; // andxCommand != 0xFF
        // andxOffset arbitrarily > 0
        ServerMessageBlock.writeInt2(80, buf, 36);
        // byteCount position: 33 + 1 + 8 = 42
        ServerMessageBlock.writeInt2(0, buf, 42);

    /*
     * This code performs one System.nanoTime() more than necessary, and in return, the time to
     * execute Queue#drainTo is not added *on top* of waiting for the timeout (which could make
     * the timeout arbitrarily inaccurate, given a queue that is slow to drain).
     */
    long deadline = System.nanoTime() + unit.toNanos(timeout);
    int added = 0;
    while (added < numElements) {

   *       com.google.common.base.Joiner#skipNulls}.
   * </ul>
   *
   * <p>Note that constructors taking a builder object cannot be tested effectively because
   * semantics of builder can be arbitrarily complex. Still, a factory class can be created in the
   * test to facilitate equality testing. For example:
   *
   * <pre>
   * public class FooTest {
   *
   *   private static final class FooFactoryForTest {

    /*
     * This code performs one System.nanoTime() more than necessary, and in return, the time to
     * execute Queue#drainTo is not added *on top* of waiting for the timeout (which could make
     * the timeout arbitrarily inaccurate, given a queue that is slow to drain).
     */
    long deadline = System.nanoTime() + unit.toNanos(timeout);
    int added = 0;
    while (added < numElements) {

    // Service b startup takes at least 353 millis, but starting the timer is delayed by at least
    // 150 milliseconds. so in a perfect world the timing would be 353-150=203ms, but since either
    // of our sleep calls can be arbitrarily delayed we should just assert that there is a time
    // recorded.
    assertThat(startupTimes.get(b)).isNotNull();
  }

  public void testServiceStartStop() {
    Service a = new NoOpService();

   * for {@code minimumBits}, will return identically-behaving {@link HashFunction} instances.
   *
   * @param minimumBits a positive integer. This can be arbitrarily large. The returned {@link
   *     HashFunction} instance may use memory proportional to this integer.
   * @return a hash function, described above, that produces hash codes of length {@code
   *     minimumBits} or greater