But when checking benchmarks and comparisons you should keep the following in mind.

## Benchmarks and speed

When you check the benchmarks, it is common to see several tools of different types compared as equivalent.

Specifically, to see Uvicorn, Starlette and FastAPI compared together (among many other tools).

  /**
   * Puts each 16-bit code unit from the {@link CharSequence} into this sink.
   *
   * <p><b>Warning:</b> This method will produce different output than most other languages do when
   * running on the equivalent input. For cross-language compatibility, use {@link #putString},
   * usually with a charset of UTF-8. For other use cases, use {@code putUnencodedChars}.
   *
   * @since 15.0 (since 11.0 as putString(CharSequence))
   */

    ImmutableMap<Class<? extends B>, B> map = ImmutableMap.<Class<? extends B>, B>of(type, value);
    return new ImmutableClassToInstanceMap<>(map);
  }

  /**
   * Returns a new builder. The generated builder is equivalent to the builder created by the {@link
   * Builder} constructor.
   */
  public static <B> Builder<B> builder() {
    return new Builder<>();
  }

  /**

 * fingerprint at 4.0 microseconds and md5 at 4.5 microseconds.
 *
 * <p>Note to maintainers: This implementation relies on signed arithmetic being bit-wise equivalent
 * to unsigned arithmetic in all cases except:
 *
 * <ul>
 *   <li>comparisons (signed values can be negative)
 *   <li>division (avoided here)
 *   <li>shifting (right shift must be unsigned)
 * </ul>
 *

   * evaluate the integral of the function from 7.0 to 10.0.
   *
   * Using integrals guarantees that the effect of a single acquire(3) is equivalent to {
   * acquire(1); acquire(1); acquire(1); }, or { acquire(2); acquire(1); }, etc, since the integral
   * of the function in [7.0, 10.0] is equivalent to the sum of the integrals of [7.0, 8.0], [8.0,
   * 9.0], [9.0, 10.0] (and so on), no matter what the function is. This guarantees that we handle

          if (e.getHash() == hash
              && entryKey != null
              && map.keyEquivalence.equivalent(key, entryKey)) {
            ValueReference<K, V> valueReference = e.getValueReference();
            V entryValue = valueReference.get();

            RemovalCause cause;
            if (map.valueEquivalence.equivalent(value, entryValue)) {
              cause = RemovalCause.EXPLICIT;

internal fun Headers.Builder.commonSet(
  name: String,
  value: String,
) = apply {
  headersCheckName(name)
  headersCheckValue(value, name)
  removeAll(name)
  commonAddLenient(name, value)
}

/** Equivalent to `build().get(name)`, but potentially faster. */
internal fun Headers.Builder.commonGet(name: String): String? {
  for (i in namesAndValues.size - 2 downTo 0 step 2) {

  static <T extends @Nullable Object> T[] newArray(T[] reference, int length) {
    T[] empty = reference.length == 0 ? reference : Arrays.copyOf(reference, 0);
    return Arrays.copyOf(empty, length);
  }

  /** Equivalent to Arrays.copyOfRange(source, from, to, arrayOfType.getClass()). */
  /*
   * Arrays are a mess from a nullness perspective, and Class instances for object-array types are

  static <T extends @Nullable Object> T[] newArray(T[] reference, int length) {
    T[] empty = reference.length == 0 ? reference : Arrays.copyOf(reference, 0);
    return Arrays.copyOf(empty, length);
  }

  /** Equivalent to Arrays.copyOfRange(source, from, to, arrayOfType.getClass()). */
  /*
   * Arrays are a mess from a nullness perspective, and Class instances for object-array types are

///

Then we can create an `AsyncClient` with the app, and send async requests to it, using `await`.

{* ../../docs_src/async_tests/test_main.py hl[9:12] *}

This is the equivalent to:

```Python
response = client.get('/')
```

...that we used to make our requests with the `TestClient`.

/// tip