*
   * <p>For the case of {@link RoundingMode#HALF_EVEN}, this implementation uses the IEEE 754
   * default rounding mode: if the two nearest representable values are equally near, the one with
   * the least significant bit zero is chosen. (In such cases, both of the nearest representable
   * values are even integers; this method returns the one that is a multiple of a greater power of
   * two.)
   *

    // The values here look like 111...11101...010 in binary, where the initial 111...1110 takes
    // up exactly as many bits as can be represented in the significand (24 for float, 53 for
    // double). That final 0 should be rounded up to 1 because the remaining bits make that number
    // slightly nearer.
    long floatConversionTest = 0xfffffe8000000002L;

  @get:JvmName("proxySelector") val proxySelector: ProxySelector,
) {
  /**
   * Returns a URL with the hostname and port of the origin server. The path, query, and fragment of
   * this URL are always empty, since they are not significant for planning a route.
   */
  @get:JvmName("url")
  val url: HttpUrl =
    HttpUrl.Builder()
      .scheme(if (sslSocketFactory != null) "https" else "http")
      .host(uriHost)
      .port(uriPort)

        There is always *some* alternative to adding a new feature to Guava, even if it's just
        forking Guava yourself.


        We want to see that new features have some significant advantage over the alternatives.
        These advantages can take
        [many forms](https://github.com/google/guava/wiki/PhilosophyExplained#utility), but taking

  - Fixing a bug without changing a public API.
  - Fixing or improving documentation.
  - Improvements to Maven configuration.

Guidelines for any code contributions:

  1. Any significant changes should be accompanied by tests. The project already
     has good test coverage, so look at some existing tests if you're unsure
     how to go about it.

Pre-compressed input typically compresses in excess of 2GiB/s per core,
so performance impact should be minimal even if precompressed data is re-compressed.
Decompressing incompressible data has no significant performance impact.

Below is a list of common files and content-types which are typically not suitable for compression.

- Extensions

 | `gz`  | (GZIP)      |
 | `bz2` | (BZIP2)     |

   */
  public abstract long asLong();

  /**
   * If this hashcode has enough bits, returns {@code asLong()}, otherwise returns a {@code long}
   * value with {@code asBytes()} as the least-significant bytes and {@code 0x00} as the remaining
   * most-significant bytes.
   *
   * @since 14.0 (since 11.0 as {@code Hashing.padToLong(HashCode)})
   */
  public abstract long padToLong();

  /**

   */
  public abstract long asLong();

  /**
   * If this hashcode has enough bits, returns {@code asLong()}, otherwise returns a {@code long}
   * value with {@code asBytes()} as the least-significant bytes and {@code 0x00} as the remaining
   * most-significant bytes.
   *
   * @since 14.0 (since 11.0 as {@code Hashing.padToLong(HashCode)})
   */
  public abstract long padToLong();

  /**

D'après la dernière <a href="https://www.jetbrains.com/research/python-developers-survey-2018/#development-tools" class="external-link" target="_blank">Enquête Développeurs Python</a>, cela couvre environ 80% des utilisateurs.

* **Uvicorn** :
    * A les meilleures performances, étant donné qu'il n'a pas beaucoup de code mis-à-part le serveur en lui-même.