// Even though the contract in System.getProperties() specifies that a value will exist for
      // all of the listed keys, for some reason the "java.compiler" key returns null in some JVMs.
      if (property == JAVA_COMPILER) {
        continue;
      }
      // Removed in Java 9:
      // https://docs.oracle.com/javase/9/migrate/toc.htm#JSMIG-GUID-2C896CA8-927C-4381-A737-B1D81D964B7B

          thrownAtomicReferenceFieldUpdaterFailure = atomicReferenceFieldUpdaterFailure;
          helper = new SynchronizedHelper();
        }
      }
    } else {
      /*
       * We avoid Unsafe, since newer JVMs produce warnings or even errors for attempts to use it.
       *
       * In guava-jre, we avoid Unsafe by using VarHandle instead. But if we have references to

     * Compares cipher suites names like "TLS_RSA_WITH_NULL_MD5" and "SSL_RSA_WITH_NULL_MD5",
     * ignoring the "TLS_" or "SSL_" prefix which is not consistent across platforms. In particular
     * some IBM JVMs use the "SSL_" prefix everywhere whereas Oracle JVMs mix "TLS_" and "SSL_".
     */
    internal val ORDER_BY_NAME =
      object : Comparator<String> {
        override fun compare(
          a: String,
          b: String,
        ): Int {

   * are used from Java 9 or higher (i.e., high enough to trigger the VarHandle code path), such a
   * lookup would fail with an IllegalAccessException. That may then trigger use of Unsafe (possibly
   * with a warning under recent JVMs), or it may fall back even further to
   * AtomicReferenceFieldUpdaterAtomicHelper, which would fail with a similar problem to
   * VarHandleAtomicHelperMaker, forcing us all the way to SynchronizedHelper.
   *

  // safely publish these objects and we won't need this whole discussion.
  // TODO(user,lukes): consider adding volatile to all these fields since in current known JVMs
  // that should resolve the issue. This comes at the cost of adding more write barriers to the
  // implementations.

  private Futures() {}

  /**