//   have observed 12 micros on 64-bit linux systems to wake up a parked thread). So if the
  //   timeout is small we shouldn't park(). This needs to be traded off with the cpu overhead of
  //   spinning, so we use SPIN_THRESHOLD_NANOS which is what AbstractQueuedSynchronizer uses for
  //   similar purposes.
  // * We want to behave reasonably for timeouts of 0

# kernels. Does not include the implementations of any ops or kernels. Instead,
# the library which loads libtensorflow_framework.so
# (e.g. _pywrap_tensorflow_internal.so for Python, libtensorflow.so for the C
# API) is responsible for registering ops with libtensorflow_framework.so. In
# addition to this core set of ops, user libraries which are loaded (via

  // way in such a visibility issue would surface is most likely as a failure of cancel() to
  // propagate to the input. Cancellation propagation is fundamentally racy so this is fine.
  //
  // Future versions of the JMM may revise safe construction semantics in such a way that we can
  // safely publish these objects and we won't need this whole discussion.

 * The values in the model remain untouched but during the process of building a project notions like inheritance and
 * interpolation can be added. This allows to have an entity which is useful in a runtime while preserving the model so
 * that it can be marshalled and unmarshalled without being tainted by runtime requirements.
 * </p>
 * <p>
 * With changes during 3.2.2 release MavenProject is closer to being immutable after construction with the removal of

 * LoadingCache}</a>, <a
 * href="https://www.javadoc.io/doc/com.github.ben-manes.caffeine/caffeine/latest/com.github.benmanes.caffeine/com/github/benmanes/caffeine/cache/CacheLoader.html">{@code
 * CacheLoader}</a>, etc.), so you can use Caffeine without needing to use any Guava types.
 * Caffeine's types are better than Guava's, especially for <a

 *
 * <pre>{@code
 * public static double sqrt(double value) {
 *   checkArgument(value >= 0, "input is negative: %s", value);
 *   // calculate square root
 * }
 * }</pre>
 *
 * <p>so that a hypothetical bad caller of this method, such as:
 *
 * <pre>{@code
 * void exampleBadCaller() {
 *   double d = sqrt(-1.0);
 * }
 * }</pre>
 *

 *
 * <pre>{@code
 * public static double sqrt(double value) {
 *   checkArgument(value >= 0, "input is negative: %s", value);
 *   // calculate square root
 * }
 * }</pre>
 *
 * <p>so that a hypothetical bad caller of this method, such as:
 *
 * <pre>{@code
 * void exampleBadCaller() {
 *   double d = sqrt(-1.0);
 * }
 * }</pre>
 *

        if (!VALID_MODEL_VERSIONS.contains(superModelVersion)) {
            // Maven 3.x is always using 4.0.0 version to load the supermodel, so
            // do the same when loading a dependency.  The model validator will also
            // check that field later.
            superModelVersion = "4.0.0";
        }

   * are guarded by a lock and operations are applied in batches to avoid lock contention. The
   * penalty of applying the batches is spread across threads so that the amortized cost is slightly
   * higher than performing just the operation without enforcing the capacity constraint.
   *
   * This implementation uses a per-segment queue to record a memento of the additions, removals,

            // and the only FIR that we have for that PSI is an implicit invoke call, that means that
            // `Foo.Bar` is definitely not a property access - otherwise it would have had its own FIR.
            // So, it does not make sense to try to resolve such parts of qualifiers as KtCall
            if ((psi as? KtExpression)?.getPossiblyQualifiedCallExpression() == null) {
                return null
            }