String[] deduped = dedupe(elements);
                return new AbstractSet<String>() {
                  @Override
                  public int size() {
                    return deduped.length;
                  }

                  @Override
                  public Iterator<String> iterator() {
                    return MinimalCollection.of(deduped).iterator();
                  }

private
fun String.mapJavaTypeToKotlinType(): String {
    val javaTypeName = this
    return primitiveTypeStrings[javaTypeName] ?: collectionTypeStrings[javaTypeName] ?: javaTypeName
}


// TODO:kotlin-dsl dedupe with KotlinTypeStrings.primitiveTypeStrings
private
val primitiveTypeStrings =
    mapOf(
        "java.lang.Object" to "Any",
        "java.lang.String" to "String",
        "java.lang.Character" to "Char",

//     optimize the storage system for the default value of 1MB minus compression object sizes. The setting simultaneously
//     affects read from source, block, file, dedup, and object storage backup targets for a specific Veeam Job. When customers
//     create a new backup job and select the object storage or a SOBR as a backup target with this setting, the job default

    R apply(double from, long index);
  }

  /**
   * Returns the last element of the specified stream, or {@link java.util.Optional#empty} if the
   * stream is empty.
   *
   * <p>Equivalent to {@code stream.reduce((a, b) -> b)}, but may perform significantly better. This
   * method's runtime will be between O(log n) and O(n), performing better on <a

   *
   * @param deque the deque to be wrapped in a synchronized view
   * @return a synchronized view of the specified deque
   * @since 15.0
   */
  @J2ktIncompatible // Synchronized
  public static <E extends @Nullable Object> Deque<E> synchronizedDeque(Deque<E> deque) {
    return Synchronized.deque(deque, null);
  }

        stream(tags).of(stream -> stream.sorted().reduce((l, r) -> l + r).ifPresent(v -> buf.append(v)));
        buf.append(CACHE_KEY_SPLITTER);
        stream(roles).of(stream -> stream.sorted().reduce((l, r) -> l + r).ifPresent(v -> buf.append(v)));
        buf.append(CACHE_KEY_SPLITTER);
        stream(fields).of(stream -> stream.sorted().reduce((l, r) -> l + r).ifPresent(v -> buf.append(v)));

import com.google.errorprone.annotations.CanIgnoreReturnValue;
import java.util.Deque;
import java.util.Iterator;
import org.jspecify.annotations.Nullable;

/**
 * A deque which forwards all its method calls to another deque. Subclasses should override one or
 * more methods to modify the behavior of the backing deque as desired per the <a
 * href="http://en.wikipedia.org/wiki/Decorator_pattern">decorator pattern</a>.
 *

/**
 * A pluggable service for processing model objects during model building.
 *
 * <p>This service allows implementations to:</p>
 * <ul>
 *   <li>Pool identical objects to reduce memory footprint</li>
 *   <li>Intern objects for faster equality comparisons</li>
 *   <li>Apply custom optimization strategies</li>
 *   <li>Transform or modify objects during building</li>
 * </ul>
 *

This container branches off from
/tensorflow/tools/tf_sig_build_dockerfiles/. However, since
hermetic CUDA and hermetic Python is now available for Tensorflow, a lot of the
requirements installed on the original container can be removed to reduce the
footprint of the container and make it more reusable across different ML

Doing this provides the following specific benefits:
- **Reduce the API surface** - We no longer need to maintain two methods.
- **Consistency** - All languages have consistent access to the same APIs and ergonomics in the DSL.
- **Reduce the size of the Gradle distribution** - We no longer need to carry multiple standard libraries for different languages.

## Decision