* return a {@code Iterable<? extends N>}, then you can use a lambda to perform a more general
 * transformation:
 *
 * {@snippet :
 * someGraphAlgorithm(startNode, node -> ImmutableList.of(node.mother(), node.father()));
 * }
 *
 * <p>Graph algorithms that need additional capabilities (accessing both predecessors and
 * successors, iterating over the edges, etc.) should declare their input to be of a type that

# Other Tools

## uv

If uv is available, use it to manage dependencies.

## Ruff

If Ruff is available, use it to lint and format the code. Consider enabling the FastAPI rules.

## ty

If ty is available, use it to check types.

## Asyncer

When needing to run blocking code inside of async functions, or async code inside of blocking functions, suggest using Asyncer.

Prefer it over AnyIO or asyncio.

Install:

    // and friends, and that we will eventually expect to see garbage-collected. The assumption
    // is that the ClassLoader of this test is a URLClassLoader, and that it loads FRQ itself
    // rather than delegating to a parent ClassLoader. If this assumption is violated the test will
    // fail and will need to be rewritten.

    Class<?> frqC = FinalizableReferenceQueue.class;

import org.eclipse.aether.ConfigurationProperties;
import org.eclipse.aether.RepositoryListener;
import org.eclipse.aether.RepositorySystem;
import org.eclipse.aether.RepositorySystemSession;
import org.eclipse.aether.RepositorySystemSession.SessionBuilder;
import org.eclipse.aether.artifact.Artifact;
import org.eclipse.aether.artifact.DefaultArtifact;
import org.eclipse.aether.collection.VersionFilter;

import org.eclipse.aether.DefaultRepositorySystemSession;
import org.eclipse.aether.internal.impl.DefaultChecksumPolicyProvider;
import org.eclipse.aether.internal.impl.DefaultRemoteRepositoryManager;
import org.eclipse.aether.internal.impl.DefaultRepositoryKeyFunctionFactory;
import org.eclipse.aether.internal.impl.DefaultUpdatePolicyAnalyzer;
import org.eclipse.aether.internal.impl.SimpleLocalRepositoryManagerFactory;

        during the python Dataset object creation instead doing it during
        Iterator creation time.
    *   Add `batch_dims` argument to `tf.gather`.
    *   The behavior of `tf.gather` is now correct when `axis=None` and
        `batch_dims<0`.
    *   Update docstring for gather to properly describe the non-empty
        `batch_dims` case.
    *   Removing of dtype in the constructor of initializers and partition_info

   * other.
   */
  private static final Correspondence<Number, Number> FINITE_QUANTILE_CORRESPONDENCE =
      Correspondence.tolerance(ALLOWED_ERROR);

  /**
   * A {@link Correspondence} which accepts either finite values within {@link #ALLOWED_ERROR} of
   * each other or identical non-finite values.
   */

   * other.
   */
  private static final Correspondence<Number, Number> FINITE_QUANTILE_CORRESPONDENCE =
      Correspondence.tolerance(ALLOWED_ERROR);

  /**
   * A {@link Correspondence} which accepts either finite values within {@link #ALLOWED_ERROR} of
   * each other or identical non-finite values.
   */

   * create, cache and return" pattern. For further improvements, use {@link LoadingCache} and its
   * {@link LoadingCache#get(Object) get(K)} method instead of this one.
   *
   * <p>Among the improvements that this method and {@code LoadingCache.get(K)} both provide are:
   *
   * <ul>
   *   <li>{@linkplain LoadingCache#get(Object) awaiting the result of a pending load} rather than
   *       starting a redundant one

                // for each phase, make sure children phases are executed between before and after steps
                lifecycle.allPhases().forEach(phase -> phase.phases().forEach(child -> {
                    steps.get(BEFORE + child.name()).executeAfter(steps.get(BEFORE + phase.name()));
                    steps.get(AFTER + phase.name()).executeAfter(steps.get(AFTER + child.name()));
                }));