Na última pesquisa do desenvolvedor Python ficou claro <a href="https://www.jetbrains.com/research/python-developers-survey-2017/#tools-and-features" class="external-link" target="_blank">que o recurso mais utilizado é o "auto completar"</a>.

Todo o _framework_ **FastAPI** é feito para satisfazer isso. Auto completação funciona em todos os lugares.

    return tearDown;
  }

  // Features

  private final Set<Feature<?>> features = new LinkedHashSet<>();

  /**
   * Configures this builder to produce tests appropriate for the given features. This method may be
   * called more than once to add features in multiple groups.
   */
  @CanIgnoreReturnValue
  public B withFeatures(Feature<?>... features) {
    return withFeatures(Arrays.asList(features));

    assertNotSame(features, FeatureUtil.impliedFeatures(features));
  }

  public void testImpliedFeatures_returnsImpliedFeatures() throws Exception {
    Set<Feature<?>> features;

    features = Sets.<Feature<?>>newHashSet(ExampleDerivedFeature.DERIVED_FEATURE_1);
    assertTrue(FeatureUtil.impliedFeatures(features).isEmpty());

        (TestMultisetGenerator<E>) parentBuilder.getSubjectGenerator();

    Set<Feature<?>> features = new HashSet<>();
    features.add(NoRecurse.SUBMULTISET);
    features.add(RESTRICTS_ELEMENTS);
    features.addAll(parentBuilder.getFeatures());

    if (!features.remove(SERIALIZABLE_INCLUDING_VIEWS)) {
      features.remove(SERIALIZABLE);
    }

      new HashMap<>();

  /**
   * Given a set of features, add to it all the features directly or indirectly implied by any of
   * them, and return it.
   *
   * @param features the set of features to expand
   * @return the same set of features, expanded with all implied features
   */
  @CanIgnoreReturnValue
  public static Set<Feature<?>> addImpliedFeatures(Set<Feature<?>> features) {
    Queue<Feature<?>> queue = new ArrayDeque<>(features);

//
// Softmax cross entropy loss is defined as follows:
//
//  loss = Sum(-labels * Log(Exp(features) / Sum(Exp(features)))
//  loss = Sum(-labels * LogSoftmax(features))
//
// Computing gradient of the loss with respect to features gives us,
//
//  backprop = (Exp(features) / Sum(Exp(features))) - labels
//  backprop = Softmax(features) - labels
//

/// Converts a TF::SoftsignGradOp to HLO.
/// SoftsignGrad(gradient, features) = gradient / ((1 + abs(features)) ^ 2)
def : Pattern<
        (TF_SoftsignGradOp AnyRankedTensor:$gradients, AnyRankedTensor:$features),
        [(CHLO_BroadcastAddOp:$add
          (MHLO_ConstantOp:$one (GetScalarOfType<1> $features)), (MHLO_AbsOp $features),
          (BinBroadcastDimensions $one, $features)
         ),
         (CHLO_BroadcastDivOp

      new HashMap<>();

  /**
   * Given a set of features, add to it all the features directly or indirectly implied by any of
   * them, and return it.
   *
   * @param features the set of features to expand
   * @return the same set of features, expanded with all implied features
   */
  @CanIgnoreReturnValue
  public static Set<Feature<?>> addImpliedFeatures(Set<Feature<?>> features) {
    Queue<Feature<?>> queue = new ArrayDeque<>(features);

--

[[sec::publishing_feature_variants]]
== Publishing features

--
Depending on the metadata file format, publishing features may be lossy:

- using {metadata-file-spec}[Gradle Module Metadata], everything is published and consumers will get the full benefit of features
- using POM metadata (Maven), features are published as **optional dependencies** and artifacts of features are published with different _classifiers_

    return tearDown;
  }

  // Features

  private final Set<Feature<?>> features = new LinkedHashSet<>();

  /**
   * Configures this builder to produce tests appropriate for the given features. This method may be
   * called more than once to add features in multiple groups.
   */
  @CanIgnoreReturnValue
  public B withFeatures(Feature<?>... features) {
    return withFeatures(Arrays.asList(features));