Should ClassifierCompositeItemWriter use covariance (extends "keyword" instead "super") in the output type of classifier property?

Expected Behavior
The ClassifierCompositeItemWriter should use covariance for classify the items. In the class the classifier property should be defined with an output type of ItemWriter<? extends T>. This way I can use a SubclassClassifier to classify the items.

Current Behavior

ClassifierCompositeItemWriter use contravariance for classify the items. In the class the classifier property is defined with an output type of ItemWriter<? super T>:

public class ClassifierCompositeItemWriter<T> implements ItemWriter<T> {

	private Classifier<T, ItemWriter<? super T>> classifier = new ClassifierSupport<>(null);

Context

I want to classify 2 type of items : PetToCreate extends PetAction{} and PetToUpdate extends PetAction{} and delegate to the corresponding writers: CreatePetWriter implements ItemWriter<PetToCreate> or UpdatePetWriter implements ItemWriter<PetToUpdate>.

So I use a SubclassClassifier:

var typeMap = Map.of(
          PetToCreate.class, createPetWriter,
          PetToUpdate.class, updatePetWriter
);
SubclassClassifier<PetAction, ItemWriter<? extends PetAction>> petActionItemWriterSubclassClassifier = new SubclassClassifier<>(typeMap, null);

But the type checking fails:

Can someone explain to me the motivation behind this choice, shouldn't it be possible to achieve it?

I also opened a question on StackOverflow: https://stackoverflow.com/questions/78459916/why-classifiercompositeitemwriter-use-contravariance-for-classify-the-items

Start with the ClassifierCompositeItemWriter classifier declaration and compare its generic type with SubclassClassifier and the PetAction example in the issue. Check whether the proposed covariance compiles for the two specialized writers without breaking existing ItemWriter usage; done means the intended type-checking scenario works and relevant tests cover it.