The most common use cases of generics are collections such as List and Map as well as single-element containers like ThreadLocal and AtomicReference. In both the collection case as well as the single-element objects there is a very finite list of types taken. This serves us well in many cases however sometimes we need additional flexibility. We may have a use case like mapping a row from a database in a type safe manner. In these cases we have to use another technique. This technique is to parameterize the key to the data and not the container entirely. That is, rather than specifying the type of the container we specify the type of a particular piece of data. By doing this we can enable additional flexibility while still preserving type safety.

Let's consider an example. Consider a Favorite class. The purpose of this class is to collect your favorite object of an arbitrary amount of types. The Class object will serve as the key into our Favorite class. This works because the Class type is generic and holds it's own type Class<T>. Thus String.class is Class<String>, Integer.class is Class<Integer>, and so on.

The API for our Favorite class is straightforward:

public class Favorites {
  public <T> void putFavorite(Class<T> type, T value);
  public <T> T getFavorite(Class<T> type);
}

and its use would look something like:

Favorites favorites = new Favorites();
favorites.putFavorite(String.class, "Hello world");
favorites.putFavorite(Integer.class, 123);
favorites.putFavorite(Class.class, Favorite.class); 

and finally let's check out the implementation:

public class Favorites {
  private Map<Class<?>, Object> favorites = new HashMap<>();
  public <T> void putFavorite(Class<T> type, T value) {
    favorites.put(Objects.requireNonNull(type), value);
  }
  public <T> T getFavorite(Class<T> type) {
     return type.cast(favorites.get(type));
  }
}

For what this class does this is not too complicated. What the Favorite class ends up representing is a type-safe as well as has a different type for each element, thus heterogeneous. This is how we get to a heterogeneous container description from the chapter title. Looking at the above implementation probably the most interesting part is the getFavorite function. Because we lose the type information once we put the value in the Map we need a way to retrieve the value in a type safe manner. This is where the Class.cast function comes in handy. If the object can't be cast to the type we will get a ClassCastException. Because we control the types that go into the Map we know this won't happen in the usual case.

What limitations does our Favorites implementation have? The first one is if a user of our class creates a raw class object they can corrupt the type safety of our class. This is no different than having a HashSet<Integer> and putting a String by accessing it as a raw HashSet. We actually can enforce runtime type safety if we are willing to pay for it. By changing the putFavorite function to the following we can enforce the type safety at runtime even if accessed via raw types.

public <T> void putFavorite(Class<T> type, T value) {
  favorites.put(Objects.requireNonNull(type), type.cast(value));
}

There are collection wrappers in java.util.Collections that do this same thing. checkedList, checkedSet, etc perform the same runtime check and can be useful when tracking down where type unsafe changes are made in code that works with parameterized as well as raw collections.

The second limitation is that our Favorites class can't be used with non-reifiable types. This means that we can store our favorite String, Integer, String[] but we cannot store our favorite List<String> and other non-reifiable types. This is because at runtime we lose the parameterized part of our type and are just left with List.class in the above cause and thus there would be no way of knowing the difference between List<String> and List<Integer> at runtime.

While often we can get by with parameterized collections and single element types sometimes we want the extra flexibility of typesafe heterogeneous containers. This is made possible via the Class type that allows type safety in a more dynamic manner.