I have a POJO specified as: MyClass, where U is the generic type parameter.
I am trying to write a utility method which accepts a class reference Class and populates a map of type Map (accepts the map to populate).
This method is implemented like:
static void populateMap(Map map, Class type) {
...
// Parses into the specified type and returns an object of that type.
T obj = parse(..., type);
map.put (key, obj);
...
return map;
}
This compiles fine. In my caller, I attempt to populate a map with any MyClass instance (irrespective of type) as the value. Hence I use the following code:
// Loses type information
Map> m = new HashMap<>();
populateMap(m, MyClass.class);
This does not compile. Compilation error:
The method populate(Map, Class) in the type ... is not applicable for the arguments (Map>, Class)
How can I fix this?
解决方案
In this case it should be safe to do an unchecked cast to Class>:
// This is okay because we're switching to a type with an unbounded wildcard -
// the behaviors of Class.newInstance and Class.cast are still safe.
@SuppressWarnings("unchecked")
Class> classWithNarrowedType =
(Class>)(Class>)MyClass.class;
populateMap(m, classWithNarrowedType);
This is a crufty solution, especially if you have many call sites like this, but there's no getting around the fact that class literals are parameterized with raw types, making their use as factories of parameterized types like MyClass inherently awkward.
A potentially cleaner solution would decouple populateMap from the use of class literals:
interface Parser {
T parse();
}
static void populateMap(Map map, Parser parser) { ... }
...
Map> m = new HashMap<>();
Parser> myClassParser = new Parser>() {
@Override
public MyClass> parse() {
return parse(..., MyClass.class);
}
};
populateMap(m, myClassParser);
As an aside I recommend a more flexible signature (see What is PECS (Producer Extends Consumer Super)? for more info):
static void populateMap(Map map, Parser parser)