Question

As a follow-up question to What are the reasons why Map.get(Object key) is not (fully) generic, why does the JDK 6 and 7 Map interface not define the “get” method as a generic method so that the compiler can use type inference on the return value?

For example, if “get” were defined as:

public <T extends V> T get(Object key)

then a caller could write:

Map<String,Object> m = new HashMap<>();
m.put("key", new Foo());
...
Foo f = m.get("key");   // type inference, implicit cast

In the snippet above, I could have defined m as Map<String,Foo>, but note that defining m as Map<String,Object> instead of Map<String,Foo> is useful in many situations, such as when m can contain values of any type, but the value type can still be inferred based on the key e.g. a simple cache or context object.

Answer 1

Because it isn’t (in general) safe to do this. A Map<String, Object> specifically declares that its values may be any type of Object, not just Foo, so in typical usage one should not be assigning its values to anything else. Doing so would be a programmer error usually, which the type system should (and does) help you avoid. If you want to do something like that, there’s always casting, which makes your intent explicit.