Invoking Java Generic Methods

generic-method, generics, java, type-erasure

Solution

The answer seems to go beyond @Telthien and @newacct' answers. I was curious to "see" for myself the difference between:

System.out.println(Util.<String>compare("a", "b"));

with explicity typing, and:

System.out.println(Util.compare(new String(""), new Long(1)));

with implicit typing.

I performed several experiments, using variations on these two previous lines. These experiments show that, short of using the anonymous/local class trick, the compiler does check the types during compilation but the generated bytecodes only refer to `Object`, even in the case of the first line.

The following piece of code shows that typecasts can be performed safely all the way to `Object` even in the case of the explicity type argument `<String>`.

public final class Example44 {
    public static void main(final String[] args) {
        System.out.println(new Util44<String>().compare("a", "b"));
        System.out.println(new Util44().compare(new String(""), new Long(1)));
    }
}

final class Util44<T> {
    private T aT;
    public boolean compare(T t1, T t2) {
        System.out.println(this.aT);
        // I was expecting the second and third assignments to fail
        // with the first invocation because T is explicitly a String
        // and then to work with the second invocation because I use
        // a raw type and the compiler must infer a common type for T.
        // Actually, all these assignments succeed with both invocation. 
        this.aT = (T) new String("z");
        this.aT = (T) new Long(0);
        this.aT = (T) new Object();
        return t1.equals(t2);
    }
}

The bytecodes of the `main` method look like:

  // Method descriptor #15 ([Ljava/lang/String;)V
  // Stack: 7, Locals: 1
  public static void main(java.lang.String[] args);
     0  getstatic java.lang.System.out : java.io.PrintStream [16]
     3  new ca.polymtl.ptidej.generics.java.Util44 [22]
     6  dup
     7  invokespecial ca.polymtl.ptidej.generics.java.Util44() [24]
    10  ldc <String "a"> [25]
    12  ldc <String "b"> [27]
    14  invokevirtual ca.polymtl.ptidej.generics.java.Util44.compare(java.lang.Object, java.lang.Object) : boolean [29]
    17  invokevirtual java.io.PrintStream.println(boolean) : void [33]
    20  getstatic java.lang.System.out : java.io.PrintStream [16]
    23  new ca.polymtl.ptidej.generics.java.Util44 [22]
    26  dup
    27  invokespecial ca.polymtl.ptidej.generics.java.Util44() [24]
    30  new java.lang.String [39]
    33  dup
    34  ldc <String ""> [41]
    36  invokespecial java.lang.String(java.lang.String) [43]
    39  new java.lang.Long [46]
    42  dup
    43  lconst_1
    44  invokespecial java.lang.Long(long) [48]
    47  invokevirtual ca.polymtl.ptidej.generics.java.Util44.compare(java.lang.Object, java.lang.Object) : boolean [29]
    50  invokevirtual java.io.PrintStream.println(boolean) : void [33]
    53  return
      Line numbers:
        [pc: 0, line: 24]
        [pc: 20, line: 25]
        [pc: 53, line: 26]
      Local variable table:
        [pc: 0, pc: 54] local: args index: 0 type: java.lang.String[]

It actually makes sense that all the calls are always to methods with `Object` as formal parameter types, as explained in another question/answer. To conlude, the compiler always uses `Object` for the generated bytecodes, not matter if there is an explicity type argument (first line) or an implicit type argument but that the objects could have a common superclass different from `Object`.

Problem

Original source