bug in spec or compiler?

843793

Hello!

I have some problems with the jsr14 specification and the compiler implementation. I am not a native english speaker, but I think I can understand the specification correctly.

In the spec8.pdf document "Adding Generics to the Java Programming Language - Participant Draft Specification" on page 16 example 20: this should not compile but it does compile with jsr14 prototype 1.2 (javac 1.4.0_01)! Why? Anyone tried this?

In the same spec page 14: 6.1: "the erasure of a type variable is [...] if the bound consists of interface tyes only, the interface among the erasures of all interface types which has the least canonical name, using lexicographic ordering."
This sounds crazy, and the compiler does it different. It always uses the first parameter (following extends)! You can see this using "javap -c" on the bytecode. So, what�s right?

In the spec, page 3: "if a type parameter X has more than one bound, then it is a compiletime error to reference a member of an object whose declared type is X, unless that member is an accessible member of class java.lang.object"
Well, why is this an error? It compiles without any errors!

And another point: in the spec are examples of type variables with the "A implements I" syntax, which is not allowed. The compiler uses "extends" only. The example code is even on the same page (Page 3). Oh, oh!

Maybe I am wrong in some parts, and you can help me. (I cant believe, that I found three bugs in five hours) Any comments?

843793

There were no answers to my questions, yet. Here is some sample code to show my problems:

In the spec8.pdf document "Adding Generics to the Java
Programming Language - Participant Draft
Specification" on page 16 example 20: this should not
compile but it does compile with jsr14 prototype 1.2
(javac 1.4.0_01)! Why? Anyone tried this?

Here is the code:
class C<A> { A id (A x) {} }
interface I<A> { A id(A x); }
// should not compile, but does compile!
public class D extends C<String> implements I<Integer> {
String id(String x) {}
Integer id(Integer x) {}
}

In the same spec page 14: 6.1: "the erasure of a type
variable is [...] if the bound consists of interface
tyes only, the interface among the erasures of all
interface types which has the least canonical name,
using lexicographic ordering."
This sounds crazy, and the compiler does it different.
It always uses the first parameter (following
extends)! You can see this using "javap -c" on the
bytecode. So, what�s right?

Here is the code:
interface IA{ public void testA(); }
interface IB{ public void testB(); }
interface IC{ public void testC(); }
public class Test<A extends IB & IA & IC>{
public A value; // javap tells me, that value has type IB!
}

In the spec, page 3: "if a type parameter X has more
than one bound, then it is a compiletime error to
reference a member of an object whose declared type is
X, unless that member is an accessible member of class
java.lang.object"
Well, why is this an error? It compiles without any
errors!

Here is the code:
public class Test2<A extends IA & IB & IC>{
private A value;
public static void main(String[] args){
value.testA(); // where is the problem
value.testB(); // it works!
}
}

And another point: in the spec are examples of type
variables with the "A implements I" syntax, which is
not allowed. The compiler uses "extends" only. The
example code is even on the same page (Page 3). Oh,
oh!

Here is the code:
class ReprChange<A implements ConvertibleTo<B>,
B implements ConvertibleTo<A>> {....

Maybe I am wrong in some parts, and you can help me.
(I cant believe, that I found three bugs in five
hours) Any comments?

Again: is anyone here who can tell me what is right?

843793

I'd go to the page http://developer.java.sun.com/developer/earlyAccess/adding_generics/ and submit your observations as comments.

843793

I'd go to the page
http://developer.java.sun.com/developer/earlyAccess/add
ng_generics/ and submit your observations as comments.

thanks for the tip. I got an answer in 3 hours. Here it is:

In the spec8.pdf document "Adding Generics to the Java
Programming Language - Participant Draft
Specification" on page 16 example 20: this should not
compile but it does compile with jsr14 prototype 1.2
(javac 1.4.0_01)! Why?

this is a compiler bug

In the same spec page 14: 6.1: "the erasure of a type
variable is [...] if the bound consists of interface
tyes only, the interface among the erasures of all
interface types which has the least canonical name,
using lexicographic ordering."
This sounds crazy, and the compiler does it different.
It always uses the first parameter (following
extends)! You can see this using "javap -c" on the
bytecode. So, what�s right?

the compiler

In the spec, page 3: "if a type parameter X has more
than one bound, then it is a compiletime error to
reference a member of an object whose declared type is
X, unless that member is an accessible member of class
java.lang.object"
Well, why is this an error? It compiles without any
errors!

compiler is right

And another point: in the spec are examples of type
variables with the "A implements I" syntax, which is
not allowed. The compiler uses "extends" only. The
example code is even on the same page (Page 3). Oh,
oh!

error in spec

There will be a revised draft spec in the next few months.