initialize a reference with an rvalue

My compiler does something I didn't expect. So I need to check what I
understand.

struct S { int x; };

S f() { return S(); }

void g() {
 S& r1 = S();  // rvalue
 S& r2 = f();  // rvalue
 S& r3 = *(new S());  // lvalue
 delete &r3;
}

(1) Is the initializer expression S() an rvalue?
(2) Is the initializer expression f() an rvalue?
(3) Is the initializer expression *(new S()) an lvalue?
(4) This compiles OK. But should r1 and r2 be const?

Thank you,
Marlene


      [ See  http://www.**--****.com/ ]
      [ comp.lang.c++.moderated.    First time posters: Do this! ]

Yes.


Yes.


Yes.


Yes.

The restriction that initializers of non-const references had to be
lvalues was an innovation, introduced around 15 years ago.  Some
compilers still aren't up to date concerning it, however, and a lot will
only generate a warning, rather than an error, in order to avoid
breaking existing code.

      [ See  http://www.**--****.com/ ]
      [ comp.lang.c++.moderated.    First time posters: Do this! ]

The second statement:

S& r2 = f();

is not ok. r2 will have a reference to a temporary object, that is destroyed 
when after the statement is executed. Using r2 later will crash the 
application (depending on how lucky you are :)

Catalin






      [ See  http://www.**--****.com/ ]
      [ comp.lang.c++.moderated.    First time posters: Do this! ]

An I forgot the first statement:

 S& r1 = S();

It has the same problem: refering a temporary object, that will be destroyed 
after the reference is set.

Catalin (once again)






      [ See  http://www.**--****.com/ ]
      [ comp.lang.c++.moderated.    First time posters: Do this! ]

IMHO, a const reference can refer to a temporary.

      [ See  http://www.**--****.com/ ]
      [ comp.lang.c++.moderated.    First time posters: Do this! ]