I read the page: http://www.digitalmars.com/d/2.0/const3.html under section "Const Member Functions"

It didn't give any example, but following the example of 'invariant', it should be:

const ReturnType memberFunc(param) {}

I think this is really confusing: is 'const' trying to specify the 'ReturnType' or the memberFunc?

And if I want to specify both, then I have to write:

const const ReturnType  memberFunc(param) {}, or
const const(ReturnType) memberFunc(param) {}

The 2 leading const in a row looks ugly and confusing.

Why not just use the C++'s postfix syntax for const member function? to put 'const' after the param, and before the body {}

ReturnType memberFunc(param) const {}

So I tried it (although undocumented); and to my glad, it worked (see code below).

But there seems to have a semantic bug: I'm modifying class field in void f(A o) const {...}, but the compiler didn't complain about it.

I think we should fix the confusing syntax, and use C++'s postfix syntax for const member function; and also fix this bug.


$ cat const.d
===============================
class A{
public:
  A m_a;
}

class B{
public:
  A m_a;
  A a() const {
    return m_a;
  }
  void f(A o) const {  // without 'const' won't compile
    m_a = o;           // BUG, I'm modifying m_a
  }
/*
  const A v() {
    m_a = null;
    return m_a;
  }
*/
}

int main() {
  const B b = new B();

  A a = b.a();
  a.m_a = null;
  b.f(a);
//b.v();

  return 0;
}
===============================
$ dmd.exe const.d
d:\project\dmd\bin\..\..\dm\bin\link.exe const,,,user32+kernel32/noi;

On 15/02/2008, im <no@body.com> wrote:
>  I think this is really confusing: is 'const' trying to specify the
>  'ReturnType' or the memberFunc?
>
>  And if I want to specify both, then I have to write:
>
>  const const ReturnType  memberFunc(param) {}, or
>  const const(ReturnType) memberFunc(param) {}

Actually, only

    const const(ReturnType) memberFunc(param) {}

would work.

This has been mentioned before, and for what it's worth, I agree with you. Just some comments though. Walter likes the existing syntax, because it means you can do:

    const
    {
        /* several functions */
    }

It is unusual ever to need to return a const(ReturnType), since why would you want to prevent the callee from modifying their copy of the return value? That said, it is very common to want to return a const(T)[] for some T, so the problem doesn't go away.

const-at-the-end syntax does work. I don't know if this is a permanent feature or not, but currently you can say:

    const(ReturnType) memberFunc(param) const {}

And finally, I have argued in the past (and still believe) that the D
syntax should be

    const(this) const(ReturnType) memberFunc(param) {}

For an extra six characters, you get complete clarity. const(this) would mean that the symbol "this" will (transitively) not be modified within its scope. Walter's desire to encompass multiple functions would thus be preserved. i.e.

    const(this)
    {
        /* several functions */
    }

and in addition, the new syntax lends itself to future expansion (such as const(outer), etc.). Unfortunately, last I heard, Walter wasn't too keen on this suggestion, possibly because it means typing six more characters.

But either way, I definitely believe that

    const ReturnType memberFunc(param) {}

/must/ have exactly one interpretation, and that interpretation should be

    const(ReturnType) memberFunc(param) {}

and not, as is currently the case

    ReturnType memberFunc(param) const {}

I agree that that /is/ confusing, and is one of those little creases in the new const system that we still need to iron out.

Janice Caron wrote:
> It is unusual ever to need to return a const(ReturnType), since why
> would you want to prevent the callee from modifying their copy of the
> return value?

How about returning a class reference that you still have your own copy to?
One common example of this might be getters for properties of class type, especially if the class instance is part of the internal state of the owning class and shouldn't be modified without going through the methods of the owner.

On 16/02/2008, Frits van Bommel <fvbommel@remwovexcapss.nl> wrote:
> Janice Caron wrote:
>  > It is unusual ever to need to return a const(ReturnType), since why
>  > would you want to prevent the callee from modifying their copy of the
>  > return value?
>
>
> How about returning a class reference that you still have your own copy to?

Good point. I'd forgotten about that one.

Not that that makes any difference to im's original point, or to my reply, of course. The unintuitive meaning of

    const ReturnType f(params...)

still remains.