This is an unofficial snapshot of the ISO/IEC JTC1 SC22 WG21 Core Issues List revision 114a. See http://www.open-std.org/jtc1/sc22/wg21/ for the official list.

2024-04-18

420. postfixexpression->scalar_type_dtor() inconsistent

[Voted into WP at April, 2006 meeting.]

Lets start with the proposed solution. In 12.4.6 [over.ref], replace line ...

postfix-expression -> id-expression

postfix-expression -> template_opt id-expression
postfix-expression -> pseudo-destructor-name

Alternatively remove the sentence "It implements class member access using ->" and the syntax line following.

Reasons:

Currently stdc++ is inconsistent when handling expressions of the form "postfixexpression->scalar_type_dtor()": If "postfixexpression" is a pointer to the scalar type, it is OK, but if "postfixexpression" refers to any smart pointer class (e.g. iterator or allocator::pointer) with class specific CLASS::operator->() returning pointer to the scalar type, then it is ill-formed; so while c++98 does allow CLASS::operator->() returning pointer to scalar type, c++98 prohibits any '->'-expression involving this overloaded operator function.

Not only is this behaviour inconsistent, but also when comparing the corresponding chapters of c++pl2 and stdc++98 it looks like an oversight and unintended result. Mapping between stdc++98 and c++pl2:

c++pl2.r.5.2 -> 5.2 [expr.post]
c++pl2.r.5.2.4 -> 5.2.4 [expr.pseudo] + 5.2.5 [expr.ref]
c++pl2.r.13.4 -> 13.3.1.2 [over.match.oper]
c++pl2.r.13.4.6 -> 13.5.6 [over.ref]

Additionally GCC32 does is right (but against 13.5.6 [over.ref]).

AFAICS this would not break old code except compilers like VC7x and Comeau4301.

It does not add new functionality, cause any expression class_type->scalar_type_dtor() even today can be substituted through (*class_type).scalar_type_dtor().

Without this fix, template functions like some_allocator<T>::destroy(p) must use "(*p).~T()" or "(*p).T::~T()" when calling the destructor, otherwise the simpler versions "p->~T()" or "p->T::~T()" could be used.

Sample code, compiled with GCC32, VC7[1] and Comeau4301:

struct A {};//any class

template <class T>
struct PTR
    {
    T& operator*  () const;
    T* operator-> () const;
    };

template <class T>
void f ()
    {
        {
        T*  p               ;
        p = new T           ;
        (*p).T::~T()        ;//OK
        p = new T           ;
        (*p).~T()           ;//OK
        p = new T           ;
        p->T::~T()          ;//OK
        p = new T           ;
        p->~T()             ;//OK
        }

        {
        PTR<T> p = PTR<T>() ;
        (*p).T::~T()        ;//OK
        (*p).~T()           ;//OK
        p.operator->()      ;//OK !!!
        p->T::~T()          ;//GCC32: OK; VC7x,Com4301: OK for A; ERROR w/ int
        p->~T()             ;//GCC32: OK; VC7x,Com4301: OK for A; ERROR w/ int
        }
    }

void test ()
    {
    f <A>  ();
    f <int>();
    }

Proposed resolution (April, 2005):

Change 12.4.6 [over.ref] paragraph 1 as indicated:

operator-> shall be a non-static member function taking no parameters. It implements the class member access using syntax that uses ->

postfix-expression -> template_opt id-expression
postfix-expression -> pseudo-destructor-name

An expression x->m is interpreted as (x.operator->())->m for a class object x of type T if T::operator->() exists and if the operator is selected as the best match function by the overload resolution mechanism (12.2 [over.match]).