Re: Same name parameter and data member initialization

James Kanze wrote:


Note, the code is unsafe for exceptions (in generic case). With the
kind of pointers you need ideally two-stage sequence of initializing:

    A::A(int id, double pay, char* name ): // auto: //:)
     id(id),
     pay(pay),

     //first stage
     //here zero for delete[] safe
     name(0)
{
  //second stage
  A::name=new char[ strlen(name) + 1 ];
}

but since C++ does not support control of auto/noauto destructor
calls, you can do like this:

    //dtor replacement
    A::do_delete(){ delete[] name; }

    //dtor
    A::~A(){ do_delete(); }

    //will call dtor
    A& A::operator= (const A&);

    //will call dtor replacement
    A::A(int id, double pay, char* name ):
     id(id),
     pay(pay),
     //first stage
     //here zero for delete[] safe
     name(0)
{
  try{
  //second stage
  A::name=new char[ strlen(name) + 1 ];
  ...
  }
  catch(...){ do_delete(); throw; }
}

Appropriate RAII wrapper instead of POD pointers is safe for
exceptions for current-style code.

POD pointers as member of class can be found, for example, while
porting old-style code without exceptions into old-style code with.

You probably can use (or write) a wrapper to replace POD pointer
behaviour for the kind of code instead of two-stage initializing (to
do porting old-style code without exceptions into new-style code
with).

Once I have refused from any of the porting because of not-stacked C++
exceptions:

class A
{
Internal a,b;

 ~A()
 {
  //can throw during throw
  Closer closer(::external,a,b);
  closer.do();
 }
};

The possible runtime error here is very hard to detect at compile
time, because "Closer" is general-purpose class - is not developed to
be used for destructors only and whole code has not been written for
"no any destructors for non-wrapper class" paradigm :)

Maksim A. Polyanin
http://grizlyk1.narod.ru/cpp_new