draft proposal for ref counting in D

[Walter Bright]

Rainer Schuetze wrote:


 >> Why is it necessary to put a lock around the pair?

 > To be more accurate, it is the assignment and the Release that have to
 > be atomic, in addition to a concurrent read with AddRef. Imagine the
 > reading thread is suspended while just having read the pointer, but not
 > incremented the reference count yet. If an assignment with release and
 > deletion is performed before the thread resumes, AddRef is called on
 > garbage.

On my way to work today, I figured that a safe but slow implementation can work, 
if the interface is not AddRef/Release, but

class C
{
   C readThis();
   void writeThis(ref C c);
}

where the function can include the necessary locks, e.g.

class C
{
   int refcnt;

   C readThis()
   {
     synchronized(this)
     {
       refcnt++;
       return this;
     }
   }
   void writeThis(ref C c)
   {
     synchronized(c)
     {
        C x = c;
        c = this;
        if (--c.refcnt == 0)
          delete c;
     }
   }
}

Reading/Writing null (e.g. when constructing or destructing a reference) would 
have to be special cased, that would not be necesessary with free functions.