Subj : Re: A question about atomic_ptr
To   : comp.programming.threads
From : Chris Thomasson
Date : Sun Apr 17 2005 07:05 am

> Chris Thomasson wrote:
>> // *** shared_ptr ***
>>
>>
>> 1. *Add this to public section:
>> -----------------
>>
>> void tricky_copy( shared_ptr &r )
>> {
>>  pn.tricky_copy( r.pn );
>>  px = r.px;
>> }
>
> But how do you avoid the usual problem of a thread being preempted after
> pn.tricky_copy and before px = r.px? r.px may have been changed and no
> longer in sync with the old r.pn.

Yup, just noticed that. DWCAS would work. However, I am currently fiddling 
around with logic like this:


  // public shared_ptr function
  void tricky_copy_to_local( shared_ptr &src )
  {
    // *this must be thread local

    if ( src.pn.pi_ )
    {
      T *obj;
      detail::sp_counted_base *_base;

      ac_lfgc_smr_hazard_t hazard =
        ac_lfgc_smr_get( ac_thread_self(), 0 );

      for ( ;; )
      {
        obj = src.px;

        do
        {
          // setup hazard
          _base =
            (detail::sp_counted_base*)
              ac_lfgc_smr_activate
                ( hazard,
                  (ac_cpu_node_t**)&src.pn.pi_ );

          if ( ! _base )
          {
            px = 0;
            if ( pn.pi_ ) { pn.pi_->release(); }
            pn.pi_ = 0;
            return;
          }

        // inc if > 0
        } while ( ! _base->tricky_add_ref() );

        // compare object after inc
        if ( obj == src.px )
        {
          px = obj;
          if ( pn.pi_ ) { pn.pi_->release(); }
          pn.pi_ = _base;
          ac_lfgc_smr_deactivate( hazard );
          break;
        }

        // retry
        _base->release();
      }
    }

    else
    {
      px = src.px;
      pn = src.pn;
    }
  }




I wonder if I am missing any race-conditions. The object compare after the 
increment should work... Humm...

.