[boost] Re: Re: Re: shared_ptr with 0 use_count

From: dynalink (dynalink_at_[hidden])
Date: 2003-02-10 14:39:38

Thanks again, Peter. That's about what I expected - a mutex around A::create
should do the trick.

Peter Krnjevic
Dynalink Technologies

"Peter Dimov" <pdimov_at_[hidden]> wrote in message
news:000d01c2d138$2d29e7d0$1d00a8c0_at_pdimov2...
> dynalink wrote:
> > Works great! How thread safe is this?
>
> Looks as thread safe as your version, where you could end up calling
> m_X.reset(new X) simultaneosly if two threads created a B at the same
time.
> IOW, not much. :-)
>
> All shared_ptr/weak_ptr operations have the same thread safety level, i.e.
> multiple reads are safe, writes need to be exclusive.
>
> You'll probably need a mutex lock in A::create to make B thread-neutral.
>
> > "Peter Dimov" <pdimov_at_[hidden]> wrote in message
> > news:001301c2d12c$30c1d210$1d00a8c0_at_pdimov2...
> >> dynalink wrote:
> >>> I need to create a single instance of  X and have it disappear when
> >>> the
> >>> final B pointer goes out of scope. The problem with this code is
> >>> that A::m_X immediately bumps the ref count and only goes out of
> >>> scope on termination.
> >>>
> >>> Here's a simple example:
> >>>
> >>> class X;    // defined elsewhere; in my case wraps a classic C
> >>> structure. typedef boost::smart_ptr<X> X_ptr;
> >>>
> >>> struct A {
> >>>     static X_ptr create() { if (!m_X.get()) m_X.reset(new X); return
> >>>     m_X; } static X_ptr m_X;
> >>> };
> >>>
> >>> struct B {
> >>>     B() : m_A(A::create()){}
> >>>     X_ptr m_A;
> >>> };
> >>
> >> You need to keep a weak_ptr in A.
> >>
> >> struct A
> >> {
> >>     static X_ptr create()
> >>     {
> >>         if(X_ptr px = make_shared(wp_))
> >>         {
> >>             return px;
> >>         }
> >>         else
> >>         {
> >>             X_ptr px(new X);
> >>             wp_ = px;
> >>             return px;
> >>         }
> >>     }
> >>
> >>     static weak_ptr<X> wp_;
> >> };
>
> _______________________________________________
> Unsubscribe & other changes:
http://lists.boost.org/mailman/listinfo.cgi/boost
>