Hi, Nick,
Thanks for your reply. It's very helpful.
Thanks and Regards,
Wenliang
On Thu, Feb 9, 2012 at 12:12 AM, Nick Mathewson <ni...@freehaven.net> wrote:
> On Wed, Feb 8, 2012 at 5:37 AM, Wenliang Zhang <wla...@gmail.com> wrote:
> > Hi, all,
>
> Personally, my first approach would be to do it with something closer
> to a reference count: in the data associated with each bufferevent,
> remember the number of messages it has outstanding, and don't actually
> free the bufferevent until it has received a close _and_ it has no
> messages waiting.
>
> Alternatively, you could use a weak-reference style system where
> instead of having a pointer to the bufferevent, each Message would get
> a pointer to a a "weak reference" object that either had a pointer to
> the bufferevent, or a NULL pointer to indicate that the bufferevent
> had gone away. You'd free these objects once the bufferevent had no
> more outstanding messages.
>
> Or you could keep a linked list of messages for each bufferevent, and,
> when about to free the bufferevent, first cancel each message or mark
> it as having to bufferevent to send to.
>
> There are probably other ways too.
>
> hth,
> --
> Nick
> ***********************************************************************
> To unsubscribe, send an e-mail to majord...@freehaven.net with
> unsubscribe libevent-users in the body.
>
