Re: [Boost-users] shared_ptr in a boost::any [config] [hpux]

From: Thomas Matelich (matelich_at_[hidden])
Date: 2006-06-28 17:38:17

On 6/27/06, David Abrahams <dave_at_[hidden]> wrote:
> "Thomas Matelich" <matelich_at_[hidden]> writes:
>
> > I use boost::any for sending data to my thread's command queue, and I
> > need to send a shared_ptr in.  I'm looking for any one to tell me this
> > is fine or bad.  Based on my understanding of boost::any I think it
> > will be fine, but I'm slightly concerned I'll screw up the reference
> > count somehow.
> >
> > Can anyone confirm or allay my fears?
>
> Allay :)

Thanks.  Unfortunately, I just ran into
boost/config/platforms/hpux.hpp and its belief that gcc and threads
don't mix on HP-UX.  I'm not sure why this is stated.  Perhaps its
left over from HP 10.x and DCE threads?  I've been using gcc and
threads for a long time, even back on 10.20 using ACE.  I currently
use both ACE thread wrappers and Qt's threads wrappers on 11i with gcc
3.2.3.  I would just go ahead and comment out that assumption, but I'm
concerned someone knows something I don't know.

It looks like Jens Maurer added this assumption 2 years ago (
http://tinyurl.com/ku9wm ).  Has anyone experimented with removing it?
 I'd really like mutexes for my shared pointers.

Short-term, I'm currently working on a patch release and wouldn't
really want to mess with every shared pointer in my whole system.  I
guess my workaround would be to wrap the shared_ptr in a
mutexed_shared_ptr and lock every option that messes with the shared
count?