Re: [Boost-users] [signals] thread-safe post-disconnect() cleanup

From: Timmo Stange (ts_at_[hidden])
Date: 2007-02-23 18:26:38

Peter Dimov wrote:

> Something along the lines of:
> 
> template<class Sig> class slot
> {
> private:
> 
>     shared_ptr< slot_impl<Sig> > pi_;
> 
> public:
> 
>     template<class F> slot( F f ): pi_( new slot_impl<Sig>( f ) ) {}
>     result_type operator()( ... ) { return (*pi_)( ... ) }
>     template<class T> void track( shared_ptr<T> pt ) { pi_->track( pt ); }
> };
> 
> template<class Sig> class signal
> {
> private:
> 
>     std::vector< slot<Sig> > slots_;
>     mutable mutex mx_;
> 
> public:
> 
>     // illustration only
> 
>     void operator()( ... ) const
>     {
>         scoped_lock lock( mx_ );
>         std::vector< slot<Sig> > tmp( slots_ );
>         lock.unlock();
> 
>         for_each( tmp.begin(), tmp.end(), bind( &slot<Sig>::operator(), _1, 
> .... ) );
>     }
> };
> 
> allows your example to be restated as:
> 
> slot<void()> make_observer_slot( ... )
> {
>     return slot<void()>( some_observer( ... ) ).track( whatever );
> }
> 
> // ...
> 
> signal<void()> sig;
> sig.connect( make_observer_slot() );
> 
> with some_observer no longer needing to have a dependency on signals.

Hm, I think that still bears a higher level of dependency between
connection site and observer - this time the observer needs to know
the signal's signature. While you can use bind(...) on an ordinary
function class, it would be difficult to forward the tracking to
the true slot (I'd call the independently created one a slot model)
after binding.

Regards

Timmo Stange