Re: [Boost-users] [signals2][review] The review of the signals2 library (formerly thread_safe_signals) begins today, Nov 1st

Subject: Re: [Boost-users] [signals2][review] The review of the signals2 library (formerly thread_safe_signals) begins today, Nov 1st
From: Nat Goodspeed (nat_at_[hidden])
Date: 2009-02-17 17:36:52

Frank Mori Hess wrote:

>> If my listen() method were able to tease apart the object returned from
>> bind(), it could detect the case of a shared_ptr<boost::trackable
>> subclass>.
> 
> It could, by using boost::visit_each.  boost::bind supports visit_each to let 
> you apply a visitor to all the objects bound inside the returned functor.  
> The signals libraries use visit_each to discover trackable objects that have 
> been bound into slots.
> 
> I would spend some time learning how to use visit_each with bind.  You could 
> implement a wrapper function that inspects slots for your own Trackable base 
> class objects, and then takes whatever step is needed to extract a shared_ptr 
> and pass it to slot::track before connecting.

Thanks for the suggestion! I'm excited about this approach, but I've run 
into a snag.

As you note above, in addition to the case of my own Trackable base 
class, I want my visitor to detect a bound boost::shared_ptr<anything> 
and pass the shared_ptr to my new slot_type object's track() method. In 
fact, for a boost::shared_ptr<SomeTrackableSubclass>, I want the 
shared_ptr tracking to take precedence. As you said, this is the safer 
mechanism.

I can in fact detect a bound shared_ptr and pass it to track() as I 
want. The problem is that binding a shared_ptr captures a copy, so the 
referenced object will live until the connection is explicitly 
disconnected! That makes the slot_type::track() mechanism moot.

It looks as though I could only achieve what I want if my visit_each() 
visitor could *modify* the boost::bind object to replace the bound 
shared_ptr with its wrapped plain pointer. I don't believe this is possible?

(Aside: I tried extending my visitor's detection to boost::weak_ptr, 
changing my test code by explicitly converting my shared_ptr to weak_ptr 
before passing it to boost::bind(). I couldn't instantiate 
slot_type(boost::bind(...boost::weak_ptr<anything>...)) until I 
specialized boost::get_pointer() myself. (I'm using Boost 1.34.1 
augmented with the Signals2 library.) In any case, though, requiring 
coders to remember to explicitly wrap any boost::smart_ptr instance in 
boost::weak_ptr feels like the wrong solution.)

Perhaps I could build a sufficiently smart boost::visit_each() visitor 
to reconstruct the original boost::bind() expression, but substituting a 
plain pointer for any smart_ptr? I'm not confident I can do that without 
relying on too many boost::bind implementation details.

I'm hoping one of you will suggest a better approach.  :-)

> Just a note: you need to be careful about connecting these transient objects, 
> since the signal can't employ shared_ptr to prevent them from destructing in 
> mid-signal invocation.

Thanks for the warning. I'd prefer to handle bound shared_ptrs, if I can 
do so without making the referenced objects immortal.