Re: [Boost-users] signal slot flexibility

From: Doug Gregor (dgregor_at_[hidden])
Date: 2004-09-07 14:38:19

On Sep 7, 2004, at 1:36 PM, Frank Maddin wrote:
> If I add a function to both Hello and World, from the “Hello World” 
> example, called alternate() and set up signals like so:
>
>  
>
>             boost::signal<void (Hello*) > sig_h;
>
>             boost::signal< void (World*) > sig_w;
>
>  
>
>             sig_h.connect(boost::mem_fn(&Hello::alternate)); 
>
>              sig_w.connect(boost::mem_fn(&World::alternate));
>
>  
>
> I have to define two different signals since the object type is part 
> of the signal definition.  Is there an easy way to get a single signal 
> to call the alternate() function of both Hello and World?

Without Hello and World being related (e.g., sharing a base class 
declaring a virtual "alternate" function), I don't see a way. Still, 
when you call these signals you have to know whether you have a "Hello" 
in hand or a "World"; perhaps you wanted to store the "Hello" and 
"World" objects within the signal, and pass it no arguments? For 
instance,

boost::signal<void()> sig;

sig.connect(boost::bind(&Hello::alternate, &my_hello));
sig.connect(boost::bind(&World::alternate, &my_world));

> I recently found this signal slot solution.  
> http://www.codeproject.com/cpp/ElmueSignalsandSlots.asp
>
> There is an extra level of indirection here, in that you define a slot 
> that points to the function.  I’d be interested to see what people on 
> this forum think about this solution. If I understand things 
> correctly, this solution doesn’t seem to fit into the categories 
> listed in the “Choice of Slot Definitions” docs.

I don't see much benefit to extra (explicit) level of indirection here. 
Most slots are short bind expressions that don't ever need to be 
revisited, so it seems that it could be a bit challenging to determine 
where some slots should reside: always in the class containing the 
member function that is called (assuming one only uses member 
functions!)?

>  Also a general signal slot question.  What are the main reasons why 
> developers don’t use an inherited pure interface for every callback 
> type?  A signal would be a template class with a list of pointers to 
> the interface and some methods to operate on the list.  Not flexible 
> enough?

You often want only one event out of five, so it's a pain to drop in 
all of the stubs needed to ignore the other events. Also, you don't get 
a choice of names to use: if the writer of the interface says it's 
"on_click", you must write a function named "on_click" even though 
you'd rather write something like "submit_current_query" that describes 
the action taken when the event fires, not the event itself (which is 
essentially irrelevant).

>   Doesn’t scale well to larger uses?  Time spent calling virtual 
> functions?  I saw this in the boost signal docs that might apply - 
> “the use of a large number of small adaptor classes containing virtual 
> functions has been found to cause an unacceptable increase in the size 
> of executables (polymorphic class types require more code than 
> non-polymorphic types)”.

This is more of an implementation detail; using function pointers 
instead of virtual functions tends to reduce the amount of code 
generated for libraries such as Boost.Signals or Boost.Function.

        Doug