Re: signals and copy constructors

[Carl Nygard <cjnygard fast net>]

On Tue, 2005-01-04 at 17:54 +0100, Jan Hudec wrote:
> On Tue, Jan 04, 2005 at 15:16:51 +0100, Antonio Coralles wrote:
> > Carl Nygard wrote:
> > 
> > >On Mon, 2005-01-03 at 23:46 +0100, Antonio Coralles wrote:
> > >>i know that this is more a libsigc++ question; but as gtkmm uses
> > >>libsigc++ heavily and the libsigc++ mailinglist is not very populated i
> > >>ask this question here too:
> > >>
> > >>is there an elegant way to "change  *p_obj in  mem_fun(*p_obj, ...)"
> > >>after the signal is allredy attached to a concrete object , so that when
> > >>the copy constructor of that object is called,  some_signal.emit(...)
> > >>calls the method in the object created by this copy constructor call ?
> > >>or is it better to think of a different design, so that this question
> > >>doesn't come up ?
> > >>
> > >>to avoid misunderstandigs, here is somekind of example:
> > >>
> > >>//just a useless signal
> > >>sigc::signal<void> someSignal;
> > >>
> > >>//just a useless class
> > >>class Useless : public sigc::trackable
> > >>{
> > >>    public:
> > >>    void call_me() { ... }
> > >>};
> > >>
> > >>//a useless scope
> > >>....
> > >>{
> > >>    Useless u1;
> > >>    someSignal.connect(sigc::mem_fun(u1,  &Useless::call_me));
> > >>   //ok now connected - but how can i automatically change the target
> > >>object to u2 - so that u2.call_me() is called due someSignal.emit() ?
> > >>   Useless u2(u1);  //imagine this call is made at a comletly different
> > >>part in my programm - for example in a clone method ....
> > >>   ....
> > >>}
> > >> ....
> > >
> > >This is a bad idea simply from the standpoint that C++ creates temporary
> > >copies in a variety of situations, many of which you are not consciously
> > >aware of.  I'd find a different design.
> > >
> > >BTW, you can also disconnect the signal if you store the connection info
> > >returned when you call connect(), so you can manually disconnect and
> > >reconnect (so long as you have the original signal).
> > >
> > Well, i guess you are right, I should find another way, allthough 
> > temporary copies aren't a problem, because i could make the copy 
> > constructor private and use it only in a virtual clone method ....; but 
> > to store the the connection and the signal in the object to wich the 
> > signal was connected seems messy and error prone.
> 
> Actualy the callback will get the signal invocant (ie. the object
> emiting the signal) as first argument no matter whether you use a method
> or function. So I suggest connecting a static function that invokes
> method on it's first argument.

Huh?  You sure you're not getting the mem_fun signature mixed up with
the sigc::signal<void>() signal signature?