[sigc] Re: Return values in libsigc++ 2.0

[Martin Schulze <martin-ml hippogriff de>]

Hi!

First of all, sorry for the delay: my former webhoster suddenly stopped  
all services including my email-accounts. It took >1 week for my domain  
to get acquired by my new provider ...

Timothy M. Shead wrote:
> I am evaluating what the upgrade from 1.0/1.2 to 2.0 would look like
for an existing application, and ran across one small snag. In libsigc+ 
+ 1.0/1.2 it was possible for a custom marshaller to specify the  
default value returned when a signal doesn't have any connections. In  
2.0 the behavior is to return a default-constructed object (the  
accumulator object is never even created in this case). Unfortunately  
in my case (slots that return bool) this is undesirable. While I could  
work around this by testing signal.empty() before calling signal.emit 
(), this would mean spreading some of my signal-related logic  
throughout client code, and is a bit of a step backwards. Is there an  
explicit mechanism for handling this, or should I derive from signal  
and override emit() ?

I'm afraid I wasn't aware that this was possible in 1.2 and would not
be possible with 2.0 any more. So currently, there is no explicit  
mechanism for handling this. Also, we won't want to change the default
for the case that no custom accumulator is specified for efficiency  
reasons.

A solution would be transfer some code from sigc::signalN::emit() into
sigc::internal::signal_emitN::emit() which has a specialization for the
case that no accumulator is specified. The behaviour would only change
for the case that an accumulator is specified: the accumulator's  
operator() would always be called like it was the case for marshallers  
in 1.2. I think we could consider this as a bug fix so it could go into
libsigc++-2.0.2 which could be released in a few days time. Does this
sound like a reasonable plan?

Regards,

Martin