Re: [boost] [serialization] rvalues and const-correctness?

Subject: Re: [boost] [serialization] rvalues and const-correctness?
From: Robert Ramey (ramey_at_[hidden])
Date: 2008-11-06 15:39:10

David Abrahams wrote:
> Consider:
>
>  struct X
>  {
>      int f();
>
>      template <class Archive>
>      void save(Archive& ar, unsigned)
>      {
>          ar << f();  // can't bind non-const reference to temporary
>      }
>
>      BOOST_SERIALIZATION_SPLIT_MEMBER()
>  };
>
>
> This error seems silly to me.

Hmmm - looks like a good idea to me.  If f() is a non const function, then
calling it while saving data could change the state of the data being saved.
The library tracks repeated saving of objects and on subsequent saves
would only save a handle to the original one.  This saves time, space
and handles multiple pointers to a single instance.  If the process
of saving can change the data, then all this breaks down.  (not to
mention saving - without locking - from multiple threads - which is
now permited by the library.  If f() can't be made "const" and is used in 
this way,
its likely that this would create an archive which would not load
correctly.  Finding the source of this error, and correcting its effects
would seem to be a nightmare to me.

Couldn't we add this overload to the
> archives, or even simply replace the existing operator with this one?
>
>    template<class T>
>    Archive & operator<<(T const & t){
>                        // ^^^^^------------- Note
>        this->This()->save_override(t, 0);
>        return * this->This();
>    }

Aaaa - I would have to spend some time to look into that.

Robert Ramey