Re: [Boost-users] Serialization and const

From: Robert Ramey (ramey_at_[hidden])
Date: 2007-02-02 11:24:51

This is explained in the "rationale" section of the documentation.

The issue is that tracking implements the behavior that only one
copy of an object is serialized even though it mght be passed
multiple times to the serialization library (e.g. as a pointer).  The
library takes the passing of a non-const to signal that the programmer
might be changing this value while its being serialized and traps
this with a compile time message.  This avoids accidently creating
archives which cannot be read in the future.

Robert Ramey

Anders Sundman wrote:
> Hi!
>
> I'm having some problems grasping how the serialization library works
> with respect to const/non-const data.
>
> Ex:
>
> Consider the following simple class that I would like to serialize:
>
> struct Simple {
> Simple() : member(0) {}
> Simple(int m) : member(m) {}
> int member;
> };
> BOOST_CLASS_EXPORT(Simple);
> BOOST_SERIALIZATION_SPLIT_FREE(Simple);
> //BOOST_CLASS_TRACKING(Simple, boost::serialization::track_never);
>
> namespace boost {
>   namespace serialization {
>     template<class Archive>
>     void save(Archive & ar, const Simple & t, unsigned int version)
>     {
>       ar << t.member;
>     }
>     template<class Archive>
>     void load(Archive & ar, Simple & t, unsigned int version) {
>       ar >> t.member;
>     }
>   }
> }
>
> But when I do the serialization, strange things happen.
>
> Ex:
>
> const Simple s1(7);
> //Simple s1(7);
> boost::archive::text_oarchive oa(ss, f);
> oa << s1;
>
> The non-const alternative above doesn't work unless I do
> BOOST_CLASS_TRACKING(Simple, boost::serialization::track_never);
>
> And,
>
> Simple * const s1 = new Simple(7);
> //Simple * s1 = new Simple(7);
> boost::archive::text_oarchive oa(ss, f);
> oa << s1;
>
> The non-const alternative above doesn't work even if I do
> BOOST_CLASS_TRACKING(Simple, boost::serialization::track_never);
>
>
> I'm obviously missing something here, please help me out.
>
> Best regards,
> Anders Sundman