Re: [Boost-users] [serialization] vector deserialization problem

26 Sep 2005


      This is a bug.  Fixed in 1.33.1 Hopefully coming soon.

Robert Ramey

Sergey Filippov wrote:
...
Hi, all.
I'm using boost 1.33.0 and I have a problem deserializing from an
archive (I'm using VC 7.1). To reproduce the problem I wrote a sample
program listed below.
#include <sstream>
#include <boost/archive/binary_oarchive.hpp>
#include <boost/archive/binary_iarchive.hpp>
#include <boost/serialization/export.hpp>
#include <boost/serialization/vector.hpp>
struct A
{
   virtual ~A() {}
friend class boost::serialization::access;
   template<class Archive>
       void serialize(Archive & ar, const unsigned int version) {}
};
struct B: A
{
   friend class boost::serialization::access;
   template<class Archive>
       void serialize(Archive & ar, const unsigned int version)
   {
       ar & boost::serialization::base_object<A>(*this);
   }
};
BOOST_CLASS_EXPORT(A);
BOOST_CLASS_EXPORT(B);
int main(int argc, char* argv[])
{
   std::string str;
{
       std::stringstream stream;
       boost::archive::binary_oarchive ar(stream);
       B b;
       std::vector<int> v(10);
ar & b;
       ar & v;
str = stream.rdbuf()->str();
   }
{
       std::stringstream stream;
stream.rdbuf()->str(str);
boost::archive::binary_iarchive ar(stream);
       B b;
       std::vector<int> v;
ar & b;
       ar & v; // causes an assert
   }
return 0;
}
During deserialization of the vector (a line with "// causes an
assert"), an assert occurs in the line 266 in basic_iarchive.cpp.
I don't understand the library source code well, but it looks
something strange for me. As far as I understood the
reset_object_address function supposed to be called with addresses of
tracked objects. In my case the new_address parameter is the address
of an element in the vector. But the vector elements are of a
primitive type and must not be tracked. Maybe reset_object_address
supposed to handle such a situation, but in the provided sample
something goes wrong.
If I skip the line with reset_object_address in archive_input_seq
(collections_load_impl.hpp, line 91) in debugger, the sample works as
expected. Also with 1.32 (in which archive_input_seq almost identical
with one in 1.33 except the call to reset_object_address) the sample
works as well. So I'm considering to patch the library by commenting
this line. What side effects will it have?
I as understood from previous posts in this forum reset_object_address
was added as a bugfix to loading of containers with pointers to
trackable objects. Is it safe to comment out the reset_object_address
call, if I do not serialize such containers?
Or maybe I'm just using the serialization library in a wrong way? Then
where is the problem in my code?
Thanks in advance.
Sergey