Re: [Boost-users] [Serialization]Can not export polymorphic types

Subject: Re: [Boost-users] [Serialization]Can not export polymorphic types
From: Robert Ramey (ramey_at_[hidden])
Date: 2008-12-30 13:50:47

  "Daniel Roberts" <impulsionaudio_at_[hidden]> wrote in message news:14125aaf0812301023u33f8da07qd1f60c7724c707aa_at_mail.gmail.com...
  That works. Some things I notice:

  1) The xml file created has an attribute in the object tag called "class_name", containing the name of the derived type. This attribute does not exist in an xml file created using derived type pointers. I guess this is the critical piece of information needed to reconstruct a derived type using a base class pointer.

  *** If an object is written through a derived type pointer, there is really no need for the "class_name" - it's redundant.  Note that this also means that the "BOOST_CLASS_EXPORT" isn't necessary either.  In general, archives hold that information - and ONLY that information required to (re)load the object.

  2) It also works if you create a derived class pointer/object, static_cast the derived type pointer to a base class pointer and pass the base class pointer to the serialization.

  *** correct, casting overrides the C++ typing system - that's the purpose of the cast.  In general, you can serialized the derived type or the base type (which will retrieve the derived type information.  What you CANNOT do is to save as one type and load as different type.

  3) someInXMLArchive >> boost::serialization::make_nvp("", someTypePtr); works, which is surprising since I am passing an empty string for a tag. It appears serializing in ignores tag names and just processes data until the class is constructed or a stream error occurs.

  ***
  The only function of the tag is to permit a name for the html tag.  It has no purpose as far as serialization itself is concerned.  As a useful aid in debugging, the starting/ending tags are matched be sure someone has munched up the xml file between the time the archive has been saved and loaded

  4) Calling close on the output file stream (e.g. std::ofstream::close) causes the closing boost_serialization tag, </boost_serialization> to NOT be written to the file. Is this an issue that needs to be resolved or am I doing something else wrong?

  ***
  The archive is "woundup" when it's destructor is called.  If the underlying stream is closed before that happens, you won't get the last tag.  What I recommend is to do something like:
  {
      std::fostream os("filename")
      {
          boost::archive::xml_archive xa(os);
          xa << ....
          // xa destructor finishs off the archive
      }
      // os.close called by os destructor
  }

  which guarentees that the archive destructor is called before the stream is closed.

  Thanks a bunch for all the help. Hopefully I'll be running on my own now. Please feel free to comment on my observations above, however.

  ***
  You're welcome.  If you have a guilty conscience about consuming so much time, feel free to submit a "trouble shooting tips and techniques" to the serialzation documentation.  Your submission can take the form of a TRAK item.
  Robert Ramey

  Regards,
  Daniel Roberts