On 12/4/05 11:22 AM, "David Parks" <davidp@cs.berkeley.edu> wrote:

First off, I apologize for not replying directly to my last message but I'm not even sure it's been posted yet -- as I don't have permission to post yet (the 'first message is moderated' deal).

Anyway, I wasn't able to make the example in my last post work with any sane process but I did get it to work by doing the following:

1) splitting the stuff into 3 files -- the main driver, the class header, and the class .cpp. I put both classes in the class files. 2) Including the */archive/* files in the class cpp immediately followed by export.hpp... then the class header. 3) Including the same archive headers in the class header followed by serialization.hpp (I'm sure this could be done lighter but...) 4) Including only the class header in main. 5) BOOST_CLASS_EXPORT(derived) in the class cpp. 6) This is where it starts to diverge from any information in the tutorials and documentation. First, I had to switch from text to xml. I'm not positive but, if I couldn't get it to work after 12 straight hours, I'd guess the text interface is broken. 7) You will get a pretty incorrect error (in the header comments) if you don't call the base class in the derived class serialization (ar & BOOST_SERIALIZATION_BASE_OBJECT_NVP(base);). This is not ideal because in many instances, like mine, you might not care about the bases serialization if you have a derived class... the derived class is 'smarter'. But this I can deal with.

Actually, step [7] is all you had to do. (But I'm not an expert, or even a novice, in Boost.Serialization.) You did steps 1-6 for nothing. I think another respondent, who created the library, just did step [7] in his post and got your sample code to work. You say that having to call the base implementation within your class's serialization is mostly problematic. I would say that not calling it would be problematic. Granted, you do lose the ability to precisely layout the base sub-object's state; but what happens if the base class has critical state data that it does _not_ expose? This important data could be private base class sub-objects without an accessor, private grand-base classes, or grand-base class sub-objects without an accessor. Your "smarter" derived class would have to be "too nosey" for its own good.

What I needed to do was serialize a vector of objects of some base class -- you know, like the useless circles and squares of shapes example in every intro C++ book. The documentation could use this as an example -- it has to be one of the most common uses for a serializer.

Oh, one goofy note that thankfully only took a couple of minutes to catch: namespaces aren't all the rage but if you make big use of them (like me), beware of BOOST_CLASS_EXPORT... it takes a flash of inspiration to realize that it's going to try to put colons in your XML tags. Even more tricky is the situation where you are calling, say, the base-class with BOOST_SERIALIZATION_BASE_OBJECT_NVP. It stringizes this without telling you -- it has to. But if you do this

Are you sure you need BOOST_CLASS_EXPORT? I don't think you did in your sample code. What's wrong with colons in XML tags? They're legal XML-name characters; they can even be the first character in such a name.

ar & BOOST_SERIALIZATION_BASE_OBJECT_NVP(A::B)

you will get an error but this (thankfully)

using namespace A; ar & BOOST_SERIALIZATION_BASE_OBJECT_NVP(B);

works. I have no idea how to make this friendly.

Maybe you can refer to your (direct) base classes with a typedef: class A: public daryle::B, private loki::C { typedef daryle::B base_type1; typedef loki::C base_type2; //... template < class Archive > void serialize( Archive &a, unsigned version ) { a & BOOST_SERIALIZATION_BASE_OBJECT_NVP(base_type1); a & BOOST_SERIALIZATION_BASE_OBJECT_NVP(base_type2); //... } //... }; Whatever name alias you use for a base class should be used for the simple tag name. I don't know what happens for virtual bases, though. -- Daryle Walker Mac, Internet, and Video Game Junkie darylew AT hotmail DOT com