Not sure whether this helps, but just from reading what BOOST_CLASS_EXPORT does I think you can not export incomplete types. If B is template, then a unique name is generated by a compiler for every specialization of B. Let's say: template<class T> struct B { ... }; B<int> is different type from B<double>. BOOST_CLASS_EXPORT will not generate export of B. You need to explicitly state which B you would like to export. This is my asumption. Good Luck, Ovanes On Mon, January 29, 2007 16:36, Marc Wäckerlin wrote:

Marc Wäckerlin wrote:

...

But it still does not step into the correct child. After some more reading, I guess, that I'd have to use BOOST_CLASS_EXPORT(B), but that does not compile, because B is a template class.

A, B, ... are in a namespace, let's say "abc". Now, I can write: BOOST_CLASS_EXPORT(::abc::B) But this must be outside of all namespaces.

Then I get: -------------------------------------------------------------------------------- /usr/include/boost/archive/detail/iserializer.hpp:197: undefined reference to `boost::archive::detail::archive_pointer_iserializerimage::BinaryReader::~archive_pointer_iserializer()' bitsets-bitsets.o: In function `~pointer_oserializer': /usr/include/boost/archive/detail/oserializer.hpp:184: undefined reference to `boost::archive::detail::archive_pointer_oserializerimage::BinaryWriter::~archive_pointer_oserializer()' --------------------------------------------------------------------------------

What's wrong, what's missing, what do I have to do?

BTW: This newsgroup is *extremely* slow. I wait for hours and hours until I see my message!

Regards Marc

_______________________________________________ Boost-users mailing list Boost-users@lists.boost.org http://lists.boost.org/mailman/listinfo.cgi/boost-users

Robert Ramey wrote:

BOOST_EXPORT must a) be invoked outside of any namespace b) not have characters like < >, :: etc in it c) refer to a complete type

Then, how to export a type that is: 1) a template class and/or 2) inside a namespace and/or 3) non-complete? Thank you Regards Marc

Marc Wäckerlin wrote:

Robert Ramey wrote:

...

BOOST_EXPORT must a) be invoked outside of any namespace b) not have characters like < >, :: etc in it c) refer to a complete type

That's not even the problem. I extended my example from the last thread by a simple class A - no template, no namespace - and it reproduces the problem.

Please have a look at the attachment.

Other question: My main problem is, that derived classes are not stored through (smart-) pointer to base.

Then BOOST_CLASS_EXPORT should not be necessary. It is only used to assign an external name to a type.

Will this problem be resolved when I'll be able to export, or what's the problem there and what's "BOOST_CLASS_EXPORT" for? Is it only for assigning a name? Then, do I really need it (because my serialiser does neither care about, nor stores meta information)?

I do not see how a serializer which does not store meta information can work. I suggest separating issues relating to serialization of classes from those related to creation of custom archives. a) test serialization of your classes one of the archives delivered with the system b) test your custom archive with some of the class serializations delivered with the system. It looks like your example fails to link. Note that the archives included with the system used explicit instantiation to generated templated code not otherwise explicitly referred to. Robert Ramey

OK. Let me re-phrase this. I believe that the only time BOOST_CLASS_EXPORT (or something like it) should be necessary is when one serializes a derived class through an abstract pointer. It has two functions. First, it ensures that code which would not be otherwise instantiated is in fact instantiated. Second, it relates an externalizable text string to the type being serialized. Robert Ramey David Abrahams wrote:

"Robert Ramey" writes:

...

BOOST_CLASS_EXPORT should not be necessary. It is only used to assign an external name to a type.

That does not correspond to my understanding.

Marc Wäckerlin wrote:

Any Idea why there's a link problem?

I received no answer until now, so I'd like to ask a bit differently: Are you able to reproduce the link-problem? Or have you any idea, why it occurs? According to the code, the missing method is inline - so why doesn't it link? It should be reproducible with the example I sent you, an the GNU-gcc. Thank you Regards Marc