[Boost-users] Re: Re: Re: Serialization when using virtual inheritance

From: Robert Ramey (ramey_at_[hidden])
Date: 2005-05-30 18:12:06

basically you have to

a) "register" or "export" any concrete classes serialized.
b) serialize the base object from every derived class - including
intermedieate ones.
c) If you use export - it shoiuld be after the *_archive.hpp files.  If its
not - it should fail to compile

That's all that should be necessary.

It has been my intention to permit "export" to be only in the header.  I
don't think that 1.32 it quite worked out that way but I'm trying to fix
this.

Robert Ramey

Eric wrote:
> Hi Robert (and Jeff):
>
>> Actually, the export.hpp has to come after any *archive.hpp
>> due to the "trick" used to instantiate the appropriate code.
>
> That was the problem!  I only used the xml_oarchive object in the
> main() function, so I only included the respective headers in the
> main.  The rest of the code included the export.hpp header.  So, I
> changed my common include file to also include the *archive.hpp
> headers and everything is working well.
>

> The only thing that I've noticed is that sometimes the objects won't
> be found when I have 2 or more abstract classes between the base and
> the final concrete object.  The concrete object is registered, but it
> looks like the (Derived,Base) pair isn't being found.  Any advice on
> where I went wrong on that is appreciated.  In the meantime, I'm
> using void_cast_register<>(...) to register the base class and the
> top derived class and everything works great so far :)
>
> -Eric
>
>>
>> On some compilers, one can get multiple symbol errors.  I
>> think the only way to handle this is to be sure its
>> included at most once or tweak the linker switches to
>> override the error.
>
>> Robert Ramey
>
>
>>> "Eric" <eric-public_at_[hidden]> wrote in
>>> message news:001601c564d9$35709a60$0a00a8c0_at_FASTBRICK...
>>> Thanks Robert.  This partially worked...  Your
>>> comment about putting the BOOST_CLASS_EXPORT
>>> macro into the header keyed me into something.
>
>>> I am using the BOOST_CLASS_EXPORT(derived) macro, but
>>> I've been placing them into the source (cpp) file instead
>>> of the header (h) file.  With my system, when I put
>>> BOOST_CLASS_EXPORT(derived) into the header file, I
>>> get a multiply-defined error at link time (the offending
>>> code is boost::archive::detail::guid_initializer<class,...>
>>> since the header is included multiple times when I use the
>>> class.
>>>
>>> As a test, I setup a sandbox with a stripped down system
>>>  and was able to get it to work when I put the
>>> BOOST_CLASS_EXPORT in the header, but not when it is
>>> in the cpp file.
>>>
>>> Q:  Why does the location of the BOOST_CLASS_EXPORT
>>> matter?  Is it doing order-depended initialization
>>> upon bringing up the system?
>>>
>>>
>>> Q:  How do I avoid the mutiple defined error in
>>> the linker?
>>>
>>>
>>> Thanks again for your help and the library :)
>>>
>>> -Eric