Re: [Boost-users] boost serialization and RTTI

From: Os (os_at_[hidden])
Date: 2007-04-20 03:45:52

Thanks Robert.

I may persist with trying to get this to work, but I don't honestly
believe it will work out.

I think it would be good to include in the tests folder a single
example that would trivially serialise and deserialise data that built
with RTTI turned off in the compiler. Otherwise this option isn't
being tested at all.

cheers,
os.

On 20/04/07, Robert Ramey <ramey_at_[hidden]> wrote:
> There is a test program which demostates that this facility fuctions:
>
> It's named "test_no_rtti" or something like that.
>
> I make the "no rtti" version as sort of an after thought.  My idea
> was the the type identification method be an attribute of the
> class as other serialization attributes were.  Thus one could
> declare everything regarding how an type is to be serialized
> in the *.h file and include it in all your programs and archives
> containing instances of the given type would be transportable.
>
> So I was concerned that a program support varying methods.
> That is, I envisioned the usage of "on rtti" as a class attribute
> rather than a program attribute.
>
> So i wrote test_no_rtti to demonstrate and test this facility.
> As an example, I made a special version of extended_type_info
> (extended_type_info_nortti) as a test.  It uses the exported
> class name as an identifier so it requires that the class be
> exported by name.
>
> So my test did/does demostrate that the serialization library
> can be used without rtti.  However.
>
> test_no_rtti - fails without RTTI enabled.  This is because the
> test is mis-named.  It should be called test_no_rtti_rtti_interoperability.
>
> In order to not use the extened_type_info_no_rtti example
> as a basis for your program, you would have to make sure
> that all your types (an all imported types) use it.  The library
> address this by including the default "extended_type_info.hpp"
> if not other extended_type_info system has been specfied.  This
> facility has been the source of some infamous header ordering
> issues which have since been addressed.
>
> So it has been addressed.  It hasn't been documented well
> nor has it been tested exhaustively.  If one had nothing else to
> do he could tweak the serialization test so it would run ALL
> the tests with no_rtti.  But no one has need it badly enough
> to to this work.
>
> I don't knowwhat else to tell you.
>
> Robert Ramey
>
>
> > Hi,
> >
> > I very much want to use boost serialization but can't enable RTTI in
> > my compiler (vc7/vc8) because of compatibility restrictions with other
> > code.
> >
> > I've seen various statements on the web that boost serialization will
> > work without RTTI, but no concrete instructions about how to make this
> > happen.
> >
> > I've tried writing a trivial app using
> > boost::archive::binary_oarchive, but I get this compiler warning
> > (which is as good as an error):
> >
> > boost\smart_cast.hpp(76) : warning C4541: 'dynamic_cast' used on
> > polymorphic type 'boost::archive::detail::basic_oarchive' with /GR-;
> > unpredictable behavior may result
> >
> > So is it the case that, while serialization doesn't require RTTI, the
> > provided archive classes do? Am I going to have to write my own
> > archive class?
> >
> > Any help appreciated.
> >
> >
> > cheers,
> > os.
>
>
>
> _______________________________________________
> Boost-users mailing list
> Boost-users_at_[hidden]
> http://lists.boost.org/mailman/listinfo.cgi/boost-users
>

-- 
os_at_[hidden]
http://www.collective.co.uk/
http://www.myspace.com/darkroomtheband