Dear Users and Developers, I am currently working on a kind of discriminated union type, e.g. XDR specific union type implementation (please refer to http://www.faqs.org/rfcs/rfc1832.html section) which maps enumeration values to types. In this case one wants to provide users as much comfort as possible for accesing this union. I have written a template class, which is defined by mpl::vector pairs of discriminant and related type. This class contains a boost::variant definition from this sequence. The problem is, that accessing the values etc. of this variant is a little bit clumsy: discriminated_union_inst.get<some const int for discrminant>(); discriminated_union_inst.set<some const int for discrminant>(new value); So I defined some accessor template. If initialized the accessor is responsible to access only one particular discriminant related type for example if discriminated_union is defined by sequence of: mpl_derived_pair<0, int>, mpl_derived_pair<1, std::string> ... mpl_derived_pair<N, MyClass> I define an accessor object: accessor<0> accessor_for_0(union_instance); later on this accessor allows the access if valid data is contained in the union. if union contains second type (mpl_derived_pair<1, string>) accessor_for_0->some_value //throws exception or if(accessor_for_0) // is true: means unions contains the correct type //do some access if true Now I must say: It was a pleasure to implement this using many boost libraries! But as some of you might have seen I have to pass to all accessors the reference or pointer to the accessed discrminated union type. So every instance requires on 32bit OS at least 4 bytes. The banal thing is that a group of accessors accesses the same type. The question/idea here would be to use or to make these objects share the reference, so that all related accessor references require 4 bytes. What do you think of this? Is it possible at all? With Kind Regards, Ovanes Markarian