Hi, I would like to propose a way that could avoid allocations and would be customizable too. A allocator could be aggregated inside the boost.exception hierarchy. This class could be a template on the number of bytes it would have, and it would do the SBO for allocators. struct allocator_base { virtual /* all allocator interface */ }; template <std::size_t size> struct allocator : allocator_base { boost::array<char, size> buffer_; // all boilerpart to handle multiple sizes }; Then it is possible to dynamic_cast<> to allocator_base, where all allocations would take place. That way the allocator would not be allocated in the heap and the user could customize to a bigger size if it knows it will have to attach more info to that throwing, or less if it believes to be wasting space. It could have a partial specialization for 0 too to disable SBO completely. I guess objects should be default-constructed and swapped too, to avoid copying. Regards, -- Felipe Magno de Almeida