DE wrote:
e.g. a symmetric matrix can have such an interface:
template<...> //parameters omitted for simplicity class symmetric_matrix : public symmetric_matrix_expression<...> { //... explicit symmetric_matrix(size_t size); //since it is square //... //there's no need for void resize(size_t size); //distinct rows/columns //etc... //numbers }
so a user will not even be able to write such nonsensical code like
symmetric1.resize(10, 10); //note unnecessary duplication
First, concrete inheritance like this == bad performances. Use CRTP I'm against the proliferation of such types cause it'll just pollute and clutter the namespace when a simple policy things is easy to do and have same set of features. And anyway, your exampel is faulty, you should ask your user to wait for any type T that is a matrix expression with symmetric features or when you'll call f( a+b*c ); you'll get spurious copy. so the actual generic way to write this should be : template<class T> typename enable_if_c< as_matrix<T>::value && is_same< settings<T,shape>, symmetric>::value >::type f( T const& m ); but we can't have user write that so a small macro thingy will be needed.
because symmetric_matix does not even provide such interface and will not leave user wandering if he missed the key and typed '(10, 19)' with unreadable compiler error somewhere in instantiated templates as far as i know policies can not do this unless it is full specialized and the latter i think is equivalent to a distinct type
You should read and learn about : SFINAE ,boost::enable_if and BOOST_MPL_ASSERT. That's ages that cryptic error messages int empalte are no more. -- ___________________________________________ Joel Falcou - Assistant Professor PARALL Team - LRI - Universite Paris Sud XI Tel : (+33)1 69 15 66 35