"David Abrahams" <dave@boost-consulting.com> wrote
vector template will be expensive. That said, I see no reason why we need to be causing any instantiation. Everything *could* be done by partial specialization matching, though it might be verbose to account for the numbered and unnumbered forms.
Just speculating: Wouldn't implementing at<> exclusively by partial specialization matching involve defining n * n / 2 specializations? at<v1<>, 0>, at<v2<>, 0>, at<v3<>, 0>, ..., at(Vn<>, 0) at<v2<>, 1>, at<v3<>, 1>, ..., at(Vn<>, 1) at<v3<>, 2>, ..., at(Vn<>, 2) In my case, where n = 255, this would result in ~30,000 specializations of at (or at_impl). Maybe this is where the actual problem is? Regards, Arkadiy