IIUC, the variant library is similar to a "dependent product" which can be thought of as a pseudo_pair<Index,TypeList>. where an instance: pseudo_pair<I,T> dpIT; is a pair of values (i,t) where: I i; T<i> t; IOW, there is only one value, but it can be one of many types depending on the valid values of I. BTW, would variant be made for readable by making variant take an enumeration: enum I{i0, i1, ...., in-1} as the discriminator? Anyway, back to the original question. Is there any interest in a "dependent function" where: dependent_function<I,T> df; where: T<i0>& df<i0>(); T<i1>& df<i1>(); ... T<in-1>& df<in-1>(): IOW, this would be like a regular struct (a record type), however, instead of: df.ix there would be: df<ix>(); BTW, this idea comes from Nuprl: http://www.cs.cornell.edu/Info/Projects/NuPrl/nuprl.html and I've recently had a need for it because I've got a template which is specialized on an enumeration; hence, this would come in handy if I wanted to have a structure containing all instances of the specializations.