I have a bunch of numerical routines in C++ that take and return boost::arrays, for example array<double, 5> strange_function(array<double, 3>); in another example, one integrates boost::function that takes an array of size N and returns arrays of size F The library is pretty general because it works in any number of dimensions. But it seems to be a syntactic waste to use those general functions in the cases where the input or the output is a single number, i.e. one has to use array<double, 1> all over the place, array constructions or ugly casts. One option would be to transform all the functions and boost::function declaration to take arbitrary combinations of double input/ouput and array input/out double strange_function(array<double, 3>); array<double,3> strange_function(double ); (actually strange_function is a template function, so what it really does depends on the template parameter strange_function<T>) ( Think for example of Mathematica functions that take numbers or vectors Integrate[{f[x],g[x]},x] or Integrate[f[x],x] ) I come up with a solution that is, instead of working with boost::array work with a similar type that simply believes it is a number when the dimension is one, here it is the implementation: template<size_t Dim> struct array : boost::array<double, Dim>{ typedef boost::array<double, Dim> base; array(base const& arr) : base(arr){} array(double const& d); operator double&(); operator double const&() const; }; template<> array<1>::array(double const& d) : base((base){{d}}){} template<> array<1>::operator double&(){return (*this)[0];} template<> array<1>::operator double const&() const{return (*this) [0];} in this way the array is transformed from and into a number if requested, and the existing code that uses boost::array still works. Just wondering for some feedback. Thank you, Alfredo