That would also have been my implementation: template <int N> struct positive_power { template <typename T> static typename tools::promote_args<T>::type result(T base) { typename tools::promote_args<T>::type power = positive_power<N/2>::result(base); return (N%2) ? base * power *power : power * power; } }; with the two specializations: template <> struct positive_power<0> { template <typename T> static typename tools::promote_args<T>::type result(T base) { return 1; } }; template <> struct positive_power<1> { template <typename T> static typename tools::promote_args<T>::type result(T base) { return base; } }; (strictly speaking, <0> only is needed, but the compiler may not know to optimize multiplying by T(1), which is one criticism of previous post by John Moeller, who proposes essentially the same solution but with a factor ((N%2) ? base : 1) that I don't like). Johan: what's wrong with it? -- Hervé Brönnimann hervebronnimann@mac.com On May 14, 2008, at 1:49 PM, Johan Råde wrote:
John Maddock wrote:
Steven Watanabe wrote:
return (N%2) ? base*positive_power<N-1>::result(base)
Shouldn't positive_power<N>(base) be reduced to square(positive_power<N/2>(base) or base * square(positive_power<N/2>(base)) depending on N%2?
--Johan
_______________________________________________ Unsubscribe & other changes: http://lists.boost.org/mailman/ listinfo.cgi/boost