Paolo Carlini wrote: > PS: the various __is_nan overloads must be inline, otherwise lots of > multiple definitions... Then, the next issue is that the various check_nan.cc tests must be adapted to not rely on isnan being always available in std::tr1. Paolo.