m_class··on Error Propagation in Scientific ComputingIs there any generic way to preserve loss of numerical accuracy for any given function?