+ // Oh dear, we have a problem as Newton and Halley steps
+ // disagree about which way we should move. Probably
+ // there is cancelation error in the calculation of the
+ // Halley step, or else the derivatives are so small
+ // that their values are basically trash. We will move
+ // in the direction indicated by a Newton step, but
+ // by no more than twice the current guess value, otherwise
+ // we can jump way out of bounds if we're not careful.
+ // See https://svn.boost.org/trac/boost/ticket/8314.