The 'cf' (continued fraction) documentation should mention that cf expansions are exact for rational numbers, but are cut off at ratepsilon (for floats) and 10^-fpprec (for bigfloats). -s