> Also, I found that numerical results by 1.16 and 1.15 are different (less > than 0.1%). I doubt it is because of revision 4771. I have modified the > source code of 1.16 to undo revision 4771. And I found that, without 4771, > results by 1.16 and 1.15 are the same with each other. So, I question > whether revision 4771 do a better job or a worse job, in terms of accuracy.
It is possible that something subtle is happening (as I've not traced through the logic in detail), but I am uncertain how copying the result of a deterministic calculation is losing accuracy relative to recomputing it. - Rhys
