mpf_sub bug (essentially same as reldiff bug)

[Kevin Ryde]

abbott at module.dima.unige.it writes:
>
>   mpf_sub(rd, a, b);   // should be about 2^(-129), but gives zero

Thanks, that's a bug.

I think this case of nearly equal values is meant to work, you can see
the loop in mpf/sub.c under ediff==0 which skips equal high limbs.
Have to check what's not happening.

-- 
All followups to gmp-bugs at swox.com please.