Cofactor canonicalisation of mpn_gcdext

[abbott]

Hi Niels,

> This summary agrees with my analysis. For the examples I found where
> strict inequality failed, the offending number was s = 1 or t = 1;
> don't think s = -1 or t = -1 can violate the strict inequality.

It's nice to have independent confirmation :-)

Yes, you are right that s=-1 cannot happen.  As I said before strictness can fail
only when  lambda < 3, but this implies that s=0 or s=1.

Since you want to document without referring to t, could you not simply say that
  if |v/g| = 2 then s=1
  otherwise   2 |s| < |v/g|
  [note that s=0 iff input is of the form (K*N,N) for some integer K]

John.