problem with NTL 5.5 when upgrading from GMP 4.2.4 to 4.3.0

[shoup at cims.nyu.edu]

Paul,

I think I was a bit pessimistic in my assessment.
I believe that with a bit of care, one can efficiently
post-process the output of any XGCD routine to
get outputs that are consistent with running
the classical extended Euclid routine.

I'll pursue this as as soon as I can.

  -- Victor

>
>>        Tobrjörn,
>>
>>> Should we slow down the code a little and get a more strictly bounded
>>> cofactor value for the price?  The price is small, but if the operands
>>> are just a few limbs, it might still hurt.
>>
>> (a) having loose bounds will help to detect bugs in applications, as
>> this
>>     example shows
>>
>> (b) having strict bounds will help applications, indeed some of them
>> might
>>     have to do some extra work to get unique cofactors, or the
>> correctness
>>     of some algorithm might depend on strict bounds
>>
>> I vote for (b) if the extra price is small.
>>
>> Paul
>>
>>
>
>
>