Bill Allombert on Wed, 20 Nov 2002 19:46:40 +0100

[Date Prev] [Date Next] [Thread Prev] [Thread Next] [Date Index] [Thread Index]

Re: Bug rnfconductor

On Sat, Nov 16, 2002 at 01:49:41PM +0100, Karim BELABAS wrote:
> I have a backward compatibility problem here: the format is
>  * undocumented (it is documented in the source code)
>  * needed for a single routine: ideallog
>  * painful to use as it stands: lists of lists of lists of ..., where you
> access different layers at the same time depending on the information you
> need.
>  * lacking needed data [nfmodprinit, exponent of (O_K/P^e)^* for each P | f ]
> that I have to recompute each time I use it. When the modulus has no huge
> prime divisors, this recomputation dominates computation time in ideallog :-(.
> Currently, I basically translate it on the fly to something usable,
> recomputing missing data, _each time_ I use it (i.e each time an ideallog()
> is computed).
> So I would like to change it to something more palatable, and clean up the
> current mess (in particular remove hardwired code and define a clean
> interface). Unfortunately, this would prevent reading in old bnr / idealstar
> data. I do not want to maintain two different (large) sets of subroutines.
> This is probably not a big problem, since the only data (from the number
> fields module) worth storing is base field data, in my opinion. Not data
> related to (abelian) extensions thereof (cheap to compute, infinitely many
> extensions to handle...)
> Opinions ?

Since nfinit format has already changed and bnrinit contains the nfinit data,
it looks like the new bnrinit format is already incompatible.
Since we have already changed the documented output, why not change
the undocumented one ? And maybe document the new one in the process ?