Re: Mixing reproducible and non-reproducible code
On 2012-12-03 15:59:01 +0000, N.M. Maclaren wrote:
> On Dec 3 2012, John Pryce wrote:
> >Nick makes valid points about determinism being (un)related to
> >correctness, etc., but I think he is not quite fair. If a bug is
> >identified, it is much easier to locate relevant faults if running in
> >reproducible mode. And it is much easier to prove a block of code correct,
> >formally or informally, if only one thread is involved. I would value
> >optional reproducibility, as a debugging aid.
>
> Grrk. I did agree that it helps a great deal with that, but the proposal
> was for universal reproducibility, which isn't critical for that and has
> two VERY serious downsides:
>
> Demanding cross-system reproducibility is an intolerable burden for
> any non-simple function.
>
> Demanding reproducibility across versions prevents any algorithmic
> improvements.
Not necessarily, if the reproducibility is based on the behavior,
not how it is obtained. For instance, in tightest accuracy mode,
results will be reproducible at the individual function level, but
how it is computed will differ from one implementation to another
one.
Perhaps an additional requirement would be that the behavior be
sufficiently documented so that anyone could build an implementation
on his system, where the same results will be obtained.
--
Vincent Lefèvre <vincent@xxxxxxxxxx> - Web: <http://www.vinc17.net/>
100% accessible validated (X)HTML - Blog: <http://www.vinc17.net/blog/>
Work: CR INRIA - computer arithmetic / AriC project (LIP, ENS-Lyon)