On Mon, Nov 17, 2008 at 8:19 AM, Benoît Jacob
<jacob@xxxxxxxxxxxxxxx> wrote:
On Monday 17 November 2008 14:34:41 Gael Guennebaud wrote:
> that's ok with me. I have not planed anything special for beta2 since
> I know I won't have time to devote to Eigen for a few more weeks....
no problem at all, of course!
> Nevertheless, I'll try to fix the #include <Array> issue in the
> tutorial, and perhaps go through the doc too.
> I would like to
> emphasize that there are some rough edges in the QR module. For
> instance I'm not 100% sure about the API of the Tridiagonalization and
> Hessenberg classes. The QR class itself is also not 100% clean. For
> instance, in tridiag, there is an "in place" function which was
> supposed to overcome the issue of reusing it in other computations
> (e.g, EigenSolver), but in practice this is only done for the 3x3
> case, and since we have the same issue in reusing QR in SVD, I guess
> we should think about how to harmonize all this stuff...
Since QR and SVD are not currently being used by near-release projects (KDE
4.2, KOffice 2.0, Avogadro 1.0), we can consider leaving these modules out of
the API stability guarantee for 2.0. (We still ship them, but they're marked
experimental and not API stable). What do you think of that?
Do you need QR and SVD to get API stable for 2.0?
By the way Keir: for libmv, is it important that the QR and SVD get API stable
soon? Do you have a rough timetable?
It doesn't matter as long as the API doesn't change dramatically. We include a copy of Eigen2 in our source tree so there are no compatibility issues. Worse case we have to update our own usage to the new API. We have pretty good tests, so updating to a new API is straightforward.
Cheers,
Keir
Cheers,
Benoit