Re: [Pkg-octave-devel] [RFU] octave-image 1.0.15-1

[Rafael Laboissiere <rafael@laboissiere.net>]

* Thomas Weber <tweber@debian.org> [2012-03-13 21:37]:

> On Mon, Mar 12, 2012 at 11:07:05PM +0100, Rafael Laboissiere wrote:
> > * Thomas Weber <tweber@debian.org> [2012-03-12 23:06]:
> > > >From my build log (clean cowbuilder):
> > > [fftconv2]
> > >   ***** assert(norm(fftconv2(b,a,'valid')-conv2(b,a,'valid')), 0, 1e6*eps)
> > >   !!!!! test failed
> > >   assert (norm (fftconv2 (b, a, 'valid') - conv2 (b, a, 'valid')),0,1e6 * eps) expected
> > >   0
> > >   but got
> > >   894.94
> > >   maximum absolute error 894.944 exceeds tolerance 2.22045e-10
> > > 
> > > Log file at http://people.debian.org/~tweber/logs/
> > > 
> > > If you don't see this problem, I don't mind uploading, though.
> > 
> > I reported this to upstream yesterday:
> > 
> > http://sourceforge.net/mailarchive/message.php?msg_id=28963968
>
> Your call: should I upload with the current status?

Yes, please, upload it.  From the followup to my message in the
octave-dev mailing list, it appears that the problem is in Octave and not
in the Image package.  Indeed, fftconv2 returns the correct value, while
conv2 returns the wrong one.

Rafael