[Pkg-octave-devel] [RFU] octave-image 2.0.0-1

[Jordi Gutiérrez Hermoso]

On 12 November 2012 08:59, Sébastien Villemot <sebastien at debian.org> wrote:
> Rafael Laboissiere <rafael at laboissiere.net> writes:
>
>> I prepared version 2.0.0-1 in experimental for upload.  I added two patches in
>> this version:
>
> Uploaded.
>
>> 1. fftconv2-usage.patch: This removes one of the bogus usage of fftconv2.  I
>> discussed this with the upstream author Carnë Draug in the octave-dev mailing
>> list.  Feel free to revert my patch if you think it is not appropriate.
>
> Since your patch fixes #664716, I kept it. I don't know how upstream
> wants to solve the issue.

Talking to Carnë in IRC, it seems clear that the problem isn't that
the test is wrong but rather that fftconv2 is wrong here. It seems
like an actual bug.

OF list: fftconv2's vector-vector-matrix three-arg call seems
incorrect and seems to have been incorrect for some time. Can anyone
please confirm? These two calls should both be equal:

    octave:1> conv2(1:3, 1:2, [1 2; 3 4; 5 6])
    ans =

        1    4    4
        5   18   16
       14   48   40
       19   62   48
       15   48   36

    octave:2> fftconv2(1:3, 1:2, [1 2; 3 4; 5 6])
    ans =

        1    4    7    6
        5   18   31   24
       11   36   61   42
       10   32   54   36


The upper result is correct. You can nobtain this result with
fftconv2(fftconv2([1 2; 3 4; 5 6,1:2),1:3), so I think this should be
fairly easy to fix.

Patches, anyone?

- Jordi G. H.