[IPOL discuss] demo base class and git submodules

[Juan Cardelino]

On Mon, Jan 23, 2012 at 6:23 PM, Nicolas Limare
<nicolas.limare at cmla.ens-cachan.fr> wrote:
> Hi,
>
>> The previous email from Nicolas raised a forgotten issue for me: the
>> size of the code needed to build your own demo. [...] but we
>> certainly don't need to download all demos with their corresponding
>> images and stuff.
>
> I completely agree, thisa code base has grown out of control. The
> reason for the code weight is the input images. The solution is to
> handle the input images like the code. Enric did it with his latest
> demo; I intend to do it systematically once I come back to active demo
> development (it's been too long..) [*].
>
> Without these input files, the repository weight would be much
> smaller. But the default code tree would still contain the 25 (and
> counting...) demos. The second step would be to split the code in 2:
> * the common code, plus an example
> * all the demos
> This way, one could download everything needed to build a demo without
> getting all the other demos.
>
>> I suggest to use git submodules for the different
>> user demos. In this way, you can download the base class and an
>> example to get started and later on you can download the rest.
>
> Are submodules needed? just 2 repositories, and one parameter in the
> common code to tell where to look for the demos, could be enough, no?
>

Yes, that would be enough, the submodules thing was only intended to
mantain the current structure of subdirectories.

> [*] The 'correct but slow' path could be
>    - adopt some guidelines extentions for automated build
>    - implement automated download and build in the demo
>    - adopt similar (bur simpler) guidelines for data sets
>    - use these quidelines, together with the automated download
>      feature, for the demo input
>

In addition, this will became worse when the N optic flow demos are
included, because they will all possibly (and maybe reasonably) share
the input images, so it realy make sense to decouple in three parts:
base demo+user demos+datasets.
And this willl happen with every major area in image processing
(segmentation, denoising, etc)

Anyway, thanks for the answer.
Best regards,
                  Juan