may be something like 3.0 (git-annex) source format?

[Chanoch (Ken) Bloom]

On Tue, 2012-01-24 at 15:46 -0400, Joey Hess wrote:
> Yaroslav Halchenko wrote:
> > I just thought to check with you on either the idea itself sounds any
> > reasonable...
> > 
> > in the frame of NeuroDebian we package "data" packages which are usually
> > quite large in size... there were numerous previous discussions on
> > viable ways to gain some convenience/size tradeoff, e.g. how to not
> > duplicate 1GB of data in .orig and .deb files.  But now that you
> > equipped us with git-annex , it might be time to research for new ways?
> > 
> > just a blunt idea -- what if 3.0 (git) format (with which I haven't
> > played before either) gets extended to provide git-annex'ed .git with
> > references to publicly (and local for buildds) available annex-ed
> > remotes and original urls (added via git annex addurl).  Then by
> > dropping of all load from .git  within .orig we achieve small
> > size, while providing ways to fetch the original load either from urls,
> > or git-annex-ed remotes.
> > 
> > Does it sound any sensible?
> 
> I hope that git-annex may be useful to teams working on large data in a
> VCS. I have only heard from a few projects doing this so far.
> 
> But, as a Debian source package goes, it's not sufficient for git-annex
> to know a remote or a url that has the data. That wouldn't help if using
> this source package on a desert island.
> 
> Now, maybe git-annex could be taught to use apt as a special remote, so
> it could know that a given file in the source package can be obtained
> by getting the corresponding binary package with apt and extracting the
> file from it. While somewhat weird and recursive, there's no technical
> reason that couldn't work. 
> 
> Whether the ftpmasters would accept such a source package, that instead
> of some of the sources, contains essentially a note pointing at the
> actual data in the binary package next to it; I don't know.

We have that already in Debian. It's called build-depends.