Bug#38902: PROPOSED] data section
Steve Greenland <stevegr@debian.org> writes:
> A few additional rules for your consideration:
>
> - The data directory shouldn't be synced to debian releases, and ought
> to be paralled to dists, not main/contrib/non-free.
> (Since there are no executables, what's the benefit of syncing it, with
> the presumed multiplying of size and hassle? If a dataset needs a particular
> program or version, a simple dependency should be enough.)
>
> A thought: Do we need to keep a source archive separate from the .deb?
We need source for license reasons. Do we need debs?
Most dataset packages are real easy to compile, i.e. copy the data to
its dir. apt can download, unpack, compile and install source packages
and for datasets this is mostly so trivial that it cannot fail. I
think source packages would be enough.
> Almost all of these package are effectively their own source, and since
> we are talking about large datasets, the burden of keeping both seems
> unnecessary. Before you light up the flamethrower, I'm not promoting the
> idea of not releasing source. But consider a package that is basically
> a reproduction of a website. Do we really need two 7Mb packages who's
> fundamental difference is that they unpack into a different location?
> Instead, we could make the .debs act like installers, except that
> they would grab the external archive from the CD or via http. Or have
> a /usr/doc/<package>/debian, whose rules file has a way to copy the
> installed data into a new tree in order to rebuild the package.
Distributing datasets as source also has the advantage that we can
distribute diffs. Its stupid to distribute 50 MB of datas in a deb
file, just because 5K of data has changed. With sources the maintainer
would make a snapshot of the data to generate a orig.tar.gz and then
make diffs to current data. If the diff grows to much, make anotehr
orig.tar.gz and so on.
May the Source be with you.
Goswin
Reply to: