Re: Bug#692000: ITP: liblbfgs -- L-BFGS solver for dense nonlinear optimization problems

[Dima Kogan <dima@secretsauce.net>]

> On Mon, 12 Nov 2012 08:36:46 +0100
> S__bastien Villemot <sebastien@debian.org> wrote:
>
> Dima Kogan <dima@secretsauce.net> writes:
> 
> >> On Sun, 11 Nov 2012 22:57:41 +0100
> >> Andreas Tille <tille@debian.org> wrote:
> >>
> >> On Sun, Nov 11, 2012 at 01:29:17PM -0800, Dima Kogan wrote:
> >> > >    dpkg-source: error: can't build with source format '3.0 (quilt)': no upstream tarball found at ../libdogleg_0.08.orig.tar.{bz2,gz,lzma,xz}
> >> > > 
> >> > > Any reason to not use pristine-tar?
> >> > 
> >> > I can use pristine-tar here, but I'm not completely clear on why it is useful.
> >> > The upstream source comes from a tag in a git repo. This tag fully describes the
> >> > sources. One could make a tarball from this tag (this is what pristine-tar would
> >> > do), but why would this be a useful thing to do? Tell me if having a
> >> > pristine-tar branch is desirable, and I'll set it up.
> >> 
> >> Pristine-tar ensures a tarball with identical MD5 sum which is pretty
> >> important if different people try to upload different package versions
> >> of the same upstream version.  In case somebody else than me who did the
> >> initial upload with MD5sum abc tries to create a Debian package '*-2'
> >> and does not download the tarball via dpkg-source he will end up with a
> >> different upstream tarball (not byte different when unpackaging but with
> >> different md5sum).  This will result in a rejection from ftpmaster.
> >
> > I'm not sure this is entirely correct. The contents of the upload are dictated
> > by the .changes file, which is constructed by dpkg-genchanges. dpkg-genchanges
> > has logic in it to only include the .orig.tar.gz when a new upstream version is
> > being built (based on debian/changelog). Thus if somebody builds a *-2 package
> > and debian/changelog says that a *-1 exists, dpkg-genchanges won't ship the
> > *.orig.tar.gz at all, since one presumably exists on the master server already.
> > I guess I don't like the redundant data that the pristine-tar archive contains,
> > and would prefer not to use it without a compelling reason.
> 
> The checksum of the orig tarball is not in the *-2 changes file, but it
> is in the *-2 dsc file that you will upload. So the problem that Andreas
> described will hit.

Aha. Thanks for clarifying. This still seems silly when used together with git,
but it looks like there are new work-in-progress source formats such as 3.0(git)
to resolve it.

I added pristine-tar to both liblbfgs and libdogleg. I did make sure that the
liblbfgs tarball matches exactly the one that Andreas uploaded earlier.

Thanks again

dima