Re: Bug#764606: dgit and upstream git repos [and 1 more messages]
Dimitri John Ledkov writes ("Bug#764606: dgit and upstream git repos [and 1 more messages]"):
> On 9 October 2014 17:24, Ian Jackson <email@example.com> wrote:
> > However,
> > $ apt-get source sword
> > $ cd sword-*
> > $ rm -rf .pc
> > $ echo 'some stupid change' >>AUTHORS
> Invalid transition. dgit/git-dpm would error out that tree does not
> match HEAD commit.
> but none the less
dgit certainly shouldn't. It is a basic principle of dgit that you
are entitled to use it to edit the source code. That's its purpose.
The primary principle is that dgit's user should not need to know
anything about the maintainers' workflow, branch structures, package
format choices, etc. etc. The dgit user should get the package's
actual source code, and be free to make any fast-forwarding commits to
it, and push the result.
Ability of people who do know about the maintainer's workflow, to do
something more sophisticated, is a subsidiary goal.
> > $ dpkg-source -b .
> > [error message about uncommitted changes]
> > $ dpkg-source --commit
> > [add commit message]
> > [hideous pile of crap from attempting to reapply all patches]
> > $8
> > The question is this, then: given the result of the first four steps2
> > above, how to produce a correct source package ?
> I believe/assert:
> (a) dpkg-source --commit should work in this case
That would be nice.
Another problem is that this means that dgit source trees might not be
buildable other than by dgit, because dpkg-buildpackage would run
dpkg-source in the `wrong' way.
> (b) may or may not need extra help with --no-preparation,
> --unapply-patches, --auto-commit, --abort-on-upstream-changes
> I'll experiment more when i have more time to dig into this. Not gonna
> burn the midnight oil over this.