Open Stack

On 12/02/2012 08:49 PM, Mark Washenberger wrote:
> What's the reason behind not commiting the versioninfo file?

Because it's generated data, similar to AUTHORS and ChangeLog. The
versioninfo file only exists to provide information in our sdist
tarballs so that access to git is not required.

A lot of the automation in this area is about ensuring that there is one
location of truth for pieces of information that we have. Remember back
when AUTHORS files were something that we stored in git? And then we had
a unit test that parsed AUTHORS to check to ensure that the people in
the git repo commit messages were all accounted for in the AUTHORS file?
That was a giant pain and people hated it - so we made the AUTHORS file
generated from the git log data at sdist time. That way, for both of the
ways you would consume the output, our released sdist tarballs, or from
git itself, you would be assured of having accurate information.

Current bug (that has a fix) notwithstanding, the work with the versions
from git is driving at a similar thing.

> It seems like git-buildpackage isn't wrong for thinking that you
> should be able to do package builds from a git-archive result.

That makes the assumption that git-archive results are usable. In our
case, they are not. You should not consume them. You will notice that we
have them turned off in github. We do produce and published properly
built tarballs to both tarballs.openstack.org and to launchpad.net. If
you want to build them yourself, it's really easy - you run "python
setup.py sdist"

So people doing debian packaging with git-buildpackage should be running
git import-orig on one of those release tarballs, or if they are working
from a local fork, they should be generating tarballs with python
setup.py sdist and not with git-archive.

Monty