Packaging ACK workflow is sub-optimal.

[Robert Bruce Park]

Currently our packaging ACK workflow looks like this:

1. Dev creates MP that modifies a source package, potentially touching files in debian/
2. Dev requests a silo, and is granted one.
3. Dev builds his new package in a silo and spends some hours/days testing that it behaves as desired.
4. in the case of RTM, QA then spends some hours verifying that the silo doesn't regress the image.
5. Once the dev and QA approve the silo, we hit the publish button, and...
6. "Packaging changes require manual ACK"

Now the core dev is put in an awkward position, because they want to ensure the highest quality of packaging standards in ubuntu, but they also don't want to invalidate the hours/days of testing that went into steps 3 and 4.

What we need to do is move step 6 earlier in the process. I'm aware of a way that this could easily be transitioned from publish time to build time, but mterry and kenvandine feel it should be even earlier than that, even before silo-assignment time.

I'm not sure how to solve this but it's a discussion we need to have.

[Michael Terry (mterry)]

I'd propose that the silo bot enforces (around the same time in the process that it enforces top-approval) that any branch that modifies debian/ must have an approved review tagged as a "packaging" review. That way, projects know they need to seek such a review at MP time or eventually be rejected.

Maybe that needs to happen in conjunction with a final (presumably quick) ACK for the whole thing at the same point in the process we do it now? Just to help with reviewing branches that interact. Not sure on that. But I like the idea of an MP-time review.

[Robert Bruce Park (robru)]

The problem with enforcing it at the same time it enforces top-approval is
that it enforces top-approval at publish time, so that really would not be
appreciably different than the current situation (it already is rejecting
publication without ack today).

People already are supposed to know that the packaging review is coming,
but they already don't bother to seek it pre-publish.

I think the best we can hope for given the current tools is to implement an
assign-time check that rejects mps that touch debian without a packaging
ack on the mp, but a) that would be a lot of extra work to fetch the
necessary info, because it's currently only available at build time, and b)
there'd have to be an education campaign to make our upstreams seek
packaging acks prior to silo assignment.

[Robert Bruce Park (robru)]

Today I landed a major build.py rewrite, and among the changes are that those packaging diffs are now exposed at build time, instead of just at publish time.

This at least makes it possible for landers to seek packaging acks sooner in the process, but it will require some education of landers to make them seek it on their own prior to publication.

[Robert Bruce Park (robru)]

I guess the build job could detect that a packaging diff was generated and set the status to 'Packages built, but remember to get an ACK!' or something.