Bug#903211: release.debian.org: How to handle unbuildable packages in buster

[Niels Thykier <niels@thykier.net>]

On Sat, 7 Jul 2018 19:17:11 +0200 (CEST) Santiago Vila <sanvila@unex.es>
wrote:
> Package: release.debian.org
> Severity: wishlist
> 
> Dear Release Managers:
> 

Hi Santiago,

Thanks for working on QA in testing.

> I'm looking for guidance and advice here.
> 
> I'm reporting FTBFS bugs in testing, but I'm skipping bugs of type
> "unmet build-depends" because I've heard somewhere that Britney takes
> care of that.
> 
> However, I don't see the procedure is working 100% correctly (maybe
> because of transitive build-dependencies/dependencies).
> 

Britney has partial support for Build-Depends at the moment.  What you
are experiencing is probably the limitations of this support.

What Britney can do is:

 * Reject an upload if it has a Build-Depends or Build-Depends-Arch
   relation that can only be satisfied in unstable and the packages
   needed from unstable are also rejected.

The limitation appear in the following cases:

 1) Build-Depends-Indep is not checked at all.  This is a known issue
    and we are working on it.  (I just committed some infrastructure
    code in britney for it yesterday, but the actual check is still
    missing)

 2) Build-Depends are not enforced at migration time.  I.e. if the
    relation can be satisfied in unstable by packages that are "not
    rejected", then an upload can migrate /before/ its Build-Depends.

 3) Build-Depends are not enforced on removal.  That is packages
    /can/ be removed while packages are build depending on them.

Limitation 2+3 are slightly more involved and may take quite a while for
us to implement.

> The question: It is still ok to submit FTBFS bugs so that the rule
> "package in buster must be buildable in buster" is met?
> 
> Thanks.
> 
> 

I wish I could say unconditionally yes, but I am not sure we are ready
for it yet.  If your testing can trivially say which relation in
Build-Depends/Build-Depends-Arch is broken, you could cross reference it
with the [excuses] and see if they are permanently rejected due (and why).

I am not sure how feasible that is for you, but it might enable you to
file RC bugs for a subset of the issues already now.

Thanks,
~Niels

[excuses]: https://release.debian.org/britney/excuses.yaml