Andre Robatino <robatino <at> fedoraproject.org> writes:
Gordon Messmer <gordon.messmer <at> gmail.com> writes:
> On 09/27/2015 04:01 PM, Andre Robatino wrote:
> > Most of those people have the updates-testing repo enabled by default, so
> > they tend not to notice broken deps like this. Some kind of automation is
> > necessary to reliably prevent it.
>
> Looking at bodhi, it doesn't appear that anyone was unaware that the
> update required a newer sqlite, and would be delayed if that update
> didn't move to stable.
Looking at
https://bodhi.fedoraproject.org/updates/FEDORA-2015-16455 , it's
clear that Martin Stransky wasn't aware of it when he submitted it
("Unfortunately I can't remove this update....") and in any case it's
policy
that groups of packages with mutual deps are supposed to be submitted
together as a single update.
Oh I see, you mean none of the testers seemed to be unaware. It's unclear
actually, it looks like there was confusion over whether it would be
possible to fix the update before it went to stable (it wasn't in this
case), and since it was a security update people didn't want to delay it
since sqlite could be installed from updates-testing if necessary. But
automation could have kept it from being submitted in the first place. If a
group of packages is submitted, either that group of packages breaks stable
deps when submitted all by itself, or it doesn't. In principle there's no
reason it can't be automated and prevent any broken deps from ever hitting
stable.