On Sat, Dec 8, 2012 at 12:25 AM, Matt Turner <matts...@gentoo.org> wrote: > I have never once been able to grab a portage snapshot and build a > stage 1, 2, 3 series from it without encountering at least a couple of > problems with the tree.
Ditto - the latest issue I've run into is: 443472. Probably won't impact the average user, and perhaps I should just modify the script to not bother reading that file and figure out what the latest build is on its own. > > I think we should consider things that break release media serious > regressions. I don't know what that entails specifically, but whether > it need be QA bashing down your door or a quick fix or revert, it sure > would be nice to get Gentoo to a place where release media always > works. Agreed. If a user can't just burn a CD and then do an emerge kde-meta there is a problem. > > In short, I think the conversation we should be having should be about > how to avoid breaking release builds and how to quickly fix problems > when they're discovered. I think those working with catalyst have the most to add regarding what breaks them. As far as detect-ability goes, do we need some kind of tinderbox that just does a daily build. Perhaps just build from stage3 to a couple of world targets, including one with some server-oriented software, one with gnome, and one with kde? I've reported a bunch of bugs with the EC2 bootstrap script described on my blog (not my original work), but it is only automated from a build perspective - testing is manual. That takes about 18 hours to build (with an emphasis on economy), and I use spot instances so it really only costs maybe a buck or two. My experience has been that if it builds it usually works. So, simply testing for whether the build completes is a pretty-good first step. Of course, for system packages our recourse isn't necessarily good, since we don't have a test branch or anything like that. If we wanted to revert we'd have to impact users who already upgraded. Obviously the goal would be to fix in place with a new revbump, assuming that were possible. Rich
