On 12/19/18 1:05 AM, Philipp Kern wrote: > In the Ubuntu PPA case you get free reign over what's in that archive > and what you backport as part of offering the package. Obviously this > might conflict with the existing set. But the same is true for a > centralized volatile archive if you need to backport a large set of > build dependencies to make the whole thing work in the first place. And > I'm sure you wouldn't just go for a gitlab source package with bundled > build dependencies. That is why I prefer it as an extension of backports where the dependencies still follow the regular release cycle, they should be in testing and that means doing proper transitions for breaking changes and only the gitlab package itself be kept in volatile. > Now the policy question of who can ship what in a way that looks > semi-officially as being part of Debian is tricky. I personally find the > notion that testing should just be the staging ground for the next > release to be unfortunate but at the same time know how we ended up with > it. Maybe there's a place for packages that cannot usefully be supported > for a stable release and hence need to live in parallel. But then again > I wonder if the answer wouldn't be an archive where the input is built > for all suites and where the dependencies are bundled - if only because > you'd track upstream closely and would through that (hopefully) pull in > necessary security updates. I think it is still possible to maintain dependencies without bundling if it is like backports, ie, we can update them to newer upstream versions. Hence the requirement to redefine volatile. If they are not accepted in backports, a lot of packages will be duplicated, but even that is okay if backports team is not happy to take the new packages. And if this discussions go no where, the only option would be to make it an installer package, like diaspora-installer. But I'm not sure if it'd add much value over upstream provided omnibus packages. > Kind regards > Philipp Kern > >  And to some degree I am unhappy with backports' team's antagonistic > view on volatile here. Stuff like gitlab would have been rejected in the > same way it would've been in backports. The useful bits live on, it > wasn't abandoned to cause more work for backports. At the same time > backports can serve as a replacement of a subset of use cases too, where > its rules fit just fine. Thanks for sharing this.
Description: OpenPGP digital signature