On 09/18/2015 01:14 PM, Alexis Ballier wrote: > On Fri, 18 Sep 2015 13:04:45 +0200 > hasufell <hasuf...@gentoo.org> wrote: > >> On 09/18/2015 12:56 PM, Alexis Ballier wrote: >>> On Fri, 18 Sep 2015 11:58:09 +0200 >>> hasufell <hasuf...@gentoo.org> wrote: >>> >>>> On 09/18/2015 11:55 AM, Duncan wrote: >>>>> Alexis Ballier posted on Fri, 18 Sep 2015 11:04:19 +0200 as >>>>> excerpted: >>>>> >>>>>>> Keep in mind what this implies when you change these >>>>>>> dependencies without bumping the ebuilds that use them. >>>>>> >>>>>> only way i see these changing is with a new ros_messages_*** >>>>>> useflag, which will cause a rebuild anyway >>>>> >>>>> ?? Only with --newuse or similar, tho? Otherwise USE (or >>>>> USE_EXPAND here) changes don't trigger rebuilds, do they? >>>>> >>>> >>>> Correct. It's not good to rely on this and expect users to have a >>>> certain update pattern or even use a particular PM. >>>> >>> >>> then they wont have the messages if they don't rebuild, and >>> cat/pkg[ros_messages_newthing] deps wont be satisfied and the pm >>> will do the right thing by rebuilding. >>> nothing to worry about, really. >>> >> >> cat/pkg[ros_messages_newthing] will not be in users VDB until you >> revbump cat/pkg, so I don't think it's that easy. >> > > and i think you're confused about what dynamic deps is and is not: > cat/pkg[ros_messages_newthing] is not in vdb, so is not satisfied, so > pm tries to find something that satisfies it from tree or dies. >
I'm not confused about dynamic deps, but about your example, because you didn't say anything about revbumps. If cat/pkg is not (rev)bumped, then there will be no check about whether cat/pkg[ros_messages_newthing] is satisfied in the first place. That was my point.
