On Friday 13 November 2009 21:46:04 Mark Knecht wrote: > On Fri, Nov 13, 2009 at 10:42 AM, Alan McKinnon <alan.mckin...@gmail.com> wrote: > > On Friday 13 November 2009 14:39:52 Neil Bothwick wrote: > >> On Thu, 12 Nov 2009 16:58:15 +0200, Alan McKinnon wrote: > >> > Almost invariably it's an automagic dependency where the offending > >> > package is not in DEPEND. If you have been through the cycle at least > >> > once, it is safe to delete /var/lib/portage/preserved_libs_registry > >> > and continue on your way. > >> > >> Won't that leave orphaned libraries hanging around since they aren't > >> removed until emerges complete successfully? I've seen this behaviour > >> before, where the list gets shorter each time and let it run its course. > >> It may take longer, but you know it's safe. > > > > Interesting point. My tests before indicated that a full --depclean > > sorted everything out, but I can't be certain. @preserved-rebuild deletes > > orphans once it's complete, but it would be nice to verify what happens > > otherwise. > > > > Unfortunately, it's been a long time since any of my machines got stuck > > in this loop. I must have earned some good joss in recent months... -- > > alan dot mckinnon at gmail dot com > > If this problem is fundamentally due to dependencies not in DEPEND > then is there any evidence that it's big a problem? I.e. - there > aren't many packages that create the loop from what I've seen so far. > > I've had this issue show up on all the machines I've updated this > week, but it was always (I think) the same packages that caused the > problems. As Neil suggested, at least on one machine the number of > offending packages did seem to go down, but it would never go to zero > as far as I can tell. (I did it 3 times on one box just to convince > myself but emerging 50 packages gets boring.) While I haven't bug > reported it I suspect someone will jump on this and a few days or > weeks from now it won't exist, at least for these packages. > > Other than disk space what's the technical downside of some libraries > being stranded. Will this somehow leave applications pointing at old > library binaries?
The basic problem is that portage's idea of the state of the machine differs from reality. For a package manager, that's not a good thing as sooner or later it will do the wrong thing. Detecting orphans is also an expensive process later so it's best to avoid it happening if possible -- alan dot mckinnon at gmail dot com