On 23/02/14 08:00 PM, David Kalnischkies wrote: > On Sat, Feb 22, 2014 at 04:59:27AM -0500, Daniel Dickinson wrote: >> As you can see from the bug history gnuplot is being installed by apt >> even though no packages which Depends on or Recommends on gnuplot are >> either directly or indirectly installed (based on binary package >> dependencies). > > There is the possibility that apt wanted to install one of the reverse > dependencies but decided against it later on and chooses another path. > apt currently can't reverse in such situations. It gets right of > "obvious" wrong-doings later on by running autoremove over the result, > but suggests are important enough to safe a package from disposal by it. Interesting. So there is a possibility that in the dependency resolution/traversal that it gets added due to it being a possible path and suggests keeps it from being removed? > That said, trying it with your "list" (it still includes packages > multiple times and also has lines with multiple packages) Sorry about that. The list is an amalgamation of package lists from a pared down version of the debian-live project's rescue list (and associated lists) and my own lists. For the list attached I just concatenated the lists and removed conditional used by live build to allow for only include items in the final list based on things like target architecture or whether or not one has enabled non-free section of archive for the build. I can't > reproduce your problem. gnuplot appears here only in the list of > suggested packages which are not installed… > (I had to remove samba from the list, as it isn't installable for me at > the moment) > > touch /tmp/status > apt-get install -o Debug::pkgDepCache::Marker=1 -o Dir::state::status=/tmp/status -st testing $(cat newlist) > > So, do you have some hints on how to reproduce your problem? I can supply the original package lists (actually tarball of entire live-build config) and you can run it through live-build (or I can send you the output of live-build, which includes the apt run, but is rather verbose and weighs in at 7.6 MB; actually I need to redo the run anyway since I have removed that build to retry, so I'll first see of running the build command without --verbose is still verbose enough for what need without the excessive size). To be clear and in case you are unaware I originally reported this against live-build which is part of the Debian-Live project and is system for generating bootable images (for example bootable CDs or USB sticks) with a minimally tweaked Debian (in my case Wheezy + a couple of backports packages) system (tweaking is mostly for doing config things like locales that are normally set up by debian-installer). The reason I reassigned to apt is that the live-build maintainers claim the package lists are just sent to apt and that they do nothing special that would cause apt to install packages that apt would not otherwise install. TBH I didn't try running apt manually I just fed the lists to live-build and it did whatever it does. I will see if I can reproduce in a chroot or linux container independent of live-build. I suspect the possibility that live-build is temporarily installing some package that causes the issue (looking at the live-build scripts and the lb build --verbose output I can see that there are packages that do get added and later removed from the chroot). The live-build maintainers didn't seem to really pay much attention to bug in the first place as they rather quickly closed with 'no bug here', and when I asked directly they said that yes (as you can see in the bug log) they claim that if there is a bug it is in apt. Perhaps it's because the part of the report where I show that grepping the output for each and every immediate ancestor (rdepends/rrecomends) of gnuplot came up empty got silently rejected by the BTS (or other mail server) due to the size of the requested build.log that I attached (and failed to notice the size due to high speed connection and not really waking up when Icedove popped up some question, which I realize now was probably a complaint about size). >> In as described previously, 'aptitude why gnuplot' on the >> resulting system claims the reason that gnuplot was installed is that >> amanda-client Suggests it. While amanda-client *is* delibarately >> installed and does suggest gnuplot, there is no reason that a Suggests >> should automatically cause a package to be installed (that is no reason >> except, if it is possible to do so, explicitly enabling such behavior, >> which is not done in this case). > > As far as I know 'aptitude why' displays the first reason it can find > and for aptitude suggests are a reason. They aren't automatically > installed by it though (but apt/aptitude have an option for it). > why more describes why a package isn't autoremoved… Ok, so it's possible that if there are multiple reasons that aptitude why wouldn't show the real one but the first one it found? In that case I will fire up aptitude and see if the info screen that lists why a package got installed (and IIRC includes every path, not just the first) has what I'm looking for. I guess I kind of assumed that aptitude why would show all reasons. >> BTW sorry for the multiple attempts to get the reassign and summary to >> apt maintainers vs live maintainers right. > > No worries, next time you can send corrections (and errors make even the > best people) to the control@ mail address only though as it avoid > cluttering the bugreport itself with messages. Will keep that in mind, thanks. Regards, Daniel
Attachment:
signature.asc
Description: OpenPGP digital signature