Received: (at submit) by bugs.debian.org; 23 Jan 1997 04:12:54 +0000 Received: (qmail 29226 invoked from network); 23 Jan 1997 04:12:44 -0000 Received: from turnbull.sk.tsukuba.ac.jp (root@130.158.99.4) by master.debian.org with SMTP; 23 Jan 1997 04:12:30 -0000 Received: from turnbull.sk.tsukuba.ac.jp by turnbull.sk.tsukuba.ac.jp with smtp (Linux Smail3.1.29.1 #3) id m0vnGPC-00001SC; Thu, 23 Jan 97 13:04 JST Message-Id: To: submit@bugs.debian.org Subject: dselect and mirrors with incomplete archives Date: Thu, 23 Jan 1997 13:04:50 +0900 From: "Stephen J. Turnbull" Package: dpkg Version: 1.4.0.6 I am using Debian 1.2, kernel version 2.0.27, and libc 5.4.13. I observed the following problem of inconsistency between Packages and the actual archive with a Debian mirror and reported it to the mirrors. Since I think it is a bug in the package management system, I am reporting it here. As I don't understand dpkg yet (I haven't successfully installed the system, so don't really have access to docs etc), I'm going to make some suggestions, but if they seem orthogonal to common sense, they probably are. I looked at titles to the bugs regarding dpkg, and didn't see anything related. With my current bandwidth, actually reading all the reports would take most of this week; sorry if I missed this one. >>>>> At 11:44 AM 23/01/97 +0900, I wrote: me> I have been having a lot of problems with installation from the me> Debian Linux mirror at ftp.kreonet.re.kr. me> The basic cause seems to be that the package list was mirrored me> _before_ getting all the packages. Thus many of the files are me> older than the version mentioned in the package list. In me> particular, ftp.kreonet.re.kr claims to have libc_5.4.17-X.deb, me> but only has libc_5.4.13-X.deb. Since many of the packages me> already updated depend on having at least libc 5.4.17, dselect me> goes crazy, and there are many errors in installation so dpkg me> can't recover. me> I suspect that this could be a problem at many overseas me> mirrors, as bandwidth really stinks recently. >>>>> "Karl" == Karl Ferguson responded: Karl> Unfortuantely we can't do much about that. If files Karl> time-out enough when mirroring and mirror decided to spit Karl> the dummy and exit, then you'll end up having updated files Karl> all over the place. It also depends on here the mirror is Karl> mirroring from. me> This really is a defect in the package management system. As me> it is not possible to guarantee timely updates of a full me> mirror, the package list should be generated on the fly at the me> mirror. Karl> Bugs with mirrors should be reported directly to the mirror Karl> admins (as you see listed in the README.mirrors file) - it's Karl> not really the mirrors responsability to have thir own Karl> package maintenance system, if we demanded this then I doubt Karl> we'd have any mirrors left at all - I'm a mirror myself, and Karl> I wouldn't want that. I agree with this; it seems to me that developing a script for generating Packages rests with the dpkg maintenance team. Since you presumably already have one, it should be possible to modify it for use by mirrors. I recognize that this could potentially be annoying because of various packages and versions of mirroring software, but since dselect is the first thing us newbies see, you could chase all of us back to RedHat and we wouldn't want that :-) Karl> Packages(.gz) etc are all updated and retrieved first during Karl> a mirror because it's first on the list. As I said before, Karl> we can't help it if mirror spits the dummy and exits due to Karl> errors. Right. At a minimum, Packages ought to be the *last* thing on the list, and *old* packages should not be purged (the Debian naming system guarantees they won't be overwritten, as far as I can see) until Packages is updated. Yes, this leaves you with an old mirror for a potentially long period of time, but at least it's consistent. An option would be to add a "stable-in-process" distribution for the adventurous. This would only appear at mirrors with relatively low bandwidth and long transition periods. Of course, both of these options require a lot more space of the mirror :-( Also, dselect should probably (optionally) recheck the dependencies after getting the packages to make sure that the packages received actually do satisfy the dependency tree. dselect doesn't seem very smart about large numbers of packages; maybe new users should be advised to do the installations incrementally. It also would be helpful to add an outline feature (hiding/showing contents below a header) to dselect. (Yeah, I know, I ought to put this in a separate report, but I don't feel like wrestling with the transmission lags to check the bug list.) Thanks for your attention to these problems. Stephen Turnbull -- Stephen J. Turnbull Institute of Policy and Planning Sciences Yaseppochi-Gumi University of Tsukuba http://turnbull.sk.tsukuba.ac.jp/ Tel: +81 (298) 53-5091; Fax: 55-3849 turnbull@sk.tsukuba.ac.jp