Received: (at submit) by bugs.debian.org; 14 May 1997 07:37:30 +0000 Received: (qmail 10703 invoked from network); 14 May 1997 07:37:29 -0000 Received: from sweat.cs.unm.edu (root@198.59.151.200) by master.debian.org with SMTP; 14 May 1997 07:37:29 -0000 Received: by sweat.cs.unm.edu id m0wRYca-0003DZC (Debian Smail-3.2 1996-Jul-4 #2); Wed, 14 May 1997 01:37:12 -0600 (MDT) Message-Id: Date: Wed, 14 May 1997 01:37:12 -0600 (MDT) From: Barak Pearlmutter To: submit@bugs.debian.org Subject: dselect feature req (not user interface!) Package: dpkg Version: 1.4.0.8 When using the ftp method to access a repository, often the Packages files do not quite match the .deb files in the distribution. Almost always this is due to version skew, and some .deb files have slightly higher version numbers than the corresponding entries in the Packages file. When attempting to INSTALL, error like getting: big-long-dir/pack_vera-verb.deb (xxx) big-long-dir/pack_vera-verb.deb: No such file OR directory. This is a very very common problem, so basically the loop for upgrading a debian box by ftp is usually: Update Select Install (manually record bad files) (manually fetch new versions) (manually install new versions) (manually delete those that installed successfully) Quit (The reason this doesn't get you into trouble is that dpkg won't let you hurt yourself due to dependency problems. Bravo dpkg.) People have complained about this, and proposed various mechanisms for ensuring that the Packages files stay more consistent with the actual packages available. But given network reliability & mirroring software, this problem cannot be fixed entirely. How about if, instead, dselect and dpkg-ftp automated the above manual process: it is easy enough to notice the missing file and instead of just reporting it, get a DIR looking for newer versions in the appropriate directory, and present the user with a message like: can find: big-long-dir/pack_vera-verb.deb, but pack_(vera+1)-1.deb is available. Should I try to use that instead? And if this results in dependency problems later, well so be it. (Maybe there should be a chance for the user to resolve dependency problems after the fetch but before actual install, when the new control files are actually available. Or maybe not, if that would be a hassle to hack up.)