On Tue, 2012-02-07 at 09:39 +0000, Manuel A. Fernandez Montecelo wrote: > 2012/2/7 Sam Morris <s...@robots.org.uk>: > > No problem. I looked into the conffile issue a bit further and I think > > that the dpkg-maintscript documentation needs to be changed. I filed > > <http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=658854> where you can > > find a brief description of the issue; in summary, I think you should > > set the 'lastversion' to 1.7.3-7~ in the next revision of the package > > (assuming that the next revision will be 1.7.3-7 of course). The > > existing documentation would suggest you use 1.7.3-6, but this causes > > the conffile removal code to be skipped in the case of a user rebuilding > > the package with a local version number: the user would be upgrading > > from 1.7.3-6local1, which is greater than 1.7.3-6 and therefore > > rm_conffile thinks that it doesn't need to do anything. Any possible > > (sane) local version number is guaranteed to be << 1.7.3-7~ however, and > > so that version number is safe to use when invoking rm_conffile. > > Actually, I'm creating a new upstream package (1.7.4-1) because it was > released in mid January.
Ok, then I suggest 1.7.4-1~ when invoking dpkg-maintscript-helper. > Yesterday when past midnight I gave up when testing this, because with > my tests I get the file renamed and backed up sometimes, e.g. when > using the file from -4 or so (cannot recall the actual details). In > -5 the paths within the plugins.cfg had to be changed because of > multi-archi-fying, and in -6 the file moved to /etc/OGRE-VERSION/ to > make possible to install different versions of OGRE at once. A bit of > a mess, yes, and ultimately to remove the file altogether. > > Maybe the helper assumes that you've installed each and every version > and just compares with the md5sum of the last version used? The version check is just there so that the code for removing/renaming a conffile is only run once. The problem I ran into is that I'd rebuilt the package, bumping the version number past that which the new package's maintainer scripts invoked rm_conffile with. Hence the bug I filed on dpkg to get the documentation sorted out. Then will come lintian checks, perhaps... :) > I'm using this right now: > ----------- > $ cat debian/libogre-VERSION.preinst > #!/bin/sh -e > > if dpkg-maintscript-helper supports rm_conffile 2>/dev/null; then > dpkg-maintscript-helper rm_conffile /etc/OGRE/plugins.cfg 1.7.4-1 > libogre-1.7.3 -- "$@" > dpkg-maintscript-helper rm_conffile /etc/OGRE-1.7.3/plugins.cfg 1.7.4-1 > libogre-1.7.3 -- "$@" > fi > > $ md5sum debian/libogre-VERSION.{preinst,postinst,postrm} > 9dc129ad3931e9a550ca9f65f7ee800e debian/libogre-VERSION.preinst > 9dc129ad3931e9a550ca9f65f7ee800e debian/libogre-VERSION.postinst > 9dc129ad3931e9a550ca9f65f7ee800e debian/libogre-VERSION.postrm > ----------- That looks fine to me (though the version number should be changed). > So I will need to test all of the possible combinations to try to > understand it and fix it for good this time. > > BTW, I moved the packaging to: > http://anonscm.debian.org/gitweb/?p=pkg-games/ogre.git;a=summary > > I have to synchronise last changes from yesterday (will try to do it > within the next hours), but maybe you find it helpful. Great. I'll test this and let you know what happens. > > Cheers. > > > PS: are you trying to backport the package or something? Yeah, just for my own build servers. We need to use Boost 1.48 hence I rebuild the package. -- Sam Morris <s...@robots.org.uk> -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org