Re: Britney error with the gossip package?
On Tue, Oct 07, 2008 at 09:02:11PM +0100, Neil Williams wrote: On Tue, 07 Oct 2008 20:42:07 +0100 Adam D. Barratt [EMAIL PROTECTED] wrote: britney only considers installability, not buildability. Maybe it should - after all, in the phase prior to a release, the I'm not sure whether britney should or should not consider this (on one hand it would be better pre-emptive QA, on the other it can make it harder to transition for reasons that can be fixed a posteriori). Still, if you want to do that, I would like to advertise that edos-debcheck, starting from version 1.0-7, has introduced the edos-builddebcheck command line tool which is able to check for unsatisfiable *build*-dependencies. The implementation is kind of hackish (it is based on plain edos-debcheck, plus some package name mangling from Sources to Packages), but it works nicely. Cheers. -- Stefano Zacchiroli -*- PhD in Computer Science \ PostDoc @ Univ. Paris 7 [EMAIL PROTECTED],pps.jussieu.fr,debian.org} -- http://upsilon.cc/zack/ I'm still an SGML person,this newfangled /\ All one has to do is hit the XML stuff is so ... simplistic -- Manoj \/ right keys at the right time signature.asc Description: Digital signature
Re: Britney error with the gossip package?
On Tue, 2008-10-07 at 20:21 +0100, Neil Williams wrote: $ rmadison libloudmouth1-0 libloudmouth1-0 |1.4.0-1 | testing | alpha, amd64, arm, armel, hppa, i386, ia64, mips, mipsel, powerpc, s390, sparc libloudmouth1-0 |1.4.2-1 | unstable | alpha, amd64, arm, armel, hppa, hurd-i386, i386, ia64, m68k, mips, mipsel, powerpc, s390, sparc How did gossip migrate with Build-Depends on a version of libloudmouth1-0 that still does not exist in Lenny? britney only considers installability, not buildability. That is to say, if package A depends on B (= 2) then an appropriate version of B must (in the absence of hints to the contrary) exist in testing, but if A /build/-depends on B (= 2) then britney does not care whether B exists in testing at all, yet alone with the correct version. Regards, Adam -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]
Re: Britney error with the gossip package?
On Tue, 07 Oct 2008 20:42:07 +0100 Adam D. Barratt [EMAIL PROTECTED] wrote: On Tue, 2008-10-07 at 20:21 +0100, Neil Williams wrote: $ rmadison libloudmouth1-0 libloudmouth1-0 |1.4.0-1 | testing | alpha, amd64, arm, armel, hppa, i386, ia64, mips, mipsel, powerpc, s390, sparc libloudmouth1-0 |1.4.2-1 | unstable | alpha, amd64, arm, armel, hppa, hurd-i386, i386, ia64, m68k, mips, mipsel, powerpc, s390, sparc How did gossip migrate with Build-Depends on a version of libloudmouth1-0 that still does not exist in Lenny? britney only considers installability, not buildability. Maybe it should - after all, in the phase prior to a release, the ability to build the entire release from source *is* important. To me, this is precisely what Britney should be able to avoid - gossip migrated (irrespective of what happened with the unblock, this could easily happen between releases too) in a condition that would not build from source in testing. This would not have been a problem if gossip had been held back from testing migration (with or without a freeze or unblock). That is to say, if package A depends on B (= 2) then an appropriate version of B must (in the absence of hints to the contrary) exist in testing, but if A /build/-depends on B (= 2) then britney does not care whether B exists in testing at all, yet alone with the correct version. gossip does build against the version in testing *if* configure and configure.ac is patched to downgrade the LOUDMOUTH_REQUIRED variable. Presumably the build-dependency was increased for a bug fix but the symbols appear unchanged and gossip is not using any new symbols that may appear in the version of libloudmouth in unstable. The binary gossip package does not specify that version of libloudmouth so that is probably a bug in the gossip Debian packaging - an explicit dependency should have been specified to ensure that the release of libloudmouth containing the bug fix needed by gossip upstream was actually available to gossip at runtime. That would have prevented Britney migrating the package. So this is a combination of bugs: The gossip binary should have had an explicit dependency on libloudmouth1-0 (= 1.4.1) in debian/control.in that overrides the value derived from dpkg_shlibdeps so that gossip users actually got the benefit of the incremented Build-Depends. Britney could be improved to act as a safety-net for packages that fail to set such dependencies in order to prevent bigger problems, like this one. -- Neil Williams = http://www.data-freedom.org/ http://www.nosoftwarepatents.com/ http://www.linux.codehelp.co.uk/ pgpC2jNWCWBLd.pgp Description: PGP signature
Re: Britney error with the gossip package?
On Tue, 7 Oct 2008 21:02:11 +0100 Neil Williams [EMAIL PROTECTED] wrote: britney only considers installability, not buildability. Maybe it should - after all, in the phase prior to a release, the ability to build the entire release from source *is* important. To me, this is precisely what Britney should be able to avoid - gossip migrated (irrespective of what happened with the unblock, this could easily happen between releases too) in a condition that would not build from source in testing. Actually, maybe lintian can come to the rescue here - if the Build-Depends version is higher than the shared library dependency isn't that always going to be a problem? The reverse is fine, of course, but Build-Depends shlibs would appear to be a problem waiting to happen. -- Neil Williams = http://www.data-freedom.org/ http://www.nosoftwarepatents.com/ http://www.linux.codehelp.co.uk/ pgp0GeWDjFN6V.pgp Description: PGP signature
Re: Britney error with the gossip package?
Neil Williams [EMAIL PROTECTED] writes: I'm assuming there are records of unblocks beyond the mailing list archive? The release team's hint files are available: http://ftp-master.debian.org/testing/hints/ In this case: testing/hints/luk:unblock gossip/1:0.31-1 We move hints that were done down (under the finished line) for archiving purposes. How did gossip migrate with Build-Depends on a version of libloudmouth1-0 that still does not exist in Lenny? britney does not, and never has, check build-dependencies. Marc -- BOFH #250: Program load too heavy for processor to lift. pgpyBEQNxDV2A.pgp Description: PGP signature
Re: Britney error with the gossip package?
Neil Williams wrote: On Tue, 7 Oct 2008 21:02:11 +0100 Neil Williams [EMAIL PROTECTED] wrote: britney only considers installability, not buildability. Maybe it should - after all, in the phase prior to a release, the ability to build the entire release from source *is* important. To me, this is precisely what Britney should be able to avoid - gossip migrated (irrespective of what happened with the unblock, this could easily happen between releases too) in a condition that would not build from source in testing. Actually, maybe lintian can come to the rescue here - if the Build-Depends version is higher than the shared library dependency isn't that always going to be a problem? The reverse is fine, of course, but Build-Depends shlibs would appear to be a problem waiting to happen. Wasn't dpkg supposed to use max(shlibs, build-depends)? The rationale, IIRC, is because a particular program might rely on a specific bugfix in a given version of the library. Since bugfixes don't cause shlibs bumps, this was a way for the maintainer to enforce the correct functioning of the program. -- Felipe Sateler -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]
Re: Britney error with the gossip package?
Le mardi 07 octobre 2008 à 17:13 -0400, Felipe Sateler a écrit : Wasn't dpkg supposed to use max(shlibs, build-depends)? The rationale, IIRC, is because a particular program might rely on a specific bugfix in a given version of the library. Since bugfixes don't cause shlibs bumps, this was a way for the maintainer to enforce the correct functioning of the program. It only does so for packages with symbol files. The shlibs system doesn’t provide a mapping between shared library packages and development packages, which .symbols files do. -- .''`. : :' : We are debian.org. Lower your prices, surrender your code. `. `' We will add your hardware and software distinctiveness to `-our own. Resistance is futile. signature.asc Description: Ceci est une partie de message numériquement signée
