Re: glibc and abi-compliance-checker break multiple KDE autopkgtests
Dear all, On 09-09-18 22:04, Paul Gevers wrote: > On 09/06/18 21:13, Paul Gevers wrote: >> So it seems they are requested by something, and because the are not >> available in testing, apt-get is not limited by our pinning to take them >> from unstable. I believe it must be a "Provides" of some sort. What I >> want to know (and I will spend some time on it) is what in the >> dependency chain makes us end up with this as an option. > > I was not able to figure out in the time I spend on it why apt-get ended > up with installing those packages. Does anybody know the right commands > and their arguments to figure out this specific question? > > Paul > PS: the last line that I used was: > apt-cache dotty $(echo $(apt-cache showsrc kf5-kdepim-apps-libs | grep > Binary | sed s/Binary:// | sed s/,//g) | sort --unique) dh-acc > exuberant-ctags | grep '++8' I went ahead and let glibc and abi-compliance-checker migrate to testing. I ran reference runs of the failing tests today to check if everything is all-right, and it is. So this is a versioned Depends/Breaks/Conflicts issue somewhere. I am very uncomfortable about this, because today qtbase-opensource-src started to be hit by the same issue. Paul @ qtbase-opensource-src maintainers, this conversation starts here: https://lists.debian.org/debian-ci/2018/09/msg00010.html signature.asc Description: OpenPGP digital signature
Re: glibc and abi-compliance-checker break multiple KDE autopkgtests
Hi all, On 09/06/18 21:13, Paul Gevers wrote: > So it seems they are requested by something, and because the are not > available in testing, apt-get is not limited by our pinning to take them > from unstable. I believe it must be a "Provides" of some sort. What I > want to know (and I will spend some time on it) is what in the > dependency chain makes us end up with this as an option. I was not able to figure out in the time I spend on it why apt-get ended up with installing those packages. Does anybody know the right commands and their arguments to figure out this specific question? Paul PS: the last line that I used was: apt-cache dotty $(echo $(apt-cache showsrc kf5-kdepim-apps-libs | grep Binary | sed s/Binary:// | sed s/,//g) | sort --unique) dh-acc exuberant-ctags | grep '++8' signature.asc Description: OpenPGP digital signature
Re: glibc and abi-compliance-checker break multiple KDE autopkgtests
Hi On 06-09-18 16:39, Antonio Terceiro wrote: >>> Does this mean that libc++-8-dev is breaking the ABI of the Qt/KDE >>> packages? Luckily libc++-8-dev will not migrate to testing due to >>> https://bugs.debian.org/714686 Does it need a "Breaks" then? >> >> Actually due to a bug in the migration process this package migrated to >> testing on 2018-08-26 despite the RC bug. It has been removed from >> testing during last night. >> >>> Does anybody know why libc++-8-dev is installed when glibc or >>> abi-compliance-checker come from unstable? It seems that package is >>> providing something that in testing is provided by libc++-dev (Or >>> somewhere else in the dependency chain this goes "wrong" and leads to >>> this outcome). >> >> I have been able to install libc++-dev along glibc 2.27-6, so I wonder >> if it is not just a matter of regenerating the testing chroot following >> the libc++-8-dev removal from testing. > > the containers on ci.debian.net are recreated from scratch once a day, > so this should solve itsef, I guess? I don't know. In the failing cases, libc++-8-dev was installed from unstable, not from testing: Get:2 http://deb.debian.org/debian unstable/main amd64 libc++abi1-8 amd64 1:8~svn340819-1 [81.3 kB] Get:3 http://deb.debian.org/debian unstable/main amd64 libc++1-8 amd64 1:8~svn340819-1 [214 kB] Get:4 http://deb.debian.org/debian unstable/main amd64 libc++-8-helpers all 1:8~svn340819-1 [28.3 kB] Get:5 http://deb.debian.org/debian unstable/main amd64 libc++-8-dev amd64 1:8~svn340819-1 [1,473 kB] So it seems they are requested by something, and because the are not available in testing, apt-get is not limited by our pinning to take them from unstable. I believe it must be a "Provides" of some sort. What I want to know (and I will spend some time on it) is what in the dependency chain makes us end up with this as an option. Paul signature.asc Description: OpenPGP digital signature
Re: glibc and abi-compliance-checker break multiple KDE autopkgtests
On Thu, Sep 06, 2018 at 03:55:04PM +0200, Aurelien Jarno wrote: > On 2018-09-06 12:13, Paul Gevers wrote: > > Hi > > > > On 06-09-18 11:53, Samuel Thibault wrote: > > > Samuel Thibault, le jeu. 06 sept. 2018 11:44:45 +0200, a ecrit: > > >> Paul Gevers, le jeu. 06 sept. 2018 11:22:46 +0200, a ecrit: > > >>> On 06-09-18 11:19, Samuel Thibault wrote: > > It'd be useful for the abi-compliance-checker test to actually output > > error messages, > > > > https://ci.debian.net/data/autopkgtest/testing/amd64/k/kf5-kdepim-apps-libs/944759/log.gz > > > > it not very talkative :) > > >>> > > >>> I agree, but I found that there are more logs in the artifacts. > > >> > > >> Ah, right. They seem to only point at c++ headers, so it'd rather be a > > >> g++ issue? > > > > > > All the passed artifacts I can find have libc++-dev 6.0.1-1, not > > > libc++-8-dev 1:8~svn340819-1. > > > > Does this mean that libc++-8-dev is breaking the ABI of the Qt/KDE > > packages? Luckily libc++-8-dev will not migrate to testing due to > > https://bugs.debian.org/714686 Does it need a "Breaks" then? > > Actually due to a bug in the migration process this package migrated to > testing on 2018-08-26 despite the RC bug. It has been removed from > testing during last night. > > > Does anybody know why libc++-8-dev is installed when glibc or > > abi-compliance-checker come from unstable? It seems that package is > > providing something that in testing is provided by libc++-dev (Or > > somewhere else in the dependency chain this goes "wrong" and leads to > > this outcome). > > I have been able to install libc++-dev along glibc 2.27-6, so I wonder > if it is not just a matter of regenerating the testing chroot following > the libc++-8-dev removal from testing. the containers on ci.debian.net are recreated from scratch once a day, so this should solve itsef, I guess? signature.asc Description: PGP signature
Re: glibc and abi-compliance-checker break multiple KDE autopkgtests
On 2018-09-06 12:13, Paul Gevers wrote: > Hi > > On 06-09-18 11:53, Samuel Thibault wrote: > > Samuel Thibault, le jeu. 06 sept. 2018 11:44:45 +0200, a ecrit: > >> Paul Gevers, le jeu. 06 sept. 2018 11:22:46 +0200, a ecrit: > >>> On 06-09-18 11:19, Samuel Thibault wrote: > It'd be useful for the abi-compliance-checker test to actually output > error messages, > > https://ci.debian.net/data/autopkgtest/testing/amd64/k/kf5-kdepim-apps-libs/944759/log.gz > > it not very talkative :) > >>> > >>> I agree, but I found that there are more logs in the artifacts. > >> > >> Ah, right. They seem to only point at c++ headers, so it'd rather be a > >> g++ issue? > > > > All the passed artifacts I can find have libc++-dev 6.0.1-1, not > > libc++-8-dev 1:8~svn340819-1. > > Does this mean that libc++-8-dev is breaking the ABI of the Qt/KDE > packages? Luckily libc++-8-dev will not migrate to testing due to > https://bugs.debian.org/714686 Does it need a "Breaks" then? Actually due to a bug in the migration process this package migrated to testing on 2018-08-26 despite the RC bug. It has been removed from testing during last night. > Does anybody know why libc++-8-dev is installed when glibc or > abi-compliance-checker come from unstable? It seems that package is > providing something that in testing is provided by libc++-dev (Or > somewhere else in the dependency chain this goes "wrong" and leads to > this outcome). I have been able to install libc++-dev along glibc 2.27-6, so I wonder if it is not just a matter of regenerating the testing chroot following the libc++-8-dev removal from testing. Aurelien -- Aurelien Jarno GPG: 4096R/1DDD8C9B aurel...@aurel32.net http://www.aurel32.net signature.asc Description: PGP signature
Re: glibc and abi-compliance-checker break multiple KDE autopkgtests
Hi On 06-09-18 11:53, Samuel Thibault wrote: > Samuel Thibault, le jeu. 06 sept. 2018 11:44:45 +0200, a ecrit: >> Paul Gevers, le jeu. 06 sept. 2018 11:22:46 +0200, a ecrit: >>> On 06-09-18 11:19, Samuel Thibault wrote: It'd be useful for the abi-compliance-checker test to actually output error messages, https://ci.debian.net/data/autopkgtest/testing/amd64/k/kf5-kdepim-apps-libs/944759/log.gz it not very talkative :) >>> >>> I agree, but I found that there are more logs in the artifacts. >> >> Ah, right. They seem to only point at c++ headers, so it'd rather be a >> g++ issue? > > All the passed artifacts I can find have libc++-dev 6.0.1-1, not > libc++-8-dev 1:8~svn340819-1. Does this mean that libc++-8-dev is breaking the ABI of the Qt/KDE packages? Luckily libc++-8-dev will not migrate to testing due to https://bugs.debian.org/714686 Does it need a "Breaks" then? Does anybody know why libc++-8-dev is installed when glibc or abi-compliance-checker come from unstable? It seems that package is providing something that in testing is provided by libc++-dev (Or somewhere else in the dependency chain this goes "wrong" and leads to this outcome). This is maybe also the reason why the autopkgtest of these packages fail already longer in unstable. Apparently the wrong libc++*-dev package gets installed. Paul PS to all: if we determine how this goes wrong and if I can't get the CI framework to do the right thing and this is outside of glibc and abi-compliance-checker, I'll ask the migration software to ignore these failures. signature.asc Description: OpenPGP digital signature
Re: glibc and abi-compliance-checker break multiple KDE autopkgtests
Samuel Thibault, le jeu. 06 sept. 2018 11:44:45 +0200, a ecrit: > Paul Gevers, le jeu. 06 sept. 2018 11:22:46 +0200, a ecrit: > > On 06-09-18 11:19, Samuel Thibault wrote: > > > It'd be useful for the abi-compliance-checker test to actually output > > > error messages, > > > > > > https://ci.debian.net/data/autopkgtest/testing/amd64/k/kf5-kdepim-apps-libs/944759/log.gz > > > > > > it not very talkative :) > > > > I agree, but I found that there are more logs in the artifacts. > > Ah, right. They seem to only point at c++ headers, so it'd rather be a > g++ issue? All the passed artifacts I can find have libc++-dev 6.0.1-1, not libc++-8-dev 1:8~svn340819-1. Samuel
Re: glibc and abi-compliance-checker break multiple KDE autopkgtests
Paul Gevers, le jeu. 06 sept. 2018 11:22:46 +0200, a ecrit: > On 06-09-18 11:19, Samuel Thibault wrote: > > It'd be useful for the abi-compliance-checker test to actually output > > error messages, > > > > https://ci.debian.net/data/autopkgtest/testing/amd64/k/kf5-kdepim-apps-libs/944759/log.gz > > > > it not very talkative :) > > I agree, but I found that there are more logs in the artifacts. Ah, right. They seem to only point at c++ headers, so it'd rather be a g++ issue? Samuel
Re: glibc and abi-compliance-checker break multiple KDE autopkgtests
Hi Samuel, all, On 06-09-18 11:19, Samuel Thibault wrote: > It'd be useful for the abi-compliance-checker test to actually output > error messages, > > https://ci.debian.net/data/autopkgtest/testing/amd64/k/kf5-kdepim-apps-libs/944759/log.gz > > it not very talkative :) I agree, but I found that there are more logs in the artifacts. Paul signature.asc Description: OpenPGP digital signature
Re: glibc and abi-compliance-checker break multiple KDE autopkgtests
Hello, Paul Gevers, le jeu. 06 sept. 2018 10:55:30 +0200, a ecrit: > the autopkgtest of libkf5calendarsupport, kf5-kdepim-apps-libs and > akonadi-calendar started to fail when run in testing with either glibc > or abi-compliance-checker from unstable. It'd be useful for the abi-compliance-checker test to actually output error messages, https://ci.debian.net/data/autopkgtest/testing/amd64/k/kf5-kdepim-apps-libs/944759/log.gz it not very talkative :) On Linux, the glibc 2.27-6 change only contains two additional static functions, so at best it'd be an issue with the toolchain. Samuel
glibc and abi-compliance-checker break multiple KDE autopkgtests
Dear glibc, abi-compliance-checker and Qt/KDE maintainers, With recent upload of glibc and abi-compliance-checker the autopkgtest of libkf5calendarsupport, kf5-kdepim-apps-libs and akonadi-calendar started to fail when run in testing with either glibc or abi-compliance-checker from unstable. These autopkgtests all run abi-compliance-checker and fail on it. Currently these regressions are contributing to the delay of the migration of glibc [1] and abi-compliance-checker [2] to testing. Could you please help to investigate the situation? Normally I would file bugs but because the similarity between the packages that fail for glibc and abi-compliance-checker and the fact that three packages fail I fell back to this e-mail, as it seems to be too much coincidence. Feel free to ask help about the CI side of things from the Debian-CI team (in CC). More information about this e-mail can be found on https://wiki.debian.org/ContinuousIntegration/RegressionEmailInformation Paul [1] https://qa.debian.org/excuses.php?package=glibc [2] https://qa.debian.org/excuses.php?package=abi-compliance-checker signature.asc Description: OpenPGP digital signature