#4084 mulilibbed packages could be inconsistet in the updates and updates-testing repos compared to the Release tree repo
Closed: Grooming 5 years ago Opened 12 years ago by mschwendt.

There is a discrepancy between what is pushed to stable dist repos and to updates-testing. With the latest "gcc" test-update, F-14 is affected, too (which means this is not specific to F-13).

gcc-gfortran.i686 is available in x86_64 development/14 but not x86_64 updates/testing/14. The version-specific dependency in it is unresolvable deps if it isn't updated.

What is the cause of it? Can it be fixed in the compose tools? Or is a self-"Obsoletes" (which is not arch-specific by default) in gcc-gfortran the only cure?

[ ] gcc-4.5.1-1.fc14.x86_64.rpm 13-Aug-2010 14:51 14M
[ ] gcc-c++-4.5.1-1.fc14.x86_64.rpm 13-Aug-2010 14:50 5.1M
[X] gcc-gfortran-4.5.1-1.fc14.i686.rpm 13-Aug-2010 14:56 4.4M
[ ] gcc-gfortran-4.5.1-1.fc14.x86_64.rpm 13-Aug-2010 14:54 5.0M
[ ] gcc-gnat-4.5.1-1.fc14.x86_64.rpm 13-Aug-2010 14:52 9.9M
[ ] gcc-java-4.5.1-1.fc14.x86_64.rpm 13-Aug-2010 14:56 4.0M
[ ] gcc-objc++-4.5.1-1.fc14.x86_64.rpm 13-Aug-2010 14:56 4.3M
[ ] gcc-objc-4.5.1-1.fc14.x86_64.rpm 13-Aug-2010 14:57 3.9M
[ ] libgcc-4.5.1-1.fc14.i686.rpm 13-Aug-2010 14:53 61K
[ ] libgcc-4.5.1-1.fc14.x86_64.rpm 13-Aug-2010 14:51 52K
[ ] libgcj-4.5.1-1.fc14.i686.rpm 13-Aug-2010 14:54 17M
[ ] libgcj-4.5.1-1.fc14.x86_64.rpm 13-Aug-2010 14:55 19M
[ ] libgcj-devel-4.5.1-1.fc14.i686.rpm 13-Aug-2010 14:50 1.5M
[ ] libgcj-devel-4.5.1-1.fc14.x86_64.rpm 13-Aug-2010 14:51 1.5M
[ ] libgcj-src-4.5.1-1.fc14.x86_64.rpm 13-Aug-2010 14:51 12M

[ ] gcc-4.5.1-3.fc14.x86_64.rpm 07-Sep-2010 12:36 14M
[ ] gcc-c++-4.5.1-3.fc14.x86_64.rpm 07-Sep-2010 12:39 5.1M
[ ] gcc-gfortran-4.5.1-3.fc14.x86_64.rpm 07-Sep-2010 12:42 5.0M
[ ] gcc-gnat-4.5.1-3.fc14.x86_64.rpm 07-Sep-2010 12:31 9.9M
[ ] gcc-java-4.5.1-3.fc14.x86_64.rpm 07-Sep-2010 12:29 4.0M
[ ] gcc-objc++-4.5.1-3.fc14.x86_64.rpm 07-Sep-2010 12:35 4.3M
[ ] gcc-objc-4.5.1-3.fc14.x86_64.rpm 07-Sep-2010 12:34 3.9M
[ ] libgcc-4.5.1-3.fc14.i686.rpm 07-Sep-2010 12:33 62K
[ ] libgcc-4.5.1-3.fc14.x86_64.rpm 07-Sep-2010 12:33 52K
[ ] libgcj-4.5.1-3.fc14.i686.rpm 07-Sep-2010 12:29 17M
[ ] libgcj-4.5.1-3.fc14.x86_64.rpm 07-Sep-2010 12:35 19M
[ ] libgcj-devel-4.5.1-3.fc14.i686.rpm 07-Sep-2010 12:29 1.5M
[ ] libgcj-devel-4.5.1-3.fc14.x86_64.rpm 07-Sep-2010 12:37 1.5M
[ ] libgcj-src-4.5.1-3.fc14.x86_64.rpm 07-Sep-2010 12:41 12M

None of the gcc-* subpackages should be ever multilibbed, they should be shipped for the primary arch only (and similarly for libgcj-src). The rest ought to be multilibbed.
So updates gets it right.

repoquery -qR netcdf-devel-0:4.1.1-3.fc14.i686


Since this is in the 'main' repo, gcc-gfortran is included as multilib there. In the updates repo, this package (and others with the same requirement) don't exist, so it's not pulled in there.

Then netcdf is broken, it shouldn't require 32-bit gcc-gfortran, when 64-bit gcc-gfortran can do the same job.


repoquery -q --whatrequires "gcc-gfortran(x86-32)"


That would be what needs fixed. And if we do that, we will need some sort of obsoletes hack, given that it's been pulled in as multilib in the wild for these in the past.

The push process is broken, too, however. Not specific to gcc, ... just as an example, any future gcc update will cause the same broken dep. gcc-gfortan.i686 will remain forever in the x86_64 release repo with no update replacing it in "updates" repo. This can happen with any other multiarch package dependencies, can't it?

Yes. (It's not a trivial fix.)

  • Once Bodhi uses koji signed repos, we should revisit this as a change for signed repos

Not a trivial fix even with signed repos, also not high priority. (Would need to be addressed in Mash, Koji, Bodhi today). If we move everything to dist-repos in the future, we would only have to fix it there and teach pungi and bodhi how to use it.

Metadata Update from @ausil:
- Issue close_status updated to: Grooming
- Issue status updated to: Closed (was: Open)

5 years ago

Login to comment on this ticket.