Comment 2 for bug 1858479

Revision history for this message
Steve Langasek (vorlon) wrote : Re: [Bug 1858479] Re: rdma-core needs to skip pandoc on i386

On Tue, Jan 07, 2020 at 01:20:21AM -0000, Rafael David Tinoco wrote:
> I understand you want to keep i386 samba binaries around, and getting
> rid of ceph OR rdma-core would be okay for i386 arch, specially now that
> we're phasing out the arch.

> Samba package already has a filter for enabling or not samba ceph
> support PER arch, so it would be very straightforward if dependencies
> are available in archive :o).

samba itself is not in the list of packages that we are committed to
keeping. It is in the seed as a build-dependency of vlc (the full
dependency chain is: samba <- vlc <- efl <- dbus-c++ <- libffado <-
jack-audo-connection-kit <- pulseaudio).

The guiding principle is that we should keep whatever (build-)dependency
trees that are required to support the packages which are mandatory to keep,
as long as these dependencies are reasonable to maintain, while introducing
delta on a minimum number of packages, and without unreasonably diminishing
the functionality of the compatibility library packages.

pandoc is not supportable because it's haskell, so must be cut.

rdma-core is uninteresting on i386 and has a single reverse-dependency
(ceph) so can be cut.

ceph is uninteresting on i386 and has a single reverse-dependency (samba) so
can be cut.

samba is in the seed because of vlc, but it also has gvfs-backends as an
unseeded binary dependency. So dropping samba would require changing two
source packages instead of one. Therefore I recommend keeping samba in and
changing it to not pull in ceph on i386.

> - removing cephfs dependency for samba i386 is straight-forward and better option;
> - it would free ceph to be removed, or not, independently;
> - if ceph is kept for i386, more work would need to be done;

Once samba has been updated to drop the build-dependency on ceph for i386,
ceph will be (more or less) automatically removed.