#1611 Continued lack of support for RPM weak dependencies in distribution tooling
Closed None Opened 2 years ago by fweimer.

= phenomenon =

ARM composes do not work because something does not implement RPM weak dependencies properly.

= background analysis =

The cause is likely this yum bug: https://bugzilla.redhat.com/show_bug.cgi?id=1360781

Unfortunately, the developer who commented on the libcrypt glibc bug refuses to provide more context:


= implementation recommendation =

We need to a decision whether we want to ban RPM weak dependencies indefinitely because some parts of the Fedora infrastructure keep using obsolete tools.

I have been waiting for complete weak dependency support since we have implemented Langpacks installation using RPM weak dependencies as Change in Fedora 24. When this Change was implemented we tested only in Fedora and it worked but forgot that bodhi masher can be problem. So, if you check Fedora 24 GA repository, it should work for packages using weak dependencies but if you try to pull f24-updates for them which are composed by bodhi on rhel7 which uses createrepo then it fails to generate weak rpm tags. Thus, inconsistency in updating Fedora 24 system.

There are few langpacks related bugs open for this issue against dnf, anaconda/glibc etc.

Please don't ban RPM weak dependencies otherwise whatever langpacks are working for a GA packages that will also not work in Fedora 25.

Replying to [comment:3 ausil]:

https://lists.fedoraproject.org/archives/list/devel-announce@lists.fedoraproject.org/thread/Q5LMMPVEORM76IOPGKYS4XJ6VZ2WLAAX/?sort=date is where fesco previously made a decision on the subject

I don't think this related. The trigger is a plain Recommends: libcrypt-nss%{_isa}, not something involving and/or/if.

Completely banning weak dependencies (I am not talking about rich boolean dependencies here) would be a major showstopper as they were already used in multiple packages to solve otherwise very hardly solvable problems. So, please, FESCo, do not do that.

The issue has been worked around in the ARM image generation:


The ARM compose is now blocked by another issue.

So, on one hand we don't want to ban weak deps or tell maintainers they cannot use them, but on the other hand we can't wave a magic wand and re-write all releng tools to fully support them. So, for the mean time:

agreed: releng will try and move all compose tools to support weak deps, but in the mean time issues with composes and weak deps will need to be handled on a case by case basis.

Login to comment on this ticket.