#6 look at how multilib should be done in pungi4 world
Opened 7 years ago by kparal. Modified 6 years ago

In depcheck/rpmdeplint, we do multilib by mashing. But I have a suspicion pungi4 now does multilib on its own, and not through mash. We need to use the same config as fedora releng. Investigate whether mash is still the proper tool to use or we should use pungi (configs) for that somehow.


According to https://pagure.io/mash/issue/4#comment-491920 Fedora Infra no longer uses mash. We should figure out how to use pungi for the same task.

Metadata Update from @kparal:
- Issue priority set to: High (was: Normal)

6 years ago

We're starting to produce invalid results. See #9.

any guesses on how long this'd take to change/fix?

Honestly no idea. The quickest way to find out is to probably talk to pungi4 maintainers and ask whether it's possible to use pungi for this very particular task (i.e. give pungi a repo or a set of RPMs and let it remove those that are not considered multilib). If pungi can do that without complications, the fix can be very fast. If we need to hack pungi to do that, well...

Do we have a volunteer to go on a little investigation quest? Talk to pungi devs, then try out the suggested approach? @lbrabec?

Lets setup a meeting so that I can explain how multilib works

Here are some links:
This is how pungi does multilib:
https://pagure.io/pungi/blob/master/f/pungi/multilib_dnf.py
https://pagure.io/pungi/blob/master/f/share/multilib
It uses this library:
https://pagure.io/releng/python-multilib
Multilib Fedora configs are either here (releases):
https://pagure.io/pungi-fedora/
or in infra ansible under bodhi roles (updates).

We can either run whole (or part of) pungi on our repos (created from pending packages), or we can use python-multilib library, but that only tells us about packages that are directly multilib, not about packages that are pulled as multilib as dependencies - those we'd need to compute manually.

I've seen more issues reported against rpmdeplint, which are clearly again caused by mash. If no one objects, I'll try to disable multilib checking in rpmdeplint for the moment, until we have a proper pungi4-style solution. It seems better to potentially let some multilib issues slip through rather than produce false negatives for completely correct packages.

+1 as well. it makes sense to me

I have disabled multilib checking in task-rpmdeplint. It has been deployed just to dev atm. I'd like to deploy new libtaskotron (with STI tasks support) to production next week, and then it will be easy to merge this to production as well.

It's now deployed into production. We don't check multilib ATM.

Sigh, disabled multilib (i.e. disabled downloading of i386 packages for x86_64 repos) is causing troubles for packages which have a hard dependency on i386 packages, like wine:
https://lists.fedoraproject.org/archives/list/devel@lists.fedoraproject.org/message/WXGS4SKKHE4AGDDONWIOWIWC3B2HUB4X/

I'll try to figure out whether something can be done quickly for those before we have the pungi fix.

Login to comment on this ticket.

Metadata