#2162 F31 System-Wide Change: No more i686 kernels or bootable images
Closed: Accepted 10 days ago by ignatenkobrain. Opened 21 days ago by bcotton.

Stop building i686 kernels, reduce the i686 package to a kernel-headers package that can be used to build 32bit versions of everything else. No kernel means we would stop producing i686 bootable images as well.


I'm still unclear on if this also means we no longer produce/ship i386 repos? Can you clarify? ie, https://dl.fedoraproject.org/pub/fedora-secondary/releases/30/Everything/i386/os/

In any case, I am +1, but would like to clarify the above.

+1 :champagne:

Also, joining question from @kevin about repos. I think we still need to produce them.

+1 :champagne:
Also, joining question from @kevin about repos. I think we still need to produce them.

Unless we want to convert to cross-compilation to be able to produce multilib libraries, I suppose we don't have much choice regarding producing the i*86 repository.

Obviously +1 here.

As for the repo, I think for F31, they need to exist. I think the next step will be figuring out what is really required for multilib, and making that a thing, possibly for the F32 or F33 timeframe.

+1 :champagne:
Also, joining question from @kevin about repos. I think we still need to produce them.

Unless we want to convert to cross-compilation to be able to produce multilib libraries, I suppose we don't have much choice regarding producing the i*86 repository.

I'm not sure we need to make a Everything i386 repo in order to do multilib. I thought pungi just gathers the needed ones from koji (but I could well be wrong).
@lsedlar can you clarify? Does pungi need to make a i386 repo in order to multilib the x86_64 repo with i386/i686?

+1 :champagne:
Also, joining question from @kevin about repos. I think we still need to produce them.
Unless we want to convert to cross-compilation to be able to produce multilib libraries, I suppose we don't have much choice regarding producing the i*86 repository.

I'm not sure we need to make a Everything i386 repo in order to do multilib. I thought pungi just gathers the needed ones from koji (but I could well be wrong).

I'm not so much concerned about producing the x86_64 repo as I am about being able to compile the i686 multilib content in the first place. Right now, since we produce those builds just as part of building a complete i686 distro, we get the multilib builds "for free". If we stop having that repo, where do we get all the build-deps from?

Let's do what is proposed only, for F31 and later we can see?

I'm not so much concerned about producing the x86_64 repo as I am about being able to compile the i686 multilib content in the first place. Right now, since we produce those builds just as part of building a complete i686 distro, we get the multilib builds "for free". If we stop having that repo, where do we get all the build-deps from?

I am not at all advocating removing the buildroot for i686. That buildroot is created and maintained by koji.

I am saying that we don't need to publish a i386 repo as gathered and collected and synced around by punji.

If we do keep publishing such a repo, we may have people who install a i686 Fedora 30 and upgrade it to Fedora 31 (or 32 or 33 or...) (but still have an old, insecure, unmaintained kernel from f30).

If the compose is configured to not create Everything.i386 repo, it will not have any effect on the x86_64 repo in terms of multilib.
Here's a test compose in stage: I disabled everything apart from Everything.x86_64. There are still multilib packages in there.
https://kojipkgs.stg.fedoraproject.org/compose/lsedlar/rawhide/Fedora-Rawhide-20190709.n.0/compose/Everything/

Metadata Update from @ignatenkobrain:
- Issue tagged with: pending announcement

13 days ago

So, where do we stand on composing/publishing the i386 trees?

Personally I think we should not produce them for f31, but it would be nice to have everyone on board with that... perhaps discuss/vote in the meeting?

I think discussion in the meeting would be a good idea. I am not opposed, but I am somewhat concerned that people might be pulling things from those trees on x86_64 systems for various reasons.

Let's discuss it this friday.

Metadata Update from @ignatenkobrain:
- Issue tagged with: meeting

12 days ago

We have discussed this today on FESCo meeting:

ACTION: nirik to put together a proposal how we can drop i686 (ignatenkobrain, 15:42:02)
AGREED: Once proposal in place, we can consider it as a late exception for F31 (ignatenkobrain, 15:42:53)

@kevin I guess you will open a new ticket and/or Change Proposal, so I'll close this.

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

10 days ago

FYI, the proposed change from me is at: https://fedoraproject.org/wiki/Changes/Noi686Repositories feel free to let me know if there's any further info I can add there...

As a SystemWideChange, it needs the releng ticket. This one especially.

There were some modularity concerns - when I have a i686 Koji modular failure, how do I debug it locally?

As a SystemWideChange, it needs the releng ticket. This one especially.

Sure. I figured since I am heavily involved in releng I would know and talk to other releng folks about it, but perhaps a formal ticket is good.
Filed https://pagure.io/releng/issue/8529 on it.

There were some modularity concerns - when I have a i686 Koji modular failure, how do I debug it locally?

I don't know. How do you debug it now? According to https://docs.fedoraproject.org/en-US/modularity/making-modules/building-modules-locally/ it looks like it assumes you are on the arch you wish to build for. It could be that setarch i686 and a i686 mock config that uses the koji buildroot repo would work on a x86_64 machine, but I don't know if this works currently with the existing i686 repos. I very much doubt everyone has a i686 machine they keep for debugging. I would need input from modularity folks on this.

Additionally, we could stop making i686 modules and avoid this case. I don't think (but don't know for sure) that there is no multilib in modules.

There were some modularity concerns - when I have a i686 Koji modular failure, how do I debug it locally?

I don't know. How do you debug it now? According to https://docs.fedoraproject.org/en-US/modularity/making-modules/building-modules-locally/ it looks like it assumes you are on the arch you wish to build for. It could be that setarch i686 and a i686 mock config that uses the koji buildroot repo would work on a x86_64 machine, but I don't know if this works currently with the existing i686 repos. I very much doubt everyone has a i686 machine they keep for debugging. I would need input from modularity folks on this.
Additionally, we could stop making i686 modules and avoid this case. I don't think (but don't know for sure) that there is no multilib in modules.

This is a really good question and one we should have an answer for handy. I'm going to ping @mprahl and @jkaluza to comment here.

Login to comment on this ticket.

Metadata