On Tue, Mar 1, 2022 at 3:07 AM Richard W.M. Jones <rjones(a)redhat.com&gt; wrote: It's not crazy. EPEL is intended to build on RHEL content, which means we can't depend on something RHEL doesn't publish. If Red Hat wants to publish their buildroot repo, then sure, we could use it. Just because it happens to exist in the CentOS Stream 9 buildroot content does not mean we would be able to rely on it once we replace CentOS Stream with RHEL for EPEL 9. Thus, we don't use the CentOS Stream 9 buildroot either. If we did, we'd wind up in a situation where packages were built once and then not buildable ever again. That already kind of happened when we initially had that buildroot repo in the EPEL build environment and it made it way harder for us to figure out what gaps we had for things to build against RHEL later. We've fortunately dealt with the small number of cases that occurred from then. -- 真実はいつも一つ！/ Always, there's only one truth!

On Tue, Mar 01, 2022 at 01:19:32PM +0000, Richard W.M. Jones wrote: But to add - ideally we wouldn't need to do this if Red Hat published the buildroot packages, and I don't understand why we don't do this. Rich. -- Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones Read my programming and virtualization blog: http://rwmj.wordpress.com libguestfs lets you edit virtual machines. Supports shell scripting, bindings from many languages. http://libguestfs.org

On Tue, 1 Mar 2022 at 08:19, Richard W.M. Jones <rjones(a)redhat.com&gt; wrote: I do not think you will find much disagreement here.. but after 3+ years of saying it and nothing changing, many of us have made our peace. The issue in the past has been that it takes manual matching at times to make it work. Koji, mock and rpmbuild will all complain in different ways when package content varies in minute ways. Someone then has to rebuild that package when that happens, which usually only is found at 2am by some very cranky engineer who posts a lot of less than polite messages about how EPEL people are complete crap. Or you find that the internal package is good enough to have built the RHEL content but still is lacking something that you expected to be there for anything NOT a RHEL content. Again that takes inspection and someone to care enough to do it. If we need that, then we might as well rebuild the content and make sure it is what we wanted in the first place. The way Fedora build system deals with RHEL packages is a bit of 'hack' compared to how it builds Fedora packages. It sees them as external packages and (mis)-uses a method which was originally only for bootstrapping a distro to see packages it did not build itself. This then requires additional hacks on top of that to keep the facade working.. those hacks break regularly and have to be manually dealt with. Usually the breakage then requires some 'we can break the koji database again so do a full backup and possibly a restore' actions by Fedora release engineering to delete some entry koji 'thinks' should be there but isn't (or vice versa). [I believe this happened at least twice when we tried mixing stream and non-stream.] -- Stephen J Smoogen. Let us be kind to one another, for most of us are fighting a hard battle. -- Ian MacClaren

To keep this a bit more specific, we're trying to build coccinelle for EPEL 9. This requires ocaml [compiler] and a bunch of ocaml packages. They are mainly in RHEL buildroot. The problem we're going to have (which to be fair is a problem somewhat specific to OCaml linking) is that an alternate OCaml compiler built for EPEL will have different hash values[1] for core libraries. If we use a different version from the RHEL compiler, say we use Fedora Rawhide version, then all the hashes will be different. If we use the same version as in RHEL, then most hash values will be the same, but if the compiler flags are even slightly different then there will be some differences. OCaml libraries in RHEL that we do ship (ocaml-libnbd and others) will not be linkable with code compiled with the EPEL toolchain. It's entirely possible we don't care about this, and I maybe even agree. But also that people using RHEL with EPEL added will get into weird situations if they try to recompile the virt tools packages. Rich. [1] Hash values are computed over the modules to prevent linking incompatible versions: $ ocamlobjinfo /usr/lib64/ocaml/unix.cma | grep Unix Unit name: Unix 49c6c492a189deeaed5bf77a6793e7fa Unix and turned into RPM dependencies: $ rpm -qR ocaml | grep Unix ocaml(Unix) = 49c6c492a189deeaed5bf77a6793e7fa -- Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones Read my programming and virtualization blog: http://rwmj.wordpress.com virt-builder quickly builds VMs from scratch http://libguestfs.org/virt-builder.1.html

On Wed, Mar 02, 2022 at 09:04:04AM -0500, Neal Gompa wrote: It's a good question. I think we chose not to do that simply because we were worried about handling CVEs in a timely way and general support (I'm not sure if CRB is officially supported or not, but there may be some "implicit" support if we're shipping stuff). I would not be opposed to it though. Rich. -- Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones Read my programming and virtualization blog: http://rwmj.wordpress.com virt-top is 'top' for virtual machines. Tiny program with many powerful monitoring features, net stats, disk stats, logging, etc. http://people.redhat.com/~rjones/virt-top

On Thu, Mar 03, 2022 at 12:42:03PM -0500, Josh Boyer wrote: I see "not supported" and "no ABI compatibility guarantee" in bold there which is what we want. I'll see if we can move the OCaml packages to CRB. It seems to be the easiest way to fix the original coccinelle build problem. Rich. -- Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones Read my programming and virtualization blog: http://rwmj.wordpress.com virt-top is 'top' for virtual machines. Tiny program with many powerful monitoring features, net stats, disk stats, logging, etc. http://people.redhat.com/~rjones/virt-top

On Fri, Mar 4, 2022 at 6:01 AM Richard W.M. Jones <rjones(a)redhat.com&gt; wrote: If you already had a ACKed BZ/Jira for these, you can probably reference that and make the necessary merge requests to comps[1] and pungi[2]. [1]: https://gitlab.com/redhat/centos-stream/release-engineering/comps [2]: https://gitlab.com/redhat/centos-stream/release-engineering/pungi-centos -- 真実はいつも一つ！/ Always, there's only one truth!

On Fri, Mar 04, 2022 at 11:14:31AM +0000, Richard W.M. Jones wrote: Currently even with this bug fixed we won't be able to build Coccinelle until RHEL 9.1 is released, which is like 9+ months away. This (if true) is ridiculous. Is there some other solution? Rich. -- Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones Read my programming and virtualization blog: http://rwmj.wordpress.com libguestfs lets you edit virtual machines. Supports shell scripting, bindings from many languages. http://libguestfs.org

On 11. 03. 22 10:36, Richard W.M. Jones wrote: Given that EPEL 9 now builds against CentOS Stream, this is not necessarily true. The other solution would have been to include the packages in CRB sooner :D -- Miro Hrončok -- Phone: +420777974800 IRC: mhroncok

On Fri, Mar 11, 2022 at 09:36:08AM +0000, Richard W.M. Jones wrote: I have now rebuilt all the packages and they're in a RHEL 9.1 erratum (RHBA-2022:89734-01 for those who have access). Is there anything we can do to get these packages into the EPEL buildroot so we don't have to wait forever to build coccinelle for EPEL? Rich. -- Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones Read my programming and virtualization blog: http://rwmj.wordpress.com libguestfs lets you edit virtual machines. Supports shell scripting, bindings from many languages. http://libguestfs.org