Description of problem:
I am attempting to build QEMU in rawhide and mock fails to install systemtap-devel because it can't find any package providing kernel-devel-uname-r
DEBUG util.py:439: Error:
DEBUG util.py:439: Problem: package systemtap-3.3-1.fc29.i686 requires systemtap-devel = 3.3-1.fc29, but none of the providers can be installed
DEBUG util.py:439: - conflicting requests
DEBUG util.py:439: - nothing provides kernel-devel-uname-r needed by systemtap-devel-3.3-1.fc29.i686
This appears to be specific to the i686 build - other roots successfully installed.
Looking at koji packages, the kenrel-devel RPM has indeed gone missing on i686
It was last present in
Version-Release number of selected component (if applicable):
Yes, there is a known issue with building the kernel for i686 so it's been disabled. This has been reported to the x86 SIG for follow up https://firstname.lastname@example.org/thread/YCBS5Y33YFBN2NPNPRUH6YJQFB5CVQ4F/
Sigh, this is a really unhelpful approach to deal with the problem. It would be much nicer to not have rebased the kernel to newer version if it was known to break on one of the archs. Disabling it on one arch isn't a fix, because this has a ripple effect across the rest of the distro in any places which have dependencies on the kernel. Now we're unable to build various other like systemtap, qemu, libvirt, packages, and the only way to unblock that is to rollback rawhide to the earlier kernel, or modify countless other packages to disable i686.
Agreed this isn't the best option. i686 has been a secondary concern for a while now and it is now an official alternative architecture. This means that issues with it need to be driven by the x86 SIG.
Created attachment 1453336 [details]
.config for kernel build
I successfully built a 4.18-rc1+ kernel from Linus' tree (81e97f01371f4e1701feeafe484665112cd9ddc2). I didn't use mock or qemu, just make on an i686 laptop. I'll look at building from https://git.kernel.org/pub/scm/linux/kernel/git/jwboyer/fedora.git/log/?h=rawhide next, to see if Fedora specific patches are causing a problem.
(In reply to Laura Abbott from comment #1)
> Yes, there is a known issue with building the kernel for i686
TWIMC: This is afaics something that happens sporadically. I have seen the problem myself a few hours before Laura disabled the i686 build; I reenabled them for my vanilla builds (
https://fedoraproject.org/wiki/Kernel_Vanilla_Repositories ) and 75 percent (rough estimate) of the i686 builds (done as scratch builds in koji) compile just fine. Makes me wonder if it's something with GCC or the builders.
I was able to install a kernel built from the Fedora git repo. I can't test booting remotely, but that isn't really needed here.
I can't quickly test kernel builds for intermittent issues.
FYI a patch was posted for a loop termination flaw in Kconfig which actually affects all archs, so please see if this is sufficient to get i686 enabled again:
Message from Jerry James here :
I pulled in the fix and i686 is building again. Closing this ticket.