Bug 12025

Summary: Mixing systemd and non-systemd builds for different MACHINEs confuses dnf
Product: [Build System, Metadata & Runtime] OE-Core Reporter: Richard Purdie <richard.purdie>
Component: devtools / tool chainAssignee: Richard Purdie <richard.purdie>
Status: RESOLVED FIXED QA Contact:
Severity: normal    
Priority: Medium+ CC: meta.mr.watcher, meta.watcher
Version: 2.2   
Target Milestone: 4.99   
Hardware: x86   
OS: Multiple   
Whiteboard:
OS type for building Yocto: --- Type of Regression: ---
Verified: Documentation change: No (bug/feature does not impact docs)

Description Richard Purdie 2017-09-01 07:29:24 UTC 
If I enable systemd:

DISTRO_FEATURES_append = " systemd"
VIRTUAL-RUNTIME_init_manager = "systemd"
DISTRO_FEATURES_BACKFILL_CONSIDERED = "sysvinit"

$ MACHINE=qemuppc bitbake systemd
$ MACHINE=qemumips bitbake systemd

<remove above systemd enablement>

$ MACHINE=qemux86-64 bitbake core-image-minimal-initramfs

I see the rootfs fail:

ERROR: core-image-minimal-initramfs-1.0-r0 do_rootfs: Could not invoke dnf. Command '/media/build1/poky/build/tmp/work/qemux86_64-poky-linux/core-image-minimal-initramfs/1.0-r0/recipe-sysroot-native/usr/bin/dnf -y -c /media/build1/poky/build/tmp/work/qemux86_64-poky-linux/core-image-minimal-initramfs/1.0-r0/rootfs/etc/dnf/dnf.conf --setopt=reposdir=/media/build1/poky/build/tmp/work/qemux86_64-poky-linux/core-image-minimal-initramfs/1.0-r0/rootfs/etc/yum.repos.d --repofrompath=oe-repo,/media/build1/poky/build/tmp/work/qemux86_64-poky-linux/core-image-minimal-initramfs/1.0-r0/oe-rootfs-repo --installroot=/media/build1/poky/build/tmp/work/qemux86_64-poky-linux/core-image-minimal-initramfs/1.0-r0/rootfs --setopt=logdir=/media/build1/poky/build/tmp/work/qemux86_64-poky-linux/core-image-minimal-initramfs/1.0-r0/temp -x busybox-syslog --nogpgcheck install initramfs-live-boot udev run-postinsts busybox initramfs-live-install initramfs-live-install-efi base-passwd' returned 1:
Added oe-repo repo from /media/build1/poky/build/tmp/work/qemux86_64-poky-linux/core-image-minimal-initramfs/1.0-r0/oe-rootfs-repo
Last metadata expiration check: 0:00:02 ago on Fri 01 Sep 2017 07:17:04 AM UTC.
Error: 
 Problem: conflicting requests
  - package udev-1:234-r0.mips32r2 does not have a compatible architecture
  - package udev-1:234-r0.ppc7400 does not have a compatible architecture

tmp/work/qemux86_64-poky-linux/core-image-minimal-initramfs/1.0-r0/oe-rootfs-repo/rpm$ find -name udev*
./ppc7400/udev-hwdb-234-r0.ppc7400.rpm
./ppc7400/udev-extraconf-1.1-r0.ppc7400.rpm
./ppc7400/udev-extraconf-dbg-1.1-r0.ppc7400.rpm
./ppc7400/udev-234-r0.ppc7400.rpm
./ppc7400/udev-extraconf-dev-1.1-r0.ppc7400.rpm
./mips32r2/udev-extraconf-1.1-r0.mips32r2.rpm
./mips32r2/udev-extraconf-dev-1.1-r0.mips32r2.rpm
./mips32r2/udev-hwdb-234-r0.mips32r2.rpm
./mips32r2/udev-234-r0.mips32r2.rpm
./mips32r2/udev-extraconf-dbg-1.1-r0.mips32r2.rpm
./i586/udev-extraconf-dev-1.1-r0.i586.rpm
./i586/udev-extraconf-dbg-1.1-r0.i586.rpm
./i586/udev-extraconf-1.1-r0.i586.rpm
./core2_64/udev-extraconf-dbg-1.1-r0.core2_64.rpm
./core2_64/udev-extraconf-1.1-r0.core2_64.rpm
./core2_64/udev-extraconf-dev-1.1-r0.core2_64.rpm

$ MACHINE=qemuppc bitbake quilt-native
$ MACHINE=qemumips bitbake quilt-native

[triggers it to clean up unreachable recipes such as systemd which is now not enabled]

I can then successfully build core-image-minimal-initramfs as the problematic rpms were removed.

tmp/work/qemux86_64-poky-linux/core-image-minimal-initramfs/1.0-r0/oe-rootfs-repo/rpm$ find -name udev*
./ppc7400/udev-extraconf-1.1-r0.ppc7400.rpm
./ppc7400/udev-extraconf-dbg-1.1-r0.ppc7400.rpm
./ppc7400/udev-extraconf-dev-1.1-r0.ppc7400.rpm
./mips32r2/udev-extraconf-1.1-r0.mips32r2.rpm
./mips32r2/udev-extraconf-dev-1.1-r0.mips32r2.rpm
./mips32r2/udev-extraconf-dbg-1.1-r0.mips32r2.rpm
./i586/udev-extraconf-dev-1.1-r0.i586.rpm
./i586/udev-extraconf-dbg-1.1-r0.i586.rpm
./i586/udev-extraconf-1.1-r0.i586.rpm
./core2_64/udev-extraconf-dbg-1.1-r0.core2_64.rpm
./core2_64/udev-extraconf-1.1-r0.core2_64.rpm
./core2_64/udev-extraconf-dev-1.1-r0.core2_64.rpm

For some reason its preferring the invalid arch rpms over the one that PROVIDES udev.
Comment 1 Alexander Kanavin 2017-09-29 12:28:26 UTC 
After a lot of digging into dnf, libdnf and libsolv, here's the report I submitted to the upstream:

https://bugzilla.redhat.com/show_bug.cgi?id=1497188

Let's see what the maintainers think.
Comment 2 Alexander Kanavin 2017-11-30 11:42:41 UTC 
I got a comment out of upstream:
"But in this particular case we don't have a decision yet. LIBDNF is now under reconstruction and we don't know if new design changes will go in favor of that bug report."

https://github.com/rpm-software-management/libdnf/pull/312#issuecomment-347641033

So let's see how their refactoring goes.
Comment 3 Alexander Kanavin 2018-01-15 11:09:17 UTC 
As it doesn't seem we'll get a fix from upstream for this soon, and RP has a patch in progress that sidesteps the issue by limiting the set of RPMs visible to dnf only to those actually needed, I guess the bug can be reassigned.
Comment 4 Richard Purdie 2018-07-07 12:57:31 UTC 
The rpm filtering code at least resolves this issue so it doesn't concern us any more, even if there are seemingly problematic issues in libdnf.