Do not package /usr/share/locale

[mvidner]

Problem

Factory submission complained:

found conflict of agama-19.pre+1110.2ad621a9f-1.1.x86_64 with filesystem-84.87-22.1.x86_64 /usr/share/locale [mode mismatch: d755 root:root, d555 root:root]

Solution

1. Omit it from the spec file.

Checked that other packages also do not declare /usr/share/locale directory, leaving it to filesystem.rpm. This comes out empty:

rpm -qlv $(rpm -qf /usr/share/locale/cs/LC_MESSAGES/*) | grep /usr/share/locale$

2. In fact, follow the Packaging Conventions and use %find_lang, thanks to @DimStar77 for the patch

Testing

Manual build to make sure the RPM does not contain /usr/share/locale

1. Edit _service to say <param name="revision">locale-packaging-conflict</param>
2. osc service manualrun
3. osc build
4. rpm -qlvp /var/tmp/build-root/openSUSE_Tumbleweed-x86_64/home/abuild/rpmbuild/RPMS/x86_64/agama-19.pre+*.rpm

Screenshots

No

Documentation

• .changes

[ancorgs]

LGTM

[mvidner]

Well, the build fails with

[ 361s] Verifying packages...
[ 361s] Preparing packages...
[ 361s] agama-common-19.pre+1154.10ebf379b-0.x86_64
[ 361s] agama-cli-19.pre+1154.10ebf379b-0.x86_64
[ 361s] agama-cli-bash-completion-19.pre+1154.10ebf379b-0.noarch
[ 361s] agama-cli-fish-completion-19.pre+1154.10ebf379b-0.noarch
[ 361s] agama-cli-zsh-completion-19.pre+1154.10ebf379b-0.noarch
[ 361s] agama-19.pre+1154.10ebf379b-0.x86_64
[ 361s] agama-scripts-19.pre+1154.10ebf379b-0.x86_64
[ 362s] agama-openapi-19.pre+1154.10ebf379b-0.x86_64
[ 362s] agama-autoinstall-19.pre+1154.10ebf379b-0.x86_64
[ 362s] ... running 50-check-binary-kernel-log
[ 362s] ... running 50-check-core-files
[ 362s] ... running 50-check-debuginfo
[ 362s] ... testing for empty debuginfo packages
[ 362s] ... running 50-check-filelist
[ 362s] ... checking filelist
[ 362s] agama-19.pre+1154.10ebf379b-0.x86_64.rpm: directories not owned by a package:
[ 362s] - /usr/share/locale
[ 362s]
[ 362s] kaonashi failed "build agama.spec" at Fri Jan 23 14:28:37 UTC 2026.

I will get back to this in a couple of hours

[DimStar77]

@mvidner Did you run your local build on a clean vm/chroot (e.g osc build --clean) ? If you ran this against an already polluted chroot, then any prior test install of agama.rpm inside the VM (which happens at the end of the build) would have 'unowned' /usr/share/locale from the filesystem package, as the mode did not match (755 vs 555) - reusing this chroot without cleaning it up keeps the wrong permission on /usr/share/locale in place, leading to the directory marked as not being owned by anything

Other than that, I'd suggest going with our default packaging way for locales by applying this patch to the spec file

Index: agama.spec
===================================================================
--- agama.spec  (revision 40d5394d588f3dd17c25132e464405c4)
+++ agama.spec  (working copy)
@@ -194,6 +194,7 @@
   libexecdir=%{_libexecdir} \
   mandir=%{_mandir} \
   %{_builddir}/agama/install.sh
+%find_lang agama
 
 %check
 PATH=$PWD/share/bin:$PATH
@@ -240,15 +241,13 @@
 %postun -n agama-scripts
 %service_del_postun_with_restart agama-scripts.service
 
-%files
+%files -f agama.lang
 %doc README.md
 %license LICENSE
 %{_bindir}/agama-web-server
 %{_pam_vendordir}/agama
 %{_unitdir}/agama-web-server.service
 %dir %{_datadir}/agama/eula
-%dir %{_datadir}/locale
-%{_datadir}/locale/*/LC_MESSAGES/agama.mo
 
 %files -n agama-common
 %dir %{_datadir}/agama/jsonnet

[mvidner]

[ 362s] ... running 50-check-filelist
[ 362s] ... checking filelist
[ 362s] agama-19.pre+1154.10ebf379b-0.x86_64.rpm: directories not owned by a package:
[ 362s] - /usr/share/locale

Ah, it was indeed a caching problem, the build succeeded after removing the build tree beforehand.

Thanks for the packaging fix, I should refresh my knowledge of the Packaging Conventions!

[imobachgs]

I am not sure how this works.

[DimStar77]

%files section, reading from the file agama.lang (-f agama.lang); the file is generated during %install by the %find_lang agama call

[imobachgs]

Thanks! And what happens with the original section? I mean, the lines below that one.

[DimStar77]

They stay fully valid.

the agama.lang looks something like (taken from gawk, as that's what I currently have on disk):

> cat gawk.lang 
%defattr (644, root, root, 755)
%lang(bg) /usr/share/locale/bg/LC_MESSAGES/gawk.mo
%lang(ca) /usr/share/locale/ca/LC_MESSAGES/gawk.mo
%lang(da) /usr/share/locale/da/LC_MESSAGES/gawk.mo
%lang(de) /usr/share/locale/de/LC_MESSAGES/gawk.mo
%lang(es) /usr/share/locale/es/LC_MESSAGES/gawk.mo
…

the lines below %files -f agama.lang will be appended after the content of the generated .lang file

(as you can see, the find_lang macro actually goes the more correct way of tagging the translations with the correct/relevant %lang prefix)

[coveralls]

coverage: 73.997%. remained the same
when pulling 022d661 on locale-packaging-conflict
into e5b3354 on master.