Just strip out non-UTF characters from %changelog, not the whole %changelog

[praiskup]

Describe the bug
The current behavior of Pulp appears to be that when a %changelog contains non-UTF-8 symbols, the entire %changelog section is stripped from the repository metadata. We should improve this by providing the otherwise valid %changelog and simply dropping the invalid non-UTF-8 characters.

To Reproduce
Place invalid symbols into the %changelog in your spec file and build it against an older target (newer targets have started stripping these symbols at the rpmbuild level).

Expected behavior
Even if the %changelog contains non-UTF-8 data, the valid portions should still be present in the RPM metadata, even if the section is technically incomplete.

Additional context
See also rpm-software-management/createrepo_c#327

[dralley]

@praiskup So, if we were to upload this package (python-pyABF-2.3.6-2.fc38.rpm) and add it to a repository, and publish, and look at the metadata, we should see missing changelog entries? Would it be only the specific changelog entries that contained the mistake, or just one string would be corrupted, or the entire changelog section (including valid changelogs) would be disincluded?

We should check whether this happens at a database level (e.g. if we still have the changelog entries in the DB but they aren't published) or if it happens before the database level (e.g. if the database is also missing those entries)

[praiskup]

or the entire changelog section (including valid changelogs) would be disincluded?

This. I've seen no single changelog entry for broken packages.

[dralley]

But it would still write a section for the package, right? Just an empty one. Like:

<package pkgid="bba430bebeec583e817372badbc957259d210e3081f5913a831594eb48f48f64" name="apr-util-bdb" arch="x86_64">
  <version epoch="0" ver="1.6.1" rel="6.el8"/>
</package>

where it ought to write a few ` entries there?

[praiskup]

See also this comment: fedora-copr/copr#3868 (comment)

I am still getting this output:

Changelog for mock-lvm-6.3.test.test-1.20251001100930995191.praiskup.issue.3868.13.g126ef34a.el8.noarch