This is needed for porting over dnfdragora from dnfdaemon.

add dnf5.conf-todo.5.rst page which documents differences between dnf4
add dnf5.conf-deprecated.5.rst page which documents deprecated options
and dnf5 config options

zchunk is not supported in RHEL, and other related packages also disable
their zchunk dependency on RHEL.

rpminspect complains on a dnf5 build:

    Subpackage dnf5-plugins on x86_64 carries 'Requires:
    libdnf5-cli.so.1()(64bit)' which comes from subpackage libdnf5-cli
    but does not carry an explicit package version requirement. Please
    add 'Requires: libdnf5-cli = %{version}-%{release}' to the spec
    file to avoid the need to test interoperability between various
    combinations of old and new subpackages.

This patch adds the dependency into dnf5.spec.
The very same dependency already exists in dnf5daemon-client and
libdnf5-cli bindings.

It prevents additional removal after succesful remove using the same spec.
Spec `foo`. The original code during the first run removes (dnf5 remove foo)
all packages with the name `foo`. In the next run (dnf5 remove foo) it
removes all packages providing `foo`. The next run will remove packages
providing binary `foo`. This behavior is inconsistent.

Resolves: https://issues.redhat.com/browse/RHEL-5747

It unify style with remove command, because it uses the same argument type.

Some files are present in primary.xml so it works for them but if we
want to filter by files present only in filelist.xml we need to load
them.

This option replaces `--resolve` option. The `--resolve` option was
using the existing formatting options (`--requires`, `--provides`,...)
to determine for which package attribute it should find providers. This
is confusing because it completely changes the meaning of those options
and their original functionality is no longer accessible.

Instead add a new `--providers-of=` option which allows specifying the
package attribute directly.

Commit a703658 (dnfdaemon: Fix
download callbacks) renamed repo_load_end signal to download_end, but
forgot to change the name in org.rpm.dnf.v0.Base interface
introspection file.

Fixes: rpm-software-management#1146

Fixes error:
repoPYTHON_wrap.cxx:5656:56: error: expected template-name before '<' token
 5656 |     struct SwigPyMapIterator_T : SwigPyIteratorClosed_T<OutIterator, ValueType, FromOper>

due to missing SWIG %fragment dependency on SwigPyIterator_T which provides the missing base class SwigPyIteratorClosed_T.

See swig/swig#2744

Like `logdir`, the `cachedir` and `system_cachedir` configuration
options should be treated as relative to the installroot, unless
specified is specified on the command line, e.g. via
--setopt=cachedir=/path/to/cachedir. This is the behavior of the
cachedir with DNF 4.

Currently, the host cache directory is used by default, conflicting with
what our docs say (man 7 dnf5-installroot):
    "cachedir, log files, releasever, and gpgkey are taken from or stored
    in the installroot."
After this change, the default system_cachedir is
$installroot/var/cache/libdnf5, which is consistent with the docs.

For https://bugzilla.redhat.com/show_bug.cgi?id=2256945

An empty string has a special meaning in gettext and marking such
string produces invalid catalogs with a duplicate message ID entry.

We cannot use simply Error("") necause Error() expects
BgettextMessage argument. Therefore I added EMPTY_MESSAGE macro which
expands to the same as M_(""), but is not caught by xgettext.

I think it's cleaner solution than overriding xgettext behaviour for
that line of code. For aesthetic reasons I also disabled clang-format
at two places.

Fixes: rpm-software-management#1158

The duplicates breaks msgmerge on the broken catalogs in CI.
This patch fixes them.

Fixes: rpm-software-management#1158

Allow passing `--allowerasing` option in group subcommands.

It is required for str() Python method

This is called when str([reldep])

It allows to use it in Python set()

Those methods are important for set(), str(), and print([package]).

The current implementation of the filter_version and filter_release
filters always uses the cmp_eq comparator. This means that these filters
only work with EQ and NEQ comparisons. The patch fixes these filters to
work correctly with LT, LTE, GT, and GTE comparisons as well.

Now the tests use also LT, LTE, GT, and GTE comparisons.

…x-issue-758-759