Python bindings for query.filter_<>() behaves strangely

[j-mracek]

Following code provides incorrect results - empty query

        for require in pkg.get_requires():
            query1 = libdnf.rpm.PackageQuery(query)
            query1.filter_provides([str(require)])

        for require in pkg.get_requires():
            query1 = libdnf.rpm.PackageQuery(query)
            query1.filter_provides(str(require))

            query1 = libdnf.rpm.PackageQuery(query)
            query1.filter_provides("dnf")

[j-mracek]

First problem
str(require) does return class instance description and not string representing reldep. The code requires to call require.to_string() to get required result.

[gotmax23]

Right, libdnf5.rpm.Reldep does not implement __str__(), so calling str() on it just returns the repr.

For the first two examples, you can call query.filter_provides(require) or just pass the ReldepList returned by pkg.get_requires() directly to .filter_provides() without looping. filter_provides() accepts Reldep and ReldepList objects and lists of strings.

For the third one, query1.filter_provides(["dnf"]) should work. The old hawkey.Query.filter(...) API used to accept plain strings in addition to lists of strings, but libdnf5 apparently does not.

I guess the two issues here are implementing the str protocol for Reldep objects and potentially supporting regular strings in addition to ists of strings for these methods.

[j-mracek]

I create the first PR that resolves the issue partially - #1092.

Remaining problem - passing string and not list of strings results in strange behavior (query1.filter_provides("dnf")).

[j-mracek]

It looks like that when string is passed query1.filter_provides("dnf") it is automatically converted to std::vector{"d", "n", "f"}. The problem is that that this incorrect conversion is not visible and hard to debug, because it returns some results time to time.

[j-mracek]

I've created two draft PR that represents alternative solutions.

The first adds an additional overload method to hpp. This approach is potentially problematic, but nicer.

The second adds the method only for Python binding.

To fix the issue completely it would require modifying a lot of methods, therefore I would like to know preferable way before implementation.

May I ask you for your opinion?

[m-blaha]

I personally prefer the additional overload. This makes the API the same across all languages.

[inknos]

https://github.com/rpm-software-management/dnf5/pull/1101adds an additional overload method to hpp. This approach is potentially problematic, but nicer.

#1101 adds the method only for Python binding.

@j-mracek please, check the links, they both go to 1101, but I believe you wanted to go to 1102

I also prefer the solution with the overload.

[j-mracek]

The issue is fixed by #1216