Arch Linux

Description:
The python-rapidfuzz package uses three header-only C++ libraries, namely rapidfuzz-cpp, jarowinkler-cpp and Taskflow. These are vendored in the source distribution obtained from PyPI, but the build will search for system-wide installed libraries and those take priority if they are found. rapidfuzz-cpp (in community), jarowinkler-cpp and cpp-taskflow (in AUR) are the packages that provide these libraries. Currently those packages are not in makedepends so the vendored versions are used. The vendored Taskflow is also an old version (2.7.0) while the latest Taskflow version is 3.4.0.

The build output shows whether the vendored version or system version is used. When the build uses vendored library X, the build prints `Using packaged version of X` and when a system library is used, it prints `Using system supplied version of X`.

Another problem is that the build uses -O3 gcc optimization level because it uses the `Release` cmake build type. As explained in the CMake package guidelines[0] this can be fixed by setting the build type to `None`. The correct option to pass to `python -m build` is `-C--global-option=--build-type=None`, where `-C` is the `--config-setting` option to python-build, `--global-option` is an undocumented feature of python-setuptools `build_meta.py` which passes the following option to the underlying setup.py call, and `--build-type=None` is a python-scikit-build option[1]. For future reference, `--global-option` will be deprecated and renamed to `--build-option` in a future python-setuptools release, see [2]. In order to see the optimization level and other compiler flags, option `-C--global-option=-DCMAKE_VERBOSE_MAKEFILE=ON` for `python -m build` can be used.

Package python-jarowinkler has very similar issues, so I will list them here. It uses jarowinkler-cpp header-only library which is vendored in the source distribution. Adding jarowinkler-cpp (currently in AUR) to makedepends fixes this, as the system-wide installed library will take priority over the vendored one. python-jarowinkler also has the same issue with -O3 optimization as python-rapidfuzz, and the fix is exactly the same.

Another issue with both packages is that they use pre-generated cython sources instead of generating those during package build. However, doing this would require cython version 3.0 which is still in alpha stage.

Additional info:
* package version(s)
- python-rapidfuzz 2.4.2-2
- python-jarowinkler 1.2.0-2
* config and/or log files etc.
* links
[0] https://wiki.archlinux.org/title/CMake_package_guidelines#CMake_can_automatically_override_the_default_compiler_optimization_flag
[1] https://scikit-build.readthedocs.io/en/latest/usage.html#usage-scikit-build-options
[2] https://github.com/pypa/setuptools/pull/3380