Packaging Question

Issue に示されている生成済みの共有オブジェクト依存関係と、pip/PyPI のパッケージングに関する問題から始めます。複数の Python バージョンに対応する、サポート対象の Linux パッケージング方式と、Boost.Python をどのように扱うべきかを明らかにします。Issue に具体的なパッケージングの方針があり、その互換性要件が明確になれば完了です。

Hi all,

I am using a library built on Boost.Python as a dependency for some of my other projects. This library (https://github.com/ompl/ompl) requires building from source and is always linked to the system Python. I would like to make it installable via pip and distribute it on PyPi. My biggest question around this is how to handle Boost.Python. When I look at the generated shared object files, they are linked to a Python-version-specific version of Boost. Namely, I see

libboost_python38.so.1.71.0 => /usr/lib/x86_64-linux-gnu/libboost_python38.so.1.71.0 (0x00007f9d6d9b7000)

I'm wondering what is the best practice for distributing a pre-built package like this? Should I statically link to a particular version of Boost.Python? Assuming I only handle Linux support to start, I don't mind asking them to use apt to install Boost. But, I'd like to enable support across multiple Python versions. If, for example, they are running Ubuntu 20.04, which has system Python version of 3.8, libboost-python-dev seems to install the 3.8-specific version. Would I have to include the entire boost source code and build from scratch when they run pip install?

I'd love to learn by example, but I haven't been able to find any PyPi-distributed packages that use Boost.Python. Can anyone point me to one?

Thanks a lot for your help!