We are wondering if there was a way to continue to use the automatic setup.py feature in the python_distribution, but still add runtime libraries to the Requires field of a package. There are several ML libraries that require models to be present, but are not code dependencies, as you can choose what model you want (english, italian, small, etc.). As such it would be nice for our code to add them as dependencies to the wheel we are building as we have a limited set of the models we want.

using a plugin, you can customize the arguments passed to

setup()

https://www.pantsbuild.org/docs/plugins-setup-py

This area of the codebase is due for an overhaul. There are tickets for this, but I want to get it to a state where you can easily inject extra setup args alongside the generated ones, without writing a plugin

yeah, going through it right now to write a plugin. Not too bad, but for consumers of the build system who would would be release engineers, this is quite a bit of work.

Indeed. That is one of the older areas of the codebase, and could use some love.