Add Support for Non-Standard Python Import Names

[abhisek]

Multiple Python dependencies such as pyyaml, scikit-learn etc. follows a non-standard package to import name mapping. For example, pyyaml package is imported as yaml. We need to maintain a map of Well Known packages that exhibit this behaviour.

This problem may be a client concern i.e. the code that uses our framework, having it supported in the framework helps in having community maintained mappings. This makes it easy for any client tools to easily leverage this mapping

Ref: https://www.tweag.io/blog/2023-09-21-fawltydeps-mapping-strategy/

[abhisek]

@OmkarPh Can you do an analysis on whats the best way of tackling this?

[OmkarPh]

Yep, I'm looking for any already existing mapping source or how others tackle this

[OmkarPh]

Some packages eg. pyusb, pillow have a optional top_level.txt file (generated by older 'egg' format)
eg. path - venv/lib/python3.12/site-packages/PyYAML-6.0.2.dist-info/top_level.txt

It includes list of top level packages which are exported by this package.

You can install in some venv pip install pyusb pillow pyan3 pyserial and
run this script to test - https://gist.github.com/kaos/1481cef9a6742654274c32b11c8e55ac

The number of packages having this is less, but we can still have it as an additional check

[OmkarPh]

To facilitate this, we can have a community supported JSON file that anyone can contribute to. It'll be a simple json, with property key as package name and value as list of top level modules available when this package is installed.
sample -

{
  "pillow": ["PIL"],
  "pyan3": ["pyan"],
  "pyserial": ["serial"],
  "pyusb": ["usb"],
  "PyYaml": ["yaml"],
  "setuptools": ["pkg_resources", "setuptools"],
  "scikit-learn": ["sklearn"],
  "opencv-python": ["cv2"]
}