QuAC: Quick Attribute-Centric Type Inference for Python

All source code for the QuAC tool proposed in the OOPSLA 2024 paper "QuAC: Quick Attribute-Centric Type Inference for Python." NOTE: This only includes QuAC's implementation, and not the benchmarks, baselines, and data analysis code. To reproduce the results in the paper, please download the reproduction package at https://doi.org/10.5281/zenodo.13367665

Requirements and Assumptions

• IMPORTANT: QuAC attempts to dynamically import the Python files under your Python project as modules! Make sure this is possible, and doing this has no nasty side effects (e.g., by wrapping Python files intended to be directly executed instead of imported under a if __name__ == '__main__' block)!
• You have a Python virtual environment with Python 3.10 or above.
• Your Python project under analysis has all modules written in pure Python (no Cython/C code, no FFIs).
• You should know the absolute path of the directory containing Python modules---from that path, the Python interpreter must be able to import every module within the Python project successfully. This depends from project to project. For example:
  • If you have cloned the NetworkX repository to /tmp/networkx, that directory should be /tmp/networkx.
  • If you have cloned the typing_extensions repository to /tmp/typing_extensions, that directory should be /tmp/typing_extensions/
• Your Python project under analysis should either have no dependencies or have all dependencies already installed under the Python environment QuAC is run.

Instructions

Again, note that QuAC attempts to dynamically import the Python files under your Python project as modules! Make sure this is possible, and doing this has no nasty side effects (e.g., by wrapping Python files intended to be directly executed instead of imported under a if __name__ == '__main__' block)!

• Set up a Python 3.10 or above virtual environment to run QuAC.
• Install QuAC's dependencies, listed in requirements.txt, and any dependencies of your Python project under analysis.
• Run quac/main.py:
  • Provide the absolute path of the directory containing Python modules.
  • Provide a module prefix. This filters out irrelevant modules that shouldn't be analyzed, such as installation files, example files, or test files. For example, given the NetworkX repository, providing the module prefix networkx allows us to analyze files such as networkx/algorithms/clique.py while skipping files such as examples/external/plot_igraph.py.
  • Provide an output JSON file.

For example, to run QuAC on the bm_math module in the example/ directory and save output to quac_output.json:

python \
quac/main.py \
--module-search-path example/ \
--module-prefix 'bm_math' \
--output-file quac_output.json

The generated quac_output.json should resemble the following:

{
    "bm_math": {
        "global": {
            "maximize": {
                "points": [
                    "typing.Sequence[bm_math.Point]"
                ],
                "return": [
                    "bm_math.Point"
                ]
            }
        },
        "Point": {
            "__init__": {
                "i": [],
                "return": []
            },
            "__repr__": {
                "return": [
                    "builtins.str"
                ]
            },
            "normalize": {
                "return": []
            },
            "maximize": {
                "other": [
                    "bm_math.Point"
                ],
                "return": [
                    "bm_math.Point"
                ]
            }
        }
    }
}

The JSON is organized into four levels:

• Top-Level Module Names
  • Global Function Names
    • Parameter Names or "return"
      • Type Predictions
  • Class Names
    • Method Names
      • Parameter Names or "return"
        • Type Predictions