LibSA4Py is a static analysis library for Python, which extracts type hints and features for training ML-based type inference models.

• Requirements
• Quick Installation
• Usage
  • Processing projects
  • Merging projects
• JSON Output

• Python 3.7 or newer (Python 3.8 is recommended)
• Watchman (for running pyre) [Optional]
• MacOS or Linux systems

git clone https://github.com/saltudelft/libsa4py.git
cd libsa4py && pip install .

Given Python repositories, run the following command to process source code files and generate JSON-formatted outputs:

libsa4py process --p $REPOS_PATH --o $OUTPUT_PATH --d $DUPLICATE_PATH --j $WORKERS_COUNT --l $LIMIT --c --no-nlp --pyre

Description:

• --p $REPOS_PATH: The path to the Python corpus or dataset.
• --o $OUTPUT_PATH: Path to store processed projects.
• --d $DUPLICATE_PATH: Path to duplicate files of the given dataset (i.e. jsonl.gz file produced by the CD4Py tool). [Optional]
• --s: Path to the CSV file for splitting the given dataset. [Optional]
• --j $WORKERS_COUNT: Number of workers for processing projects. [Optional, default=no. of available CPU cores]
• --l $LIMIT: Number of projects to be processed. [Optional]
• --c: Whether to ignore processed projects. [Optional, default=False]
• --no-nlp: Whether to apply standard NLP techniques to extracted identifiers. [Optional, default=True]
• --pyre: Whether to run pyre to infer the types of variables for given projects. [Optional, default=False]
• --tc: Whether to type-check type annotations in projects. [Optional, default=False]

To merge all the processed JSON-formatted projects into a single dataframe, run the following command:

libsa4py merge --o $OUTPUT_PATH --l $LIMIT

Description:

• --o $OUTPUT_PATH: Path to the processed projects, used in the previous processing step.
• --l $LIMIT: Number of projects to be merged. [Optional]

To apply Pyre's inferred types to projects, run the following command:

libsa4py apply --p $REPOS_PATH --o $OUTPUT_PATH

Description:

• --p $REPOS_PATH: The path to the Python corpus or dataset.
• --o $OUTPUT_PATH: Path to the processed projects, used in the previous processing step.

After processing each project, a JSON-formatted file is produced, which is described here.