PyTy: Repairing Static Type Errors in Python

PyTy is an automated program repair approach specifically designed for Python type errors. It utilizes a learning-based model trained on a dataset of Python type error fixes called PyTyDefects.

Purpose

Submission for ICSE 2024 Artifact: - Available Badge: We provide the artifact with a permanent DOI from Zenodo and also maintain a public GitHub repository for the project. - Reusable Badge: We describe how to reproduce the paper’s results using Docker and use the tool to fix new bugs in other repositories.

Provenance

The source code and data are publicly available on Zenodo (link) and GitHub.
The last GitHub commit before submitting the artifact is: 5207f1419724fd5a690cfa9e039678fca0773141.

Data

We include the dataset we collected, named PyTyDefects. The full dataset in JSON format is available in the folder: ./src/Input. Each JSON file represents a commit containing one or more type error fixes. The fixes are isolated using a variant of delta debugging. This dataset can be reused for other studies and approaches.

Setup

Hardware: To run the script in “FAST MODE”, a normal computer suffices. For “SLOW MODE”, we used (and recommend) a server with 250 GB RAM, 48 Intel Xeon CPU cores with 2.2Ghz, and an NVIDIA Telta V100 GPU.
Software:
Ubuntu OS
Docker (installation instructions)

Important files

You can find the detailed instructions in: - [README] The file README.md contains instructions to reproduce results and to reuse the code with a different dataset. - [LICENSE] The file LICENSE.md contains the open source license for the source code. - [REQUIREMENT] The file requirements.txt contains all the requirements to run the source code. - [PAPER] The file paper.pdf contains the paper accepted at ICSE 2024.