Towards a Large-Scale Empirical Study of Python Static Type Annotations

Python, as one of the most popular and important programming languages in the era of data science, has recently introduced a syntax for static type annotations with PEP 484, to improve code maintainability, quality, and readability. However, it is still unknown whether and how static type annotations are used in practical Python projects.This paper presents, to the best of our knowledge, the first and most comprehensive empirical study on the defects, evolution and rectification of static type annotations in Python projects. We first designed and implemented a software prototype dubbed PYSCAN, then used it to scan notable Python projects with diverse domains and sizes and type annotation manners, which add up to 19,478,428 lines of Python code. The empirical results provide interesting findings and insights, such as: 1) we proposed a taxonomy of Python type annotation-related defects, by classifying defects into four categories; 2) we investigated the evolution of type annotation-related defects; and 3) we proposed automatic defect rectification strategies, generating rectification suggestions for 82 out of 110 (74.55%) defects successfully. We suggest that: 1) Python language designers should clarify the type annotation specification; 2) checking tool builders should improve their tools to suppress false positives; and 3) Python developers should integrate such checking tools into their development workflow to catch type annotation-related defects at an early development stage.We have reported our findings and suggestions to Python language designers, checking tool builders, and Python developers. They have acknowledged us and taken actions based on our suggestions. We believe these guidelines would improve static type annotation practices and benefit the Python ecosystem in general.