Finding Reference-Counting Errors in Python']Python/C Programs with Affine Analysis

Python is a popular programming language that uses reference counting to manage heap objects. Python also has a Foreign Function Interface (FFI) that allows Python extension modules to be written in native code such as C and C++. Native code, however, is outside Python's system of memory management; therefore extension programmers are responsible for making sure these objects are reference counted correctly. This is an error prone process when code becomes complex. In this paper, we propose Pungi, a system that statically checks whether Python objects' reference counts are adjusted correctly in Python/C interface code. Pungi transforms Python/C interface code into affine programs with respect to our proposed abstractions of reference counts. Our system performs static analysis on transformed affine programs and reports possible reference counting errors. Our prototype implementation found over 150 errors in a set of Python/C programs.