Using dynamic analysis to infer Python programs and convert them into database programs

Abstract

I present Nero, a new system that automatically infers and regenerates programs that access databases. The developer first implements a Python program that uses lists and dictionaries to implement the database functionality. Nero then instruments the Python list and dictionary implementations and uses active learning to generate inputs that enable it to infer the behavior of the program. The program can be implemented in any arbitrary style as long as it implements behavior expressible in the domain specific language that characterizes the behaviors that Nero is designed to infer. The regenerated program replaces the Python lists and dictionaries with database tables and contains all code required to successfully access the databases. Results from several inferred and regenerated applications highlight the ability of Nero to enable developers with no knowledge of database programming to obtain programs that successfully access databases.