Reaching Python from Racket

Abstract

Racket is a descendant of Scheme, a language that has been widely used to teach computer science. Recently, the Python language has taken over this role, mainly due to its huge standard library and the great number of third-party libraries available. Given that the development of equivalent libraries for Racket is an enormous task that cannot be currently done in an acceptable time frame, the next best option is to allow the Racket platform to use Python programs and libraries. We have been developing an implementation of Python for the Racket platform based on a source-to-source compiler. In order to provide good performance and good interoperability with the Racket platform, the runtime libraries are being implemented over Racket data-types. This, however, entails implementing all of Python's standard library on Racket and it does not provide access to popular Python libraries implemented using C module extensions (such as Numpy and SciPy). This paper presents an alternative approach that allows libraries from Python's reference implementation to be imported and used in our Racket implementation of Python, immediately providing access to all of Python's standard library and every third-party library, including NumPy and SciPy. The proposed solution involves importing Python module objects directly from Python's virtual machine, by calling the Python/C API through Racket's Foreign Function Interface, and converting them to objects usable by our Racket runtime libraries, making them compatible with the Racket platform. This compatibility layer therefore relies on relatively expensive foreign function calls to Python's libraries, but our performance tests show that the overhead introduced by them is quite low and, for most cases, it can be minimized in order to attain the same performance as Python's reference implementation.