Writing Reusable Infopipes Using DirectFlow

An information flow program is one that processes or transforms streams of data packets. The object-based Infopipes system allows an information flow program to be constructed by connecting objects with explicit data channels, but a phenomenon known as port polarity makes code reuse difficult by requiring the programmer to maintain multiple versions of the same Infopipe that differ slightly in their interfaces. I present an experimental language called DirectFlow that frees the programmer from managing port polarity directly, and show how to translate a DirectFlow program back into objects with suitable port polarity configurations.