Abstract

We present a new approach to code compression which was designed as part of a larger infrastructure for mobile code deployment. Our approach operates on abstract syntax trees and can be parameterized by abstract grammars, making it applicable to any source language without further language-specific modifications. It employs a variation of prediction by partial match to drive an arithmetic coder adaptively, thereby achieving unmatched compression results without requiring prior statistical analysis. We have developed a prototype implementation and have used it to compress Java programs. It reduces the size of compressed Java classes/packages by 5--50 per cent compared to the best published compression scheme, which was specifically tailored towards Java.

Citations