Conditional Dyck-CFL Reachability Analysis for Complete and Efficient Library Summarization

摘要

Library summarization is an effective way to accelerate the analysis of client code. However, information about the client is unknown at the library summarization, preventing complete summarization of the library. An existing approach utilizes tree-adjoining languages (TALs) to provide conditional summaries, enabling the summarization of a library under certain premises. However, the use of TAL imposes several problems, preventing a complete summarization of a library and reducing the efficiency of the analysis. In this paper we propose a new conditional summarization technique based on the context-free language (CFL) reachability analysis. Our technique overcomes the above two limitations of TAL, and is more accessible since CFL reachability is much more efficient and widely-used than TAL reachability. Furthermore, to overcome the high cost from premise combination, we also provide a technique to confine the number of premises while maintaining full summarization of the library. We empirically compared our approach with the state-of-art TAL conditional summarization technique on 12 Java benchmark subjects from the SPECjvm2008 benchmark suite. The results demonstrate that our approach is able to significantly outperform TAL on both efficiency and precision.