Program Analysis for Modern Software Features

This is an umbrella project aiming at a variety of techniques for making static and dynamic program analyses more usable and useful for the complex systems that exist in real-world software development. This effort includes both theoretical work (e.g., dataflow frameworks for analysis of large-scale software) and tool building (e.g., analysis implementations capable of analyzing thousands of classes with acceptable cost and precision). The different sub-projects are summarized below. For all references, please see the PRESTO list of publications. Some of this work is funded by a 5-year NSF CAREER grant, by a recent 3-year NSF grant, by an IBM Software Quality Innovation Award, and by an IBM Eclipse Innovation Award.

The long-term research agenda of this work is to enable the use of static and dynamic analyses in real-world software tools, with the ultimate goal of helping software developers and testers to do their work faster and better. Several significant challenges exist for this effort:

Beyond whole-program analysis: The traditional model of whole-program dataflow analysis is poorly suited for modern software systems. We have developed and successfully used a model of component-level dataflow analysis which addresses several of the fundamental limitations of whole-program analysis.
Challenging software features: Our current work considers other challenging aspects of modern software including bloated applications [PhD11, PLDI11, FOSER10, PLDI10-1, PLDI10-2, PLDI09, ICSE08], long-running applications [FSE07], run-time adaptation through dynamic language features [SCAM11, PhD09, JASE09, SCAM07], and aspect orientation [AOSD08, ICSE07]. The latest efforts are focused on analyzing various performance problems created by run-time bloat (more details here).
Scalability and precision: Static analysis researchers are on a never-ending quest to come up with analyses that can process larger programs, or that can achieve higher precision. Our work on scalability and precision has been focused on several important analyses for C and Java software (e.g., analysis of pointers and side effects) [ISSTA11, ECOOP09, ISSTA08, TOSEM05, JASE04, SCAM02, ISSTA02, OOPSLA01, PLDI00]. However, we believe that the "arms race" for improving whole-program analyses is not sustainable, and instead the focus should be on modular analyses (e.g., component-level analyses).
Distributed software: [currently inactive] We developed a variety of static analyses for Java applications that use the remote method invocation (RMI) mechanism. Such applications present numerous new challenges to static analysis researchers. We have successfully designed and implemented a general points-to analysis and side-effect analysis for RMI-based Java applications [PhD07, TSE06, ICSM05]. We have also worked on dynamic analysis of Enterprise Java applications [PhD08, WODA07].

main page