9804065 Software tools should have the choice of tunable, scalable, compile-time analyses, which provide adjustable levels of precision for predictable cost. New analysis techniques are needed to scale up to {it industrial-sized} systems (i.e., > 100,000 lines of code). Previous work seems conclusive: whole- program analysis does not scale suitably with a reasonable degree of precision for program transformation, data-flow-based testing applications, or program-understanding uses in certain circumstances. Thus, new techniques are needed. Program decomposition, uses a coarse-grained alias analysis of a C program to subdivide it into independently analyzable segments. Specific analyses, that vary in cost and precision, can be applied to different parts of a program. Initial experiments with this technique have enabled considerable gains in precision with modest increased cost for side-effect analysis of C programs. Research will focus on (i) development of techniques for selection of a specific analysis to apply to each program unit and (ii) empirical experimentation with these techniques using the PROLANGS Analysis Framework (PAF), a software platform for program analysis which currently includes flow- and context- sensitive side-effect and alias analysis for C programs.***
