Demand-driven Pointer Analysis on Explicit Dependence Graphs

In this thesis I present a demand-driven pointer analysis for Java that aims at applications which require memory and time efficiency of the pointer analysis, such as tools for IDEs or JIT compilers. It furthermore allows clients to specify the precision needed and is context-sensitive and flow-sensitive. The demand-driven aspect of my approach allows the analysis to ignore irrelevant parts of the program. It is also client-driven via an iterative refinement method that allows the client to terminate the analysis early in case sufficient precision is achieved. The experimental evaluation of the approachhowever shows that the pointer analysis cannot be precise while terminating within a time budget. 66% of the analyzed queries exceeded even large budgets on average. I show that the flow-sensitive analysis of load statements requires too much computation without careful restriction of analyzed statements. Furthermore, I demonstrate that the refinement idea cannot be used easily in a flow-sensitive demand-driven pointer analysis.