Static Data Flow Analysis for Realtime Java Runtime Error Detection

International audienceThe strict and clear semantics of Java make it an ideal language for static analysis. Nevertheless, the use of programwide pointer analysis for proving the absence of Java runtime error conditions such as null pointer uses or illegal array indices is still not widespread. Current uses of program-wide pointer analysis focus on extracting information for optimisations in compilers. In this case, imprecise analysis results only in less agressive optimisation, which is often tolerable.Existing implementations of program-wide data flow analysis either lack the required accuracy to prove the absence of large enough numbers of certain errors or cause an explosion in analysis time and space requirements for non-trivial applications. Low accuracy leads to large numbers of “false positives”, i.e., code for which the analysis fails to prove that a certain error condition does not occur. An explosion in the analysis effort causes the analysis not to produce any useful results within reasonable time at all.The approach presented in this paper applies the analysis to create a correctness proof of safety-critical Java applications. This includes the absence of runtime errors (null pointer use, division by zero, etc.), the absence of potential deadlocks, the correctness of region-based memory management and the determination of resource constraints (heap and stack use)