Multivariant Assertion-based Guidance in Abstract Interpretation
share
Approximations during program analysis are a necessary evil, as they ensure essential properties, such as analysis soundness and termination, but they also imply that the analysis is not guaranteed to always produce useful results. In such cases it is necessary to have some means for users to provide information to guide analysis and thus to improve precision and/or shorten analysis time. This allows dealing with, e.g., constructs for which the analysis is not complete and loses precision or for which the source is only partially available. We present techniques for supporting within an abstract interpretation framework a rich set of assertions that can deal with multivariance/context-sensitivity, and can handle different run-time semantics for those assertions that cannot be discharged at compile time. We show how the proposed approach can be applied to both improving precision and accelerating convergence, and provide a procedure to decide whether the invariants used to guide the analyzer are checked, incompatible, or neither. We also provide some formal results on the effects of such assertions on the analysis results.
READ FULL TEXT