Some trade-offs in reducing the overhead of assertion run-time checks via static analysis

Abstract

A number of approaches for helping programmers detect incorrect program behaviors are based on combining language-level constructs (such as procedure-level assertions/contracts, program-point assertions, or gradual types) with a number of associated tools (such as code analyzers and run-time verification frameworks) that automatically check the validity of such constructs. However, these constructs and tools are often not used to their full extent in practice due to excessive run-time overhead, limited expressiveness, and/or limitations in the effectiveness of the tools. Verification frameworks that combine static and dynamic verification techniques and are based on abstraction offer the potential to bridge this gap. In this paper we explore the effectiveness of abstract interpretation in detecting parts of program specifications that can be statically simplified to true or false, as well as in reducing the cost of the run-time checks required for the remaining parts of these specifications. Starting with a semantics for programs with assertion checking, and for assertion simplification based on static analysis information obtained via abstract interpretation, we propose and study a number of practical assertion checking “modes,” each of which represents a trade-off between code annotation depth, execution time slowdown, and program safety. We then explore these modes in two typical, library-oriented scenarios. We also propose program transformation-based methods for taking advantage of the run-time checking semantics to improve the precision of the analysis. Finally, we study experimentally the performance of these techniques. Our experiments illustrate the benefits and costs of each of the assertion checking modes proposed, as well as the benefits obtained from analysis and the proposed transformations in these scenarios. • Most program code instrumentations with run-time checks generalize into 3 scenarios. • Run-time checks when input to static program analysis increase its precision. • Checks treatment by static analysis is achieved via a simple program transformation. • Major run-time checks overhead reductions are observed in the most costly scenario.