Acm Workshop on Strategic Directions in Computing Research the Abstract Interpretation Perspective Program Analysis Position Statement: 1. Situation of Program Analysis 1.1 Deenition 1.2 from Programs to Systems

Semantic analysis (of programs, systems, etc.) should become a mature discipline to be included in educational curricula in computer science; This requires the discipline to evolve from an ill-assorted collection of methods and algorithms to a large scope formal reasoning and design methodology; In particular, program analysis methods should be explained in language, program encoding, semantics, property, property encoding, and approximation independent way, to be of very broad scope and wide usefulness ; This might be possible by reasoning on the approximation of the structures involved in semantical speciica-tions; This eeort would culminate in a general theory of se-mantical approximation leading to a tractable compos-able parameterized semantical analysis design methodology ; From a more practical point of view, the methodology would be applied to the uniform design of: { hierarchies of semantics, { hierarchies of abstract algebras at various levels of abstraction/reenement; This would lead to generic program analysis tools most of which could be included in reusable and composable libraries; An ultimate practical goal would be to automate the generation of such semantic analysis tools primarily for program analysis and more generally for system speci-cation analysis. An extended argumentation in favor of this position statement is given below. Program semantic analysis consists of designing and writing program analyzers. A program analyzer is a program which takes as input data a (possibly annotated) program (written in some language) and (fully automatically) produces as output answers to questions about runtime properties which are valid for all (or sometimes for some) possible executions of that program. Because of undecidability or complexity , these answers are necessarily partial, but should be irrefutable. Traditional applications of program analysis concern: Program compilation (e.g. guaranteeing the correctness of a compiler optimization 58]); Program debugging (e.g. automatic test case generation, abstract debugging 3] (where data abstract properties are used instead of data values in classical debugging) or automatic veriication of assertions); Program transformation (e.g. ensuring termination of partial evaluations 40]). The methods used in program analysis, which do cope with undecidability, have a much broader potential scope of application which is still largely unexplored. For example: Analysis of high conndence reactive, real-time, parallel, or distributed innnite state computer systems (e.g. decomposition and simpliication into a partial analysis by abstraction); Debugging of software speciications (e.g. of air traac con-Program analysis is both a theoretical activity (close to formal speciication methods, semantics, etc.) and a practical activity (close to …