Synthèse de programmes | EPFL Graph Search

En informatique, la synthèse de programmes consiste à construire automatiquement un programme à partir d'une spécification. La spécification est décrite dans un langage logique, par exemple en logique temporelle linéaire. La synthèse de programmes s'appuie sur des techniques de vérification formelle de programmes. Le problème de synthèse de programmes remonte aux travaux d'Alonzo Church. Synthèse à partir d'une spécification en logique du premier ordre Manna et Waldinger ont proposé une méthode déductive pour synthétiser un programme à partir d'une spécification en logique du premier ordre. Synthèse à partir d'une spécification en logique temporelle linéaire vignette|Un robot, comme l'Atlas, peut exécuter un programme synthétisé afin de réaliser une tâche et pouvoir réagir à toutes les éventualités de l'environnement. Le programme de synthèse de programmes consiste à générer un plan pour un agent (par exemple un robot) qui réussit face à toutes les éventualités

Scaling Functional Synthesis and Repair

Emmanouil Koukoutos

Program synthesis was first proposed a few decades ago, but in the last decade it has gained increased momentum in the research community. The increasing complexity of software has dictated the urgent need for improved supporting tools that verify the software's correctness, and that automatically generate code from a formal contract provided by the programmer, along with a proof of the generated code's correctness. In addition, recent technological developments have provided tools that have enabled researchers to revisit the synthesis problem. The recent rise of SMT solvers has given synthesis tools a reliable and automated way to verify synthesized programs against contracts. The introduction of counter-example guided inductive synthesis has provided researchers with a flexible synthesis algorithm that they can adapt according to their specific domain. In this dissertation, we develop new algorithms to synthesize recursive functional programs with algebraic data types from formal specifications and/or input-output examples. We manage to scale beyond the reach of other similar tools to synthesize nontrivial functional programs, with a focus on data structure transformations. First, we address the problem of precisely specifying the desired space of candidate programs, described by context free grammars (CFGs). We implement and evaluate a method for reducing the program space by describing axioms of the target language and other domain-specific restrictions on the level of the CFG, without explicitly generating and rejecting undesirable programs. We provide a method that extracts a program model from a corpus of code and that builds a probabilistic CFG from it. We showcase the usefulness, both individually and in tandem, of these methods. Second, we develop an algorithm to efficiently traverse a possibly unbounded space of candidate programs generated from a probabilistic CFG. This algorithm is an implementation of the A* best-first search algorithm on the derivation graph generated from the CFG, with a number of domain-specific optimizations. We evaluate the efficiency of the algorithm as well as the effectiveness of the optimizations. Finally, we describe a program repair framework that locates and fixes bugs in erroneous functional programs. Our novel fault localization technique detects erroneous snippets with concrete execution and eliminates false positives by analyzing dependencies between execution traces. After the erroneous code snippet is discovered, a modified version of our synthesis algorithm generates fixes for it by introducing modifications to the original erroneous code.

EPFL2019

Deductive Synthesis and Repair

Etienne Kneuss

In this thesis, we explore techniques for the development of recursive functional programs over unbounded domains that are proved correct according to their high-level specifications. We present algorithms for automatically synthesizing executable code, starting from the speci- fication alone. We implement these algorithms in the Leon system. We augment relational specifications with a concise notation for symbolic tests, which are are helpful to characterize fragments of the functionsâ behavior. We build on our synthesis procedure to automatically repair invalid functions by generating alternative implementations. Our approach therefore formulates program repair in the framework of deductive synthesis and uses the existing program structure as a hint to guide synthesis. We rely on user-specified tests as well as automatically generated ones to localize the fault. This localization enables our procedure to repair functions that would otherwise be out of reach of our synthesizer, and ensures that most of the original behavior is preserved. We also investigate multiple ways of enabling Leon programs to interact with external, un- trusted code. For that purpose, we introduce a precise inter-procedural effect analysis for arbitrary Scala programs with mutable state, dynamic object allocation, and dynamic dispatch. We analyzed the Scala standard library containing 58000 methods and classified them into sev- eral categories according to their effects. Our analysis proves that over one half of all methods are pure, identifies a number of conditionally pure methods, and computes summary graphs and regular expressions describing the side effects of non-pure methods. We implement the synthesis and repair algorithms within the Leon system and deploy them as part of a novel interactive development environment available as a web interface. Our implementation is able to synthesize, within seconds, a number of useful recursive functions that manipulate unbounded numbers and data structures. Our repair procedure automatically locates various kinds of errors in recursive functions and fixes them by synthesizing alternative implementations.

EPFL2016

On Deductive Program Repair in Leon

Etienne Kneuss, Viktor Kuncak

We present an approach to program repair and its application to programs with recursive functions over unbounded data types. Our approach formulates program repair in the framework of deductive synthesis that uses existing program structure as a hint to guide synthesis. We introduce a new specification construct for symbolic tests. We rely on such user-specified tests as well as automatically generated ones to localize the fault and speed up synthesis. Our implementation is able to eliminate errors within seconds from a variety of functional programs, including symbolic computation code and implementations of functional data structures. The resulting programs are formally verified by the Leon system.

2015