Modifications to the data representation of an abstract data type (ADT) can require significant semantic refactoring of the code. Motivated by this observation, this paper presents a new method to automate semantic code refactoring tasks. Our method takes as input the original ADT implementation, a new data representation, and a so-called relational representation invariant (relating the old and new data representations), and automatically generates a new ADT implementation that is semantically equivalent to the original version. Our method is based on counterexample-guided inductive synthesis (CEGIS) but leverages three key ideas that allow it to handle real-world refactoring tasks. First, our approach reduces the underlying relational synthesis problem to a set of (simpler) programming-by-example problems, one for each method in the ADT. Second, it leverages symbolic reasoning techniques, based on logical abduction, to deduce code snippets that should occur in the refactored version. Finally, it utilizes a notion of partial equivalence to make inductive synthesis much more effective in this setting. We have implemented the proposed approach in a new tool called Revamp for automatically refactoring Java classes and evaluated it on 30 Java class mined from Github. Our evaluation shows that Revamp can correctly refactor the entire ADT in 97% of the cases and that it can successfully re-implement 144 out of the 146 methods that require modifications.
{"title":"Semantic Code Refactoring for Abstract Data Types","authors":"Shankara Pailoor, Yuepeng Wang, Işıl Dillig","doi":"10.1145/3632870","DOIUrl":"https://doi.org/10.1145/3632870","url":null,"abstract":"Modifications to the data representation of an abstract data type (ADT) can require significant semantic refactoring of the code. Motivated by this observation, this paper presents a new method to automate semantic code refactoring tasks. Our method takes as input the original ADT implementation, a new data representation, and a so-called relational representation invariant (relating the old and new data representations), and automatically generates a new ADT implementation that is semantically equivalent to the original version. Our method is based on counterexample-guided inductive synthesis (CEGIS) but leverages three key ideas that allow it to handle real-world refactoring tasks. First, our approach reduces the underlying relational synthesis problem to a set of (simpler) programming-by-example problems, one for each method in the ADT. Second, it leverages symbolic reasoning techniques, based on logical abduction, to deduce code snippets that should occur in the refactored version. Finally, it utilizes a notion of partial equivalence to make inductive synthesis much more effective in this setting. We have implemented the proposed approach in a new tool called Revamp for automatically refactoring Java classes and evaluated it on 30 Java class mined from Github. Our evaluation shows that Revamp can correctly refactor the entire ADT in 97% of the cases and that it can successfully re-implement 144 out of the 146 methods that require modifications.","PeriodicalId":20697,"journal":{"name":"Proceedings of the ACM on Programming Languages","volume":null,"pages":null},"PeriodicalIF":1.8,"publicationDate":"2024-01-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"139382656","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Subtypes and quotient types are dual type abstractions. However, while subtypes are widely used both explicitly and implicitly, quotient types have not seen much practical use outside of proof assistants. A key difficulty to wider adoption of quotient types lies in the significant burden of proof-obligations that arises from their use. In this article, we address this issue by introducing a class of quotient types for which the proof-obligations are decidable by an SMT solver. We demonstrate this idea in practice by presenting Quotient Haskell, an extension of Liquid Haskell with support for quotient types.
{"title":"Quotient Haskell: Lightweight Quotient Types for All","authors":"Brandon Hewer, Graham Hutton","doi":"10.1145/3632869","DOIUrl":"https://doi.org/10.1145/3632869","url":null,"abstract":"Subtypes and quotient types are dual type abstractions. However, while subtypes are widely used both explicitly and implicitly, quotient types have not seen much practical use outside of proof assistants. A key difficulty to wider adoption of quotient types lies in the significant burden of proof-obligations that arises from their use. In this article, we address this issue by introducing a class of quotient types for which the proof-obligations are decidable by an SMT solver. We demonstrate this idea in practice by presenting Quotient Haskell, an extension of Liquid Haskell with support for quotient types.","PeriodicalId":20697,"journal":{"name":"Proceedings of the ACM on Programming Languages","volume":null,"pages":null},"PeriodicalIF":1.8,"publicationDate":"2024-01-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"139383062","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Verifying safety and liveness over array systems is a highly challenging problem. Array systems naturally capture parameterized systems such as distributed protocols with an unbounded number of processes. Such distributed protocols often exploit process IDs during their computation, resulting in array systems whose element values range over an infinite domain. In this paper, we develop a novel framework for proving safety and liveness over array systems. The crux of the framework is to overapproximate an array system as a string rewriting system (i.e. over a finite alphabet) by means of a new predicate abstraction that exploits the so-called indexed predicates. This allows us to tap into powerful verification methods for string rewriting systems that have been heavily developed in the last two decades or so (e.g. regular model checking). We demonstrate how our method yields simple, automatically verifiable proofs of safety and liveness properties for challenging examples, including Dijkstra's self-stabilizing protocol and the Chang-Roberts leader election protocol.
{"title":"Regular Abstractions for Array Systems","authors":"Chih-Duo Hong, Anthony W. Lin","doi":"10.1145/3632864","DOIUrl":"https://doi.org/10.1145/3632864","url":null,"abstract":"Verifying safety and liveness over array systems is a highly challenging problem. Array systems naturally capture parameterized systems such as distributed protocols with an unbounded number of processes. Such distributed protocols often exploit process IDs during their computation, resulting in array systems whose element values range over an infinite domain. In this paper, we develop a novel framework for proving safety and liveness over array systems. The crux of the framework is to overapproximate an array system as a string rewriting system (i.e. over a finite alphabet) by means of a new predicate abstraction that exploits the so-called indexed predicates. This allows us to tap into powerful verification methods for string rewriting systems that have been heavily developed in the last two decades or so (e.g. regular model checking). We demonstrate how our method yields simple, automatically verifiable proofs of safety and liveness properties for challenging examples, including Dijkstra's self-stabilizing protocol and the Chang-Roberts leader election protocol.","PeriodicalId":20697,"journal":{"name":"Proceedings of the ACM on Programming Languages","volume":null,"pages":null},"PeriodicalIF":1.8,"publicationDate":"2024-01-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"139383519","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
On any modern computer architecture today, parallelism comes with a modest cost, born from the creation and management of threads or tasks. Today, programmers battle this cost by manually optimizing/tuning their codes to minimize the cost of parallelism without harming its benefit, performance. This is a difficult battle: programmers must reason about architectural constant factors hidden behind layers of software abstractions, including thread schedulers and memory managers, and their impact on performance, also at scale. In languages that support higher-order functions, the battle hardens: higher order functions can make it difficult, if not impossible, to reason about the cost and benefits of parallelism. Motivated by these challenges and the numerous advantages of high-level languages, we believe that it has become essential to manage parallelism automatically so as to minimize its cost and maximize its benefit. This is a challenging problem, even when considered on a case-by-case, application-specific basis. But if a solution were possible, then it could combine the many correctness benefits of high-level languages with performance by managing parallelism without the programmer effort needed to ensure performance. This paper proposes techniques for such automatic management of parallelism by combining static (compilation) and run-time techniques. Specifically, we consider the Parallel ML language with task parallelism, and describe a compiler pipeline that embeds "potential parallelism" directly into the call-stack and avoids the cost of task creation by default. We then pair this compilation pipeline with a run-time system that dynamically converts potential parallelism into actual parallel tasks. Together, the compiler and run-time system guarantee that the cost of parallelism remains low without losing its benefit. We prove that our techniques have no asymptotic impact on the work and span of parallel programs and thus preserve their asymptotic properties. We implement the proposed techniques by extending the MPL compiler for Parallel ML and show that it can eliminate the burden of manual optimization while delivering good practical performance.
在当今任何现代计算机体系结构中，并行性都会产生一定的成本，这来自于线程或任务的创建和管理。如今，程序员通过手动优化/调整代码来降低并行成本，同时又不损害并行的优势--性能。这是一场艰苦的战斗：程序员必须推理隐藏在层层软件抽象（包括线程调度器和内存管理器）背后的架构常量因素，以及它们对性能（同样是大规模性能）的影响。在支持高阶函数的语言中，这场战斗更加艰苦：高阶函数可能会使推理并行性的成本和收益变得困难，甚至不可能。在这些挑战和高级语言众多优势的激励下，我们认为，自动管理并行性以最小化其成本和最大化其收益已变得至关重要。这是一个极具挑战性的问题，即使是根据具体情况和特定应用来考虑也是如此。但是，如果有可能找到一种解决方案，那么它就可以通过管理并行性，将高级语言的许多正确性优势与性能结合起来，而无需程序员为确保性能而付出努力。本文结合静态（编译）和运行时技术，提出了自动管理并行性的技术。具体来说，我们考虑了具有任务并行性的 Parallel ML 语言，并描述了一种编译器流水线，该流水线可将 "潜在并行性 "直接嵌入调用堆栈，并在默认情况下避免任务创建的成本。然后，我们将该编译管道与运行时系统配对，后者可将潜在并行性动态转换为实际并行任务。编译器和运行时系统共同保证了并行的低成本，同时又不失其优势。我们证明，我们的技术对并行程序的工作和跨度没有渐进影响，因此保留了它们的渐进特性。我们通过为并行 ML 扩展 MPL 编译器来实现所提出的技术，并证明它可以消除手动优化的负担，同时提供良好的实用性能。
{"title":"Automatic Parallelism Management","authors":"Sam Westrick, M. Fluet, Mike Rainey, Umut A. Acar","doi":"10.1145/3632880","DOIUrl":"https://doi.org/10.1145/3632880","url":null,"abstract":"On any modern computer architecture today, parallelism comes with a modest cost, born from the creation and management of threads or tasks. Today, programmers battle this cost by manually optimizing/tuning their codes to minimize the cost of parallelism without harming its benefit, performance. This is a difficult battle: programmers must reason about architectural constant factors hidden behind layers of software abstractions, including thread schedulers and memory managers, and their impact on performance, also at scale. In languages that support higher-order functions, the battle hardens: higher order functions can make it difficult, if not impossible, to reason about the cost and benefits of parallelism. Motivated by these challenges and the numerous advantages of high-level languages, we believe that it has become essential to manage parallelism automatically so as to minimize its cost and maximize its benefit. This is a challenging problem, even when considered on a case-by-case, application-specific basis. But if a solution were possible, then it could combine the many correctness benefits of high-level languages with performance by managing parallelism without the programmer effort needed to ensure performance. This paper proposes techniques for such automatic management of parallelism by combining static (compilation) and run-time techniques. Specifically, we consider the Parallel ML language with task parallelism, and describe a compiler pipeline that embeds \"potential parallelism\" directly into the call-stack and avoids the cost of task creation by default. We then pair this compilation pipeline with a run-time system that dynamically converts potential parallelism into actual parallel tasks. Together, the compiler and run-time system guarantee that the cost of parallelism remains low without losing its benefit. We prove that our techniques have no asymptotic impact on the work and span of parallel programs and thus preserve their asymptotic properties. We implement the proposed techniques by extending the MPL compiler for Parallel ML and show that it can eliminate the burden of manual optimization while delivering good practical performance.","PeriodicalId":20697,"journal":{"name":"Proceedings of the ACM on Programming Languages","volume":null,"pages":null},"PeriodicalIF":1.8,"publicationDate":"2024-01-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"139384261","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Software contracts empower programmers to describe functional properties of components. When it comes to constraining effects, though, the literature offers only one-off solutions for various effects. It lacks a universal principle. This paper presents the design of an effectful contract system in the context of effect handlers. A key metatheorem shows that contracts cannot unduly interfere with a program's execution. An implementation of this design, along with an evaluation of its generality, demonstrates that the theory can guide practice.
{"title":"Effectful Software Contracts","authors":"Cameron Moy, Christos Dimoulas, Matthias Felleisen","doi":"10.1145/3632930","DOIUrl":"https://doi.org/10.1145/3632930","url":null,"abstract":"Software contracts empower programmers to describe functional properties of components. When it comes to constraining effects, though, the literature offers only one-off solutions for various effects. It lacks a universal principle. This paper presents the design of an effectful contract system in the context of effect handlers. A key metatheorem shows that contracts cannot unduly interfere with a program's execution. An implementation of this design, along with an evaluation of its generality, demonstrates that the theory can guide practice.","PeriodicalId":20697,"journal":{"name":"Proceedings of the ACM on Programming Languages","volume":null,"pages":null},"PeriodicalIF":1.8,"publicationDate":"2024-01-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"139382540","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Xing Zhang, Ruifeng Xie, Guanchen Guo, Xiao He, Tao Zan, Zhenjiang Hu
Bidirectional live programming systems (BLP) enable developers to modify a program by directly manipulating the program output, so that the updated program can produce the manipulated output. One state-of-the-art approach to BLP systems is operation-based, which captures the developer's intention of program modifications by taking how the developer manipulates the output into account. The program modifications are usually hard coded for each direct manipulation in these BLP systems, which are difficult to extend. Moreover, to reflect the manipulations to the source program, these BLP systems trace the modified output to appropriate code fragments and perform corresponding code transformations. Accordingly, they require direct manipulation users be aware of the source code and how it is changed, making "direct" manipulation (on output) be "indirect". In this paper, we resolve this problem by presenting a novel operation-based framework for bidirectional live programming, which can automatically fuse direct manipulations into the source code, thus supporting code-insensitive direct manipulations. Firstly, we design a simple but expressive delta language DM capable of expressing common direct manipulations for output values. Secondly, we present a fusion algorithm that propagates direct manipulations into the source functional programs and applies them to the constants whenever possible; otherwise, the algorithm embeds manipulations into the "proper positions" of programs. We prove the correctness of the fusion algorithm that the updated program executes to get the manipulated output. To demonstrate the expressiveness of DM and the effectiveness of our fusion algorithm, we have implemented FuseDM, a prototype SVG editor that supports GUI-based operations for direct manipulation, and successfully designed 14 benchmark examples starting from blank code using FuseDM.
{"title":"Fusing Direct Manipulations into Functional Programs","authors":"Xing Zhang, Ruifeng Xie, Guanchen Guo, Xiao He, Tao Zan, Zhenjiang Hu","doi":"10.1145/3632883","DOIUrl":"https://doi.org/10.1145/3632883","url":null,"abstract":"Bidirectional live programming systems (BLP) enable developers to modify a program by directly manipulating the program output, so that the updated program can produce the manipulated output. One state-of-the-art approach to BLP systems is operation-based, which captures the developer's intention of program modifications by taking how the developer manipulates the output into account. The program modifications are usually hard coded for each direct manipulation in these BLP systems, which are difficult to extend. Moreover, to reflect the manipulations to the source program, these BLP systems trace the modified output to appropriate code fragments and perform corresponding code transformations. Accordingly, they require direct manipulation users be aware of the source code and how it is changed, making \"direct\" manipulation (on output) be \"indirect\". In this paper, we resolve this problem by presenting a novel operation-based framework for bidirectional live programming, which can automatically fuse direct manipulations into the source code, thus supporting code-insensitive direct manipulations. Firstly, we design a simple but expressive delta language DM capable of expressing common direct manipulations for output values. Secondly, we present a fusion algorithm that propagates direct manipulations into the source functional programs and applies them to the constants whenever possible; otherwise, the algorithm embeds manipulations into the \"proper positions\" of programs. We prove the correctness of the fusion algorithm that the updated program executes to get the manipulated output. To demonstrate the expressiveness of DM and the effectiveness of our fusion algorithm, we have implemented FuseDM, a prototype SVG editor that supports GUI-based operations for direct manipulation, and successfully designed 14 benchmark examples starting from blank code using FuseDM.","PeriodicalId":20697,"journal":{"name":"Proceedings of the ACM on Programming Languages","volume":null,"pages":null},"PeriodicalIF":1.8,"publicationDate":"2024-01-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"139383233","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Itsaka Rakotonirina, Gilles Barthe, Clara Schneidewind
The formal analysis of cryptographic protocols traditionally focuses on trace and equivalence properties, for which decision procedures in the symbolic (or Dolev-Yao, or DY) model are known. However, many relevant security properties are expressed as DY hyperproperties that involve quantifications over both execution paths and attacker computations (which are constrained by the attacker's knowledge in the underlying model of computation). DY hyperproperties generalise hyperproperties, for which many decision procedures exist, to the setting of DY models. Unfortunately, the subtle interactions between both forms of quantifications have been an obstacle to lifting decision procedures from hyperproperties to DY hyperproperties. The central contribution of the paper is the first procedure for deciding DY hyperproperties, in the usual setting where the number of protocol sessions is bounded and where the equational theory modelling cryptography is subterm-convergent. We prove that our decision procedure can decide the validity of any hyperproperty in which quantifications over messages are guarded and quantifications over attacker computations are limited to expressing the attacker's knowledge. We also establish the complexity of the decision problem for several important fragments of the hyperlogic. Further, we illustrate the techniques and scope of our contributions through examples of related hyperproperties.
密码协议的形式分析传统上侧重于跟踪和等价特性，而符号模型（或 Dolev-Yao 模型，或 DY 模型）中的决策程序是已知的。然而，许多相关的安全属性都可以用 DY 超属性来表示，其中涉及对执行路径和攻击者计算的量化（受攻击者在底层计算模型中的知识限制）。DY 超属性将超属性泛化到了 DY 模型的环境中，而对于超属性，已有许多决策程序。遗憾的是，两种量化形式之间微妙的相互作用一直是将决策程序从超属性提升到 DY 超属性的障碍。本文的核心贡献在于，在协议会话次数有界、等式理论建模密码学是子项收敛的通常情况下，首次提出了 DY 超属性决策程序。我们证明了我们的判定过程可以判定任何超属性的有效性，在这些超属性中，对信息的量化是受保护的，对攻击者计算的量化仅限于表达攻击者的知识。我们还为超逻辑的几个重要片段确定了决策问题的复杂性。此外，我们还通过相关超属性的例子来说明我们所贡献的技术和范围。
{"title":"Decision and Complexity of Dolev-Yao Hyperproperties","authors":"Itsaka Rakotonirina, Gilles Barthe, Clara Schneidewind","doi":"10.1145/3632906","DOIUrl":"https://doi.org/10.1145/3632906","url":null,"abstract":"The formal analysis of cryptographic protocols traditionally focuses on trace and equivalence properties, for which decision procedures in the symbolic (or Dolev-Yao, or DY) model are known. However, many relevant security properties are expressed as DY hyperproperties that involve quantifications over both execution paths and attacker computations (which are constrained by the attacker's knowledge in the underlying model of computation). DY hyperproperties generalise hyperproperties, for which many decision procedures exist, to the setting of DY models. Unfortunately, the subtle interactions between both forms of quantifications have been an obstacle to lifting decision procedures from hyperproperties to DY hyperproperties. The central contribution of the paper is the first procedure for deciding DY hyperproperties, in the usual setting where the number of protocol sessions is bounded and where the equational theory modelling cryptography is subterm-convergent. We prove that our decision procedure can decide the validity of any hyperproperty in which quantifications over messages are guarded and quantifications over attacker computations are limited to expressing the attacker's knowledge. We also establish the complexity of the decision problem for several important fragments of the hyperlogic. Further, we illustrate the techniques and scope of our contributions through examples of related hyperproperties.","PeriodicalId":20697,"journal":{"name":"Proceedings of the ACM on Programming Languages","volume":null,"pages":null},"PeriodicalIF":1.8,"publicationDate":"2024-01-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"139383285","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Yanis Sellami, Guillaume Girol, Frédéric Recoules, Damien Couroussé, S. Bardin
Characterization of bugs and attack vectors is in many practical scenarios as important as their finding. Recently, Girol et. al. have introduced the concept of robust reachability, which ensures a perfect reproducibility of the reported violations by distinguishing inputs that are under the control of the attacker (controlled inputs) from those that are not (uncontrolled inputs), and proposed first automated analysis for it. While it is a step toward distinguishing severe bugs from benign ones, it fails for example to describe violations that are mostly reproducible, i.e., when triggering conditions are likely to happen, meaning that they happen for all uncontrolled inputs but a few corner cases. To address this issue, we propose to leverage theory-agnostic abduction techniques to generate constraints on the uncontrolled program inputs that ensure that a target property is robustly satisfied. Our proposal comes with an extension of robust reachability that is generic on the type of trace property and on the technology used to verify the properties. We show that our approach is complete w.r.t its inference language, and we additionally discuss strategies for the efficient exploration of the inference space. We demonstrate the feasibility of the method and its practical ability to refine the notion of robust reachability with an implementation that uses robust reachability oracles to generate constraints on standard benchmarks from software verification and security analysis. We illustrate the use of our implementation to a vulnerability characterization problem in the context of fault injection attacks. Our method overcomes a major limitation of the initial proposal of robust reachability, without complicating its definition. From a practical view, this is a step toward new verification tools that are able to characterize program violations through high-level feedback.
{"title":"Inference of Robust Reachability Constraints","authors":"Yanis Sellami, Guillaume Girol, Frédéric Recoules, Damien Couroussé, S. Bardin","doi":"10.1145/3632933","DOIUrl":"https://doi.org/10.1145/3632933","url":null,"abstract":"Characterization of bugs and attack vectors is in many practical scenarios as important as their finding. Recently, Girol et. al. have introduced the concept of robust reachability, which ensures a perfect reproducibility of the reported violations by distinguishing inputs that are under the control of the attacker (controlled inputs) from those that are not (uncontrolled inputs), and proposed first automated analysis for it. While it is a step toward distinguishing severe bugs from benign ones, it fails for example to describe violations that are mostly reproducible, i.e., when triggering conditions are likely to happen, meaning that they happen for all uncontrolled inputs but a few corner cases. To address this issue, we propose to leverage theory-agnostic abduction techniques to generate constraints on the uncontrolled program inputs that ensure that a target property is robustly satisfied. Our proposal comes with an extension of robust reachability that is generic on the type of trace property and on the technology used to verify the properties. We show that our approach is complete w.r.t its inference language, and we additionally discuss strategies for the efficient exploration of the inference space. We demonstrate the feasibility of the method and its practical ability to refine the notion of robust reachability with an implementation that uses robust reachability oracles to generate constraints on standard benchmarks from software verification and security analysis. We illustrate the use of our implementation to a vulnerability characterization problem in the context of fault injection attacks. Our method overcomes a major limitation of the initial proposal of robust reachability, without complicating its definition. From a practical view, this is a step toward new verification tools that are able to characterize program violations through high-level feedback.","PeriodicalId":20697,"journal":{"name":"Proceedings of the ACM on Programming Languages","volume":null,"pages":null},"PeriodicalIF":1.8,"publicationDate":"2024-01-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"139383522","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Liron Cohen, Adham Jabarin, Andrei Popescu, R. Rowe
Cyclic proof systems, in which induction is managed implicitly, are a promising approach to automatic verification. The soundness of cyclic proof graphs is ensured by checking them against a trace-based Infinite Descent property. Although the problem of checking Infinite Descent is known to be PSPACE-complete, this leaves much room for variation in practice. Indeed, a number of different approaches are employed across the various cyclic proof systems described in the literature. In this paper, we study criteria for Infinite Descent in an abstract, logic-independent setting. We look at criteria based on Büchi automata encodings and relational abstractions, and determine their parameterized time complexities in terms of natural dimensions of cyclic proofs: the numbers of vertices of the proof-tree graphs, and the vertex width—an upper bound on the number of components (e.g., formulas) of a sequent that can be simultaneously tracked for descent. We identify novel algorithms that improve upon the parameterised complexity of the existing algorithms. We implement the studied criteria and compare their performance on various benchmarks.
{"title":"The Complex(ity) Landscape of Checking Infinite Descent","authors":"Liron Cohen, Adham Jabarin, Andrei Popescu, R. Rowe","doi":"10.1145/3632888","DOIUrl":"https://doi.org/10.1145/3632888","url":null,"abstract":"Cyclic proof systems, in which induction is managed implicitly, are a promising approach to automatic verification. The soundness of cyclic proof graphs is ensured by checking them against a trace-based Infinite Descent property. Although the problem of checking Infinite Descent is known to be PSPACE-complete, this leaves much room for variation in practice. Indeed, a number of different approaches are employed across the various cyclic proof systems described in the literature. In this paper, we study criteria for Infinite Descent in an abstract, logic-independent setting. We look at criteria based on Büchi automata encodings and relational abstractions, and determine their parameterized time complexities in terms of natural dimensions of cyclic proofs: the numbers of vertices of the proof-tree graphs, and the vertex width—an upper bound on the number of components (e.g., formulas) of a sequent that can be simultaneously tracked for descent. We identify novel algorithms that improve upon the parameterised complexity of the existing algorithms. We implement the studied criteria and compare their performance on various benchmarks.","PeriodicalId":20697,"journal":{"name":"Proceedings of the ACM on Programming Languages","volume":null,"pages":null},"PeriodicalIF":1.8,"publicationDate":"2024-01-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"139383610","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Convex hulls are commonly used to tackle the non-linearity of activation functions in the verification of neural networks. Computing the exact convex hull is a costly task though. In this work, we propose a fast and precise approach to over-approximating the convex hull of the ReLU function (referred to as the ReLU hull), one of the most used activation functions. Our key insight is to formulate a convex polytope that ”wraps” the ReLU hull, by reusing the linear pieces of the ReLU function as the lower faces and constructing upper faces that are adjacent to the lower faces. The upper faces can be efficiently constructed based on the edges and vertices of the lower faces, given that an n-dimensional (or simply nd hereafter) hyperplane can be determined by an (n−1)d hyperplane and a point outside of it. We implement our approach as WraLU, and evaluate its performance in terms of precision, efficiency, constraint complexity, and scalability. WraLU outperforms existing advanced methods by generating fewer constraints to achieve tighter approximation in less time. It exhibits versatility by effectively addressing arbitrary input polytopes and higher-dimensional cases, which are beyond the capabilities of existing methods. We integrate WraLU into PRIMA, a state-of-the-art neural network verifier, and apply it to verify large-scale ReLU-based neural networks. Our experimental results demonstrate that WraLU achieves a high efficiency without compromising precision. It reduces the number of constraints that need to be solved by the linear programming solver by up to half, while delivering comparable or even superior results compared to the state-of-the-art verifiers.
{"title":"ReLU Hull Approximation","authors":"Zhongkui Ma, Jiaying Li, Guangdong Bai","doi":"10.1145/3632917","DOIUrl":"https://doi.org/10.1145/3632917","url":null,"abstract":"Convex hulls are commonly used to tackle the non-linearity of activation functions in the verification of neural networks. Computing the exact convex hull is a costly task though. In this work, we propose a fast and precise approach to over-approximating the convex hull of the ReLU function (referred to as the ReLU hull), one of the most used activation functions. Our key insight is to formulate a convex polytope that ”wraps” the ReLU hull, by reusing the linear pieces of the ReLU function as the lower faces and constructing upper faces that are adjacent to the lower faces. The upper faces can be efficiently constructed based on the edges and vertices of the lower faces, given that an n-dimensional (or simply nd hereafter) hyperplane can be determined by an (n−1)d hyperplane and a point outside of it. We implement our approach as WraLU, and evaluate its performance in terms of precision, efficiency, constraint complexity, and scalability. WraLU outperforms existing advanced methods by generating fewer constraints to achieve tighter approximation in less time. It exhibits versatility by effectively addressing arbitrary input polytopes and higher-dimensional cases, which are beyond the capabilities of existing methods. We integrate WraLU into PRIMA, a state-of-the-art neural network verifier, and apply it to verify large-scale ReLU-based neural networks. Our experimental results demonstrate that WraLU achieves a high efficiency without compromising precision. It reduces the number of constraints that need to be solved by the linear programming solver by up to half, while delivering comparable or even superior results compared to the state-of-the-art verifiers.","PeriodicalId":20697,"journal":{"name":"Proceedings of the ACM on Programming Languages","volume":null,"pages":null},"PeriodicalIF":1.8,"publicationDate":"2024-01-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"139380971","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：