Integrated program debugging, verification, and optimization using abstract interpretation (and the Ciao system preprocessor)

The technique of Abstract Interpretation has allowed the development of very sophisticated global program analyses which are at the same time provably correct and practical. We present in a tutorial fashion a novel program development framework which uses abstract interpretation as a fundamental tool. The framework uses modular, incremental abstract interpretation to obtain information about the program. This information is used to validate programs, to detect bugs with respect to partial specifications written using assertions (in the program itself and/or in system libraries), to generate and simplify run-time tests, and to perform high-level program transformations such as multiple abstract specialization, parallelization, and resource usage control, all in a provably correct way. In the case of validation and debugging, the assertions can refer to a variety of program points such as procedure entry, procedure exit, points within procedures, or global computations. The system can reason with much richer information than, for example, traditional types. This includes data structure shape (including pointer sharing), bounds on data structure sizes, and other operational variable instantiation properties, as well as procedure-level properties such as determinacy, termination, nonfailure, and bounds on resource consumption (time or space cost). CiaoPP, the preprocessor of the Ciao multi-paradigm programming system, which implements the described functionality, will be used to illustrate the fundamental ideas.

Program development using abstract interpretation (and the ciao system preprocessor)

The technique of Abstract Interpretation has allowed the development of very sophisticated global program analyses which are at the same time provably correct and practical. We present in a tutorial fashion a novel program development framework which uses abstract interpretation as a fundamental tool. The framework uses modular, incremental abstract interpretation to obtain information about the program. This information is used to validate programs, to detect bugs with respect to partial specifications written using assertions (in the program itself and/or in system librarles), to genérate and simplify run-time tests, and to perform high-level program transformations such as múltiple abstract specialization, parallelization, and resource usage control, all in a provably correct way. In the case of validation and debugging, the assertions can refer to a variety of program points such as procedure entry, procedure exit, points within procedures, or global computations. The system can reason with much richer information than, for example, traditional types. This includes data structure shape (including pointer sharing), bounds on data structure sizes, and other operational variable instantiation properties, as well as procedure-level properties such as determinacy, termination, non-failure, and bounds on resource consumption (time or space cost). CiaoPP, the preprocessor of the Ciao multi-paradigm programming system, which implements the described functionality, will be used to illustrate the fundamental ideas.

Program analysis, debugging and optimization using the ciao system preprocessor

We present a tutorial overview of Ciaopp, the Ciao system preprocessor. Ciao is a public-domain, next-generation logic programming system, which subsumes ISO-Prolog and is specifically designed to a) be highly extensible via librarles and b) support modular program analysis, debugging, and optimization. The latter tasks are performed in an integrated fashion by Ciaopp. Ciaopp uses modular, incremental abstract interpretation to infer properties of program predicates and literals, including types, variable instantiation properties (including modes), non-failure, determinacy, bounds on computational cost, bounds on sizes of terms in the program, etc. Using such analysis information, Ciaopp can find errors at compile-time in programs and/or perform partial verification. Ciaopp checks how programs cali system librarles and also any assertions present in the program or in other modules used by the program. These assertions are also used to genérate documentation automatically. Ciaopp also uses analysis information to perform program transformations and optimizations such as múltiple abstract specialization, parallelization (including granularity control), and optimization of run-time tests for properties which cannot be checked completely at compile-time. We illustrate "hands-on" the use of Ciaopp in all these tasks. By design, Ciaopp is a generic tool, which can be easily tailored to perform these and other tasks for different LP and CLP dialects.

A tutorial on program development and optimization using the Ciao preprocessor

We present in a tutorial fashion CiaoPP, the preprocessor of the Ciao multi-paradigm programming system, which implements a novel program development framework which uses abstract interpretation as a fundamental tool. The framework uses modular, incremental abstract interpretation to obtain information about the program. This information is used to validate programs, to detect bugs with respect to partial specifications written using assertions (in the program itself and/or in system libraries), to generate and simplify run-time tests, and to perform high-level program transformations such as multiple abstract specialization, parallelization, and resource usage control, all in a provably correct way. In the case of validation and debugging, the assertions can refer to a variety of program points such as procedure entry, procedure exit, points within procedures, or global computations. The system can reason with much richer information than, for example, traditional types. This includes data structure shape (including pointer sharing), bounds on data structure sizes, and other operational variable instantiation properties, as well as procedure-level properties such as determinacy, termination, non-failure, and bounds on resource consumption (time or space cost).

Towards description and optimization of abstract machines in an extension of prolog

Competitive abstract machines for Prolog are usually large, intricate, and incorpórate sophisticated optimizations. This makes them difñcult to code, optimize, and, especially, maintain and extend. This is partly due to the fact that efñciency considerations make it necessary to use low-level languages in their implementation. Writing the abstract machine (and ancillary code) in a higher-level language can help harness this inherent complexity. In this paper we show how the semantics of basic components of an efficient virtual machine for Prolog can be described using (a variant of) Prolog which retains much of its semantics. These descriptions are then compiled to C and assembled to build a complete bytecode emulator. Thanks to the high level of the language used and its closeness to Prolog the abstract machine descriptions can be manipulated using standard Prolog compilation and optimization techniques with relative ease. We also show how, by applying program transformations selectively, we obtain abstract machine implementations whose performance can match and even exceed that of highly-tuned, hand-crafted emulators.

Abstract interpretation with specialized definitions

The relationship between abstract interpretation and partial evaluation has received considerable attention and (partial) integrations have been proposed starting from both the partial evaluation and abstract interpretation perspectives. In this work we present what we argüe is the first generic algorithm for efñcient and precise integration of abstract interpretation and partial evaluation from an abstract interpretation perspective. Taking as starting point state-of-the-art algorithms for context-sensitive, polyvariant abstract interpretation and (abstract) partial evaluation of logic programs, we present an algorithm which combines the best of both worlds. Key ingredients include the accurate success propagation inherent to abstract interpretation and the powerful program transformations achievable by partial deduction. In our algorithm, the calis which appear in the analysis graph are not analyzed w.r.t. the original definition of the procedure but w.r.t. specialized definitions of these procedures. Such specialized definitions are obtained by applying both unfolding and abstract executability. Also, our framework is parametric w.r.t. different control strategies and abstract domains. Different combinations of these parameters correspond to existing algorithms for program analysis and specialization. Our approach efficiently computes strictly more precise results than those achievable by each of the individual techniques. The algorithm is one of the key components of CiaoPP, the analysis and specialization system of the Ciao compiler.

Determinacy analysis for logic programs using mode and type information

We propose an analysis for detecting procedures and goals that are deterministic (i.e. that produce at most one solution), or predicates whose clause tests are mutually exclusive (which implies that at most one of their clauses will succeed) even if they are not deterministic (because they cali other predicates that can produce more than one solution). Applications of such determinacy information include detecting programming errors, performing certain high-level program transformations for improving search efñciency, optimizing low level code generation and parallel execution, and estimating tighter upper bounds on the computational costs of goals and data sizes, which can be used for program debugging, resource consumption and granularity control, etc. We have implemented the analysis and integrated it in the CiaoPP system, which also infers automatically the mode and type information that our analysis takes as input. Experiments performed on this implementation show that the analysis is fairly accurate and efncient.

Multivariant non-failure analysis via standard abstract interpretation

Non-failure analysis aims at inferring that predicate calis in a program will never fail. This type of information has many applications in functional/logic programming. It is essential for determining lower bounds on the computational cost of calis, useful in the context of program parallelization, instrumental in partial evaluation and other program transformations, and has also been used in query optimization. In this paper, we re-cast the non-failure analysis proposed by Debray et al. as an abstract interpretation, which not only allows to investígate it from a standard and well understood theoretical framework, but has also several practical advantages. It allows us to incorpórate non-failure analysis into a standard, generic abstract interpretation engine. The analysis thus benefits from the fixpoint propagation algorithm, which leads to improved information propagation. Also, the analysis takes advantage of the multi-variance of the generic engine, so that it is now able to infer sepárate non-failure information for different cali patterns. Moreover, the implementation is simpler, and allows to perform non-failure and covering analyses alongside other analyses, such as those for modes and types, in the same framework. Finally, besides the precisión improvements and the additional simplicity, our implementation (in the Ciao/CiaoPP multiparadigm programming system) also shows better efRciency.

Combined static and dynamic assertion-based debugging of constraint logic programs

We propose a general framework for assertion-based debugging of constraint logic programs. Assertions are linguistic constructions for expressing properties of programs. We define several assertion schemas for writing (partial) specifications for constraint logic programs using quite general properties, including user-defined programs. The framework is aimed at detecting deviations of the program behavior (symptoms) with respect to the given assertions, either at compile-time (i.e., statically) or run-time (i.e., dynamically). We provide techniques for using information from global analysis both to detect at compile-time assertions which do not hold in at least one of the possible executions (i.e., static symptoms) and assertions which hold for all possible executions (i.e., statically proved assertions). We also provide program transformations which introduce tests in the program for checking at run-time those assertions whose status cannot be determined at compile-time. Both the static and the dynamic checking are provably safe in the sense that all errors flagged are definite violations of the pecifications. Finally, we report briefly on the currently implemented instances of the generic framework.

A framework for assertion-based debugging in constraint logic programming

We propose a general framework for assertion-based debugging of constraint logic programs. Assertions are linguistic constructions which allow expressing properties of programs. We define assertion schemas which allow writing (partial) specifications for constraint logic programs using quite general properties, including user-defined programs. The framework is aimed at detecting deviations of the program behavior (symptoms) with respect to the given assertions, either at compile-time or run-time. We provide techniques for using information from global analysis both to detect at compile-time assertions which do not hold in at least one of the possible executions (i.e., static symptoms) and assertions which hold for all possible executions (i.e., statically proved assertions). We also provide program transformations which introduce tests in the program for checking at run-time those assertions whose status cannot be determined at compile-time. Both the static and the dynamic checking are provably safe in the sense that all errors flagged are definite violations of the specifications. Finally, we report on an implemented instance of the assertion language and framework.

Estimating the computational cost of logic programs

Information about the computational cost of programs is potentially useful for a variety of purposes, including selecting among different algorithms, guiding program transformations, in granularity control and mapping decisions in parallelizing compilers, and query optimization in deductive databases. Cost analysis of logic programs is complicated by nondeterminism: on the one hand, procedures can return múltiple Solutions, making it necessary to estímate the number of solutions in order to give nontrivial upper bound cost estimates; on the other hand, the possibility of failure has to be taken into account while estimating lower bounds. Here we discuss techniques to address these problems to some extent.

A generic framework for the analysis and specialization of logic programs

The relationship between abstract interpretation and partial deduction has received considerable attention and (partial) integrations have been proposed starting from both the partial deduction and abstract interpretation perspectives. In this work we present what we argüe is the first fully described generic algorithm for efñcient and precise integration of abstract interpretation and partial deduction. Taking as starting point state-of-the-art algorithms for context-sensitive, polyvariant abstract interpretation and (abstract) partial deduction, we present an algorithm which combines the best of both worlds. Key ingredients include the accurate success propagation inherent to abstract interpretation and the powerful program transformations achievable by partial deduction. In our algorithm, the calis which appear in the analysis graph are not analyzed w.r.t. the original definition of the procedure but w.r.t. specialized definitions of these procedures. Such specialized definitions are obtained by applying both unfolding and abstract executability. Our framework is parametric w.r.t. different control strategies and abstract domains. Different combinations of such parameters correspond to existing algorithms for program analysis and specialization. Simultaneously, our approach opens the door to the efñcient computation of strictly more precise results than those achievable by each of the individual techniques. The algorithm is now one of the key components of the CiaoPP analysis and specialization system.

An integration of partial evaluation in a generic abstract interpretation framework

Information generated by abstract interpreters has long been used to perform program specialization. Additionally, if the abstract interpreter generates a multivariant analysis, it is also possible to perform múltiple specialization. Information about valúes of variables is propagated by simulating program execution and performing fixpoint computations for recursive calis. In contrast, traditional partial evaluators (mainly) use unfolding for both propagating valúes of variables and transforming the program. It is known that abstract interpretation is a better technique for propagating success valúes than unfolding. However, the program transformations induced by unfolding may lead to important optimizations which are not directly achievable in the existing frameworks for múltiple specialization based on abstract interpretation. The aim of this work is to devise a specialization framework which integrates the better information propagation of abstract interpretation with the powerful program transformations performed by partial evaluation, and which can be implemented via small modifications to existing generic abstract interpreters. With this aim, we will relate top-down abstract interpretation with traditional concepts in partial evaluation and sketch how the sophisticated techniques developed for controlling partial evaluation can be adapted to the proposed specialization framework. We conclude that there can be both practical and conceptual advantages in the proposed integration of partial evaluation and abstract interpretation.

Optimization of logic programs with dynamic scheduling

Dynamic scheduling increases the expressive power of logic programming languages, but also introduces some overhead. In this paper we present two classes of program transformations designed to reduce this additional overhead, while preserving the operational semantics of the original programs, modulo ordering of literals woken at the same time. The first class of transformations simplifies the delay conditions while the second class moves delayed literals later in the rule body. Application of the program transformations can be automated using information provided by compile-time analysis. We provide experimental results obtained from an implementation of the proposed techniques using the CIAO prototype compiler. Our results show that the techniques can lead to substantial performance improvement.

Operational Aspects of Full Reduction in Lambda Calculi

Esta tesis estudia la reducción plena (‘full reduction’ en inglés) en distintos cálculos lambda. 1 En esencia, la reducción plena consiste en evaluar los cuerpos de las funciones en los lenguajes de programación funcional con ligaduras. Se toma el cálculo lambda clásico (i.e., puro y sin tipos) como el sistema formal que modela el paradigma de programación funcional. La reducción plena es una técnica fundamental cuando se considera a los programas como datos, por ejemplo para la optimización de programas mediante evaluación parcial, o cuando algún atributo del programa se representa a su vez por un programa, como el tipo en los demostradores automáticos de teoremas actuales. Muchas semánticas operacionales que realizan reducción plena tienen naturaleza híbrida. Se introduce formalmente la noción de naturaleza híbrida, que constituye el hilo conductor de todo el trabajo. En el cálculo lambda la naturaleza híbrida se manifiesta como una ‘distinción de fase’ en el tratamiento de las abstracciones, ya sean consideradas desde fuera o desde dentro de si mismas. Esta distinción de fase conlleva una estructura en capas en la que una semántica híbrida depende de una o más semánticas subsidiarias. Desde el punto de vista de los lenguajes de programación, la tesis muestra como derivar, mediante técnicas de transformación de programas, implementaciones de semánticas operacionales que reducen plenamente a partir de sus especificaciones. Las técnicas de transformación de programas consisten en transformaciones sintácticas que preservan la equivalencia semántica de los programas. Se ajustan las técnicas de transformación de programas existentes para trabajar con implementaciones de semánticas híbridas. Además, se muestra el impacto que tiene la reducción plena en las implementaciones que utilizan entornos. Los entornos son un ingrediente fundamental en las implementaciones realistas de una máquina abstracta. Desde el punto de vista de los sistemas formales, la tesis desvela una teoría novedosa para el cálculo lambda con paso por valor (‘call-by-value lambda calculus’ en inglés) que es consistente con la reducción plena. Dicha teoría induce una noción de equivalencia observacional que distingue más puntos que las teorías existentes para dicho cálculo. Esta contribución ayuda a establecer una ‘teoría estándar’ en el cálculo lambda con paso por valor que es análoga a la ‘teoría estándar’ del cálculo lambda clásico propugnada por Barendregt. Se presentan resultados de teoría de la demostración, y se sugiere como abordar el estudio de teoría de modelos. ABSTRACT This thesis studies full reduction in lambda calculi. In a nutshell, full reduction consists in evaluating the body of the functions in a functional programming language with binders. The classical (i.e., pure untyped) lambda calculus is set as the formal system that models the functional paradigm. Full reduction is a prominent technique when programs are treated as data objects, for instance when performing optimisations by partial evaluation, or when some attribute of the program is represented by a program itself, like the type in modern proof assistants. A notable feature of many full-reducing operational semantics is its hybrid nature, which is introduced and which constitutes the guiding theme of the thesis. In the lambda calculus, the hybrid nature amounts to a ‘phase distinction’ in the treatment of abstractions when considered either from outside or from inside themselves. This distinction entails a layered structure in which a hybrid semantics depends on one or more subsidiary semantics. From a programming languages standpoint, the thesis shows how to derive implementations of full-reducing operational semantics from their specifications, by using program transformations techniques. The program transformation techniques are syntactical transformations which preserve the semantic equivalence of programs. The existing program transformation techniques are adjusted to work with implementations of hybrid semantics. The thesis also shows how full reduction impacts the implementations that use the environment technique. The environment technique is a key ingredient of real-world implementations of abstract machines which helps to circumvent the issue with binders. From a formal systems standpoint, the thesis discloses a novel consistent theory for the call-by-value variant of the lambda calculus which accounts for full reduction. This novel theory entails a notion of observational equivalence which distinguishes more points than other existing theories for the call-by-value lambda calculus. This contribution helps to establish a ‘standard theory’ in that calculus which constitutes the analogous of the ‘standard theory’ advocated by Barendregt in the classical lambda calculus. Some prooftheoretical results are presented, and insights on the model-theoretical study are given.