Davide Ancona's most recent papers

We propose a novel approach to constraint-based type inference based on coinductive logic programming. That is, constraint generation corresponds to translation into a conjunction of Horn clauses P, and constraint satisfaction is defined in terms of the maximal coinductive Herbrand model of P. We illustrate the approach by formally defining this translation for a small object-oriented language similar to Featherweight Java, where type annotations in field and method declarations can be omitted. In this way, we obtain a very precise type inference and provide new insights into the challenging problem of type inference for object-oriented programs. Since the approach is deliberately declarative, we define in fact a formal specification for a general class of algorithms, which can be a useful road maps to researchers. Moreover, despite we consider here a particular language, the methodology could be used in general for providing abstract specifications of type inference for different kinds of programming languages.

We propose a novel approach based on coinductive logic to specify type systems of programming languages. The approach consists in encoding programs in Horn formulas which are interpreted w.r.t. their coinductive Herbrand model. We illustrate the approach by first specifying a standard type system for a small object-oriented language similar to Featherweight Java. Then we define an idealized type system for a variant of the language where type annotations can be omitted. The type system involves infinite terms and proof trees not representable in a finite way, thus providing a theoretical limit to type inference of object-oriented programs, since only sound approximations of the system can be implemented. Approximation is naturally captured by the notions of subtyping and subsumption; indeed, rather than increasing the expressive power of the system, as it usually happens, here subtyping is needed for approximating infinite non regular types and proof trees with regular ones.

This paper discusses an application of intelligent software agents and ontologies to solve the problem of semi-automatic porting of Java programs. We have designed a system for aiding users to adapt Java code in a type- and meaning-safe way, when an application has to migrate to new libraries which are not fully compatible with the legacy ones. To achieve this, we propose an approach based on an integration of the two type-theoretic notions of subtyping and type isomorphism with ontology matching. While the former notions are needed to ensure flexible adaptation in the presence of type-safety, the latter supports the user to preserve the meaning of names that appear in the program to be adapted. Intelligent agents control the different components of the system and interact with other agents in order to provide the final user with the semi-automatic porting service he/she required.

The Common Language Infrastructure (CLI) is a virtual machine expressly designed for implementing statically typed languages such as C#, therefore programs written in dynamically typed languages are typically much slower than C# when executed on .NET. Recent developments show that Just In Time (JIT) compilers can exploit runtime type information to generate quite efficient code. Unfortunately, writing a JIT compiler is far from being simple. In this paper we report our positive experience with automatic generation of JIT compilers as supported by the PyPy infrastructure, by focusing on JIT compilation for .NET. Following this approach, we have in fact added a second layer of JIT compilation, by allowing dynamic generation of more efficient .NET bytecode, which in turn can be compiled to machine code by the .NET JIT compiler. The main and novel contribution of this paper is to show that this two-layers JIT technique is effective, since programs written in dynamic languages can run on .NET as fast as (and in some cases even faster than) the equivalent C# programs. The practicality of the approach is demonstrated by showing some promising experiments done with benchmarks written in a simple dynamic language.

Writing an optimizing static compiler for dynamic languages is not an easy task, since quite complex static analysis is required. On the other hand, recent developments show that JIT compilers can exploit runtime type information to generate quite efficient code. Unfortunately, writing a JIT compiler is far from being simple. In this paper we report our positive experience with automatic generation of JIT compilers as supported by the PyPy infrastructure, by focusing on JIT compilation for .NET. The paper presents two main and novel contributions: we show that partial evaluation can be used in practice for generating a JIT compiler, and we experiment with the potentiality offered by the ability to add a further level of JIT compilation on top of .NET. The practicality of the approach is demonstrated by showing some promising experiments done with benchmarks written in a simple dynamic language.

This paper proposes a solution to the problem of semi-automatic porting of Java programs. In particular, our work aims at the design of tools able to aid users to adapt Java code in a type-safe way, when an application has to migrate to new libraries which are not fully compatible with the legacy ones. To achieve this, we propose an approach based on an integration of the two type-theoretic notions of subtyping and type isomorphism with ontology matching. While the former notions are needed to ensure flexible adaptation in the presence of type-safety, the latter supports the user to preserve the semantics of the program to be adapted.

Although coinductive logic programming (Co-LP) has proved to have several applications, including verification of infinitary properties, model checking and bisimilarity proofs, type inference via Co-LP has not been properly investigated yet. In this paper we show a novel approach to solve the problem of type inference in the context of Java-like languages, that is, object-oriented languages based on nominal subtyping. The proposed approach follows a generic scheme: first, the program P to be analyzed is translated into an approximating logic program P' and a goal G; then, the solution of the type inference problem corresponds to find an instantiation of the goal G which belongs to the greatest model of P', that is, the coinductive model of P'. Operationally, this corresponds to find a co-SLD derivation of G in P', according to the operational semantics of Co-LP recently defined by Simon et al. [ICLP06,ICALP07]. A complete formalization of an instantiation of this scheme is considered for a simple object-oriented language and a corresponding type soundness theorem is stated. A prototype implementation based on a meta-interpreter of co-SLD has been implemented. Finally, we address scalability issues of the approach, by sketching an instantiation able to deal with a much more expressive object-oriented language.

We define a framework of components based on Java-like languages, where components are binary mixin modules. Basic components can be obtained from a collection of classes by compiling such classes in isolation; for allowing that, requirements in the form of type constraints are associated with each class. Requirements are specified by the user who, however, is assisted by the compiler that can generate missing constraints essential to guarantee type safety. Basic components can be composed together by using a set of expressive typed operators; thanks to soundness results, such a composition is always type safe. The framework is designed as a separate layer that can be instantiated on top of any Java-like language; to show the effectiveness of the approach, an instantiation on a small Java subset is provided, together with a prototype implementation. Besides safety, the approach achieves great flexibility in reusing components for two reasons: (1) type constraints generated for a single component exactly capture all possible contexts where it can be safely used; (2) composition of components is not limited to conventional linking, but is achieved by means of a set of powerful operators typical of mixin modules.

We present a simple parametric calculus of processes which exchange open mobile code, that is, code which may contain free variables to be bound by the receiver's code. Type safety is ensured by a combination of static and dynamic checks. That is, internal consistency of each process is statically verified, by relying on local type assumptions on missing code; then, when code is sent from a process to another, a runtime check based on a subtyping relation ensures that it can be successfully received, without requiring re-inspection of the code. In order to refuse communication in as few cases as possible, the runtime check accepts even mobile code which would be rejected if statically available, by automatically inserting coercions driven by the subtyping relation, as in the so-called Penn translation. The calculus is parametric in some ingredients which can vary depending on the specific language or system. Notably, we abstract away from the specific nature of the code to be exchanged, and of the static and dynamic checks. We formalize the notion of type safety in our general framework and provide sufficient conditions on the above ingredients which guarantee this property. We illustrate our approach on a simple lambda-calculus with records, where type safe exchange of mobile code is made problematic by conflicts due to components which were not explicitly required. In particular, we show that the standard coercion semantics given in the literature, with other aims, for this calculus, allows to detect and eliminate conflicts due to inner components, thus solving a problem which was left open in previous work on type-safe exchange of mobile code.

We present a translation from Fickle, a small object-oriented language allowing objects to change their class at runtime, into Java. The translation is provenly correct in the sense that it preserves the static and dynamic semantics. Moreover, it is compatible with separate compilation, since the translation of a Fickle class does not depend on the implementation of used classes. Based on the formal system, we have developed an implementation. The translation turned out to be a more subtle problem than we expected. In this article, we discuss four possible approaches we considered for the design of the translation and to justify our choice, we present formally the translation and proof of preservation of the static and dynamic semantics, and discuss the prototype implementation. Moreover, we outline an alternative translation based on generics that avoids most of the casts (but not all) needed in the previous translation. The language Fickle has undergone and is still undergoing several phases of development. In this article we are discussing the translation of FickleII.

In languages like C++, Java and C#, typechecking algorithms require methods to be annotated with their parameter and result types, which are either fixed or constrained by a bound. We show that, surprisingly enough, it is possible to infer the polymorphic type of a method where parameter and result types are left unspecified, as happens in most functional languages. These types intuitively capture the (less restrictive) requirements on arguments needed to safely apply the method. We formalize our ideas on a minimal Java subset, for which we define a type system with polymorphic types and prove its soundness. We then describe an algorithm for type inference and prove its soundness and completeness. A prototype implementing inference of polymorphic types is available.

We define a general framework for compositional compilation, meant as the ability of building an executable application by separate compilation and linking of single fragments, opposed to global compilation of the complete source application code. More precisely, compilation of a source code fragment in isolation generates a corresponding binary fragment equipped with type information, formally modeled as a typing, allowing type safe linking of fragments without re-inspecting code. We formally define a notion of soundness and completeness of compositional compilation w.r.t. global compilation, and show how linking can be in practice expressed by an entailment relation on typings. Then, we provide a sufficient condition on such entailment to ensure soundness and completeness of compositional compilation, and compare this condition with the principal typings property. Furthermore, we show that this entailment relation can often be modularly expressed by an entailment relation on type environments and a subtyping relation.

Although the C-based interpreter of Python is reasonably fast, implementations on the CLI or the JVM platforms offers some advantages in terms of robustness and interoperability. Unfortunately, because the CLI and JVM are primarily designed to execute statically typed, object-oriented languages, most dynamic language implementations cannot use the native bytecodes for common operations like method calls and exception handling; as a result, they are not able to take full advantage of the power offered by the CLI and JVM. We describe a different approach that attempts to preserve the flexibility of Python, while still allowing for efficient execution. This is achieved by limiting the use of the more dynamic features of Python to an initial, bootstrapping phase. This phase is used to construct a final RPython (Restricted Python) program that is actually executed. RPython is a proper subset of Python, is statically typed, and does not allow dynamic modification of class or method definitions; however, it can still take advantage of Python features such as mixins and first-class methods and classes. This paper presents an overview of RPython, including its design and its translation to both CLI and JVM bytecode. We show how the bootstrapping phase can be used to implement advanced features, like extensible classes and generative programming. We also discuss what work remains before RPython is truly ready for general use, and compare the performance of RPython with that of other approaches.

We define a framework of components based on Java-like languages, where components are binary mixin modules. Basic components can be obtained from a collection of classes by compiling such classes in isolation; for allowing that, requirements in the form of type constraints are associated with each class. Requirements are specified by the user who, however, is assisted by the compiler which can generate missing constraints essential to guarantee type safety. Basic components can be composed together by using a set of expressive typed operators; thanks to soundness results, such a composition is always type safe. The framework is designed as a separate layer which can be instantiated on top of any Java-like language; a prototype implementation is available for a small Java subset. Besides safety, the approach achieves great flexibility in reusing components for two reasons: (1) type constraints generated for a single component exactly capture all possible contexts where it can be safely used; (2) composition of components is not limited to conventional linking, but is achieved by means of a set of powerful operators typical of mixin modules.

We present SmartJavaMod, a language of mixin modules supporting compositional compilation, and constructed on top of the Java language. More in detail, this means that basic modules are collections of Java classes which can be typechecked in isolation, inferring constraints on missing classes and allowing safe reuse of the module in as many contexts as possible. Furthermore, it is possible to write structured module expressions by means of a set of module operators, and a type system at the module level ensures type safety, in the sense that we can always reduce a module expression to a well-formed collection of Java classes. What we obtain is a module language which is extremely flexible and allows the encoding (without any need of enriching the core level, that is, the Java language) of a variety of constructs supporting software reuse and extensibility.

We define compositional compilation as the ability to typecheck source code fragments in isolation, generate corresponding binaries, and link together fragments whose mutual assumptions are satisfied, without reinspecting the code. Even though compositional compilation is a highly desirable feature, in Java-like languages it can hardly be achieved. This is due to the fact that the bytecode generated for a fragment (say, a class) is not uniquely determined by its source code, but also depends on the compilation context. We propose a way to obtain compositional compilation for Java, by introducing a polymorphic form of bytecode containing type variables (ranging over class names) and equipped with a set of constraints involving type variables. Thus, polymorphic bytecode provides a representation for all the (standard) bytecode that can be obtained by replacing type variables with classes satisfying the associated constraints. We illustrate our proposal by developing a typing and a linking algorithm. The typing algorithm compiles a class in isolation generating the corresponding polymorphic bytecode fragment and constraints on the classes it depends on. The linking algorithm takes a collection of polymorphic bytecode fragments, checks their mutual consistency, and possibly simplifies and specializes them. In particular, linking a self-contained collection of fragments either fails, or produces standard bytecode (the same as what would have been produced by standard compilation of all fragments).

The notion of an intelligent agent as an entity which appears to be the subject of mental attitudes like beliefs, desires and intentions (hence, the BDI acronym) is well known and accepted by many researchers. Besides the definition of various BDI logics, many languages and integrated environments for programming BDI-style agents have been proposed since the early nineties. In this reasoned bibliography, nine languages and implemented systems, namely PRS, dMARS, JACK, JAM, Jadex, AgentSpeak(L), 3APL, Dribble, and Coo-BDI, are discussed and compared. References to other systems and languages based on the BDI model are also provided, as well as pointers to surveys dealing with related topics.

We define compositional compilation as the ability to typecheck source code fragments in isolation, generate corresponding binaries, and link together fragments whose mutual assumptions are satisfied, without reinspecting the code. Even though compositional compilation is a highly desirable feature, in Java-like languages it can hardly be achieved. This is due to the fact that the bytecode generated for a fragment (say, a class) is not uniquely determined by its source code, but also depends on the compilation context. We propose a way to obtain compositional compilation for Java, by introducing a polymorphic form of bytecode containing type variables (ranging over class names) and equipped with a set of constraints involving type variables. Thus, polymorphic bytecode provides a representation for all the (standard) bytecode that can be obtained by replacing type variables with classes satisfying the associated constraints. We illustrate our proposal by developing a typing and a linking algorithm. The typing algorithm compiles a class in isolation generating the corresponding polymorphic bytecode fragment and constraints on the classes it depends on. The linking algorithm takes a collection of polymorphic bytecode fragments, checks their mutual consistency, and possibly simplifies and specializes them. In particular, linking a self-contained collection of fragments either fails, or produces standard bytecode (the same as what would have been produced by standard compilation of all fragments).

Dynamic rebinding is the ability of changing the definitions of names at execution time. While dynamic rebinding is clearly useful in practice, and increasingly needed in modern systems, most programming languages provide only limited and ad-hoc mechanisms, and no adequate semantic understanding currently exists. Here, we provide a simple and powerful mechanism for dynamic rebinding by means of a calculus of mixin modules (mutually recursive modules allowing redefinition of components) where, differently from the traditional approach, module operations can be performed after selecting and executing a module component: in this way, execution can refer to virtual components, which can be rebound when module operators are executed. In particular, in our calculus module operations are performed on demand, when execution would otherwise get stuck. We provide a sound type system, which ensures that execution never tries to access module components which cannot become available, and show how the calculus can be used to encode a variety of real-world dynamic rebinding mechanisms.

Mainstream research in Web Services is currently looking at two main aspects, namely formally describing interactions among services, and finding and combining services. Much work made in the intelligent agents area is being applied to these issues. In this paper, we investigate the application of agent research to Web Services from a different perspective, that is, procedural learning. The final objective is to enable an adaptive system (an agent in our terminology) to discover or being fed with knowledge concerning how to solve a specific set of problems in a specific software or physical environment. Our work is a very preliminary step into the issue, with the main objective of assessing how current Web Services technology can support a component, described in terms of beliefs, desires and intentions, dynamically adapting its behaviour to new environments.

The first part of the paper gives a brief overview of meta-programming, in particular program generation, and its use in software development. The second part introduces a basic calculus, related to FreshML, that supports program generation (as described through examples and a translation of MetaML into it) and programming in-the-large (this is demonstrated by a translation of CMS into it).

Building on our previous work, we present a simple module calculus where execution steps of a module component can be interleaved with reconfiguration steps (that is, reductions at the module level), and where execution can partly control precedence between these reconfiguration steps. This is achieved by means of a low priority link operator which is only performed when a certain component, which has not been linked yet, is both available and really needed for execution to proceed, otherwise precedence is given to the outer operators. We illustrate the expressive power of this mechanism by a number of examples. We ensure soundness by combining a static type system, which prevents errors in applying module operators, and a dynamic check which raises a linkage error if the running program needs a component which cannot be provided by reconfiguration steps. In particular no linkage errors can be raised if all components are potentially available.