Section: Research Program

Rule-based programming languages

Programming languages are formalisms used to describe programs, applications, or software which aim to be executed on a given hardware. In principle, any Turing complete language is sufficient to describe the computations we want to perform. However, in practice the choice of the programming language is important because it helps to be effective and to improve the quality of the software. For instance, a web application is rarely developed using a Turing machine or assembly language. By choosing an adequate formalism, it becomes easier to reason about the program, to analyze, certify, transform, optimize, or compile it. The choice of the programming language also has an impact on the quality of the software. By providing high-level constructs as well as static verifications, like typing, we can have an impact on the software design, allowing more expressiveness, more modularity, and a better reuse of code. This also improves the productivity of the programmer, and contributes to reducing the presence of errors.

The quality of a programming language depends on two main factors. First, the intrinsic design, which describes the programming model, the data model, the features provided by the language, as well as the semantics of the constructs. The second factor is the programmer and the application which is targeted. A language is not necessarily good for a given application if the concepts of the application domain cannot be easily manipulated. Similarly, it may not be good for a given person if the constructs provided by the language are not correctly understood by the programmer.

In the Pareo group we target a population of programmers interested in improving the long-term maintainability and the quality of their software, as well as their efficiency in implementing complex algorithms. Our privileged domain of application is large since it concerns the development of transformations. This ranges from the transformation of textual or structured documents such as XML, to the analysis and the transformation of programs and models. This also includes the development of tools such as theorem provers, proof assistants, or model checkers, where the transformations of proofs and the transitions between states play a crucial role. In that context, the expressiveness of the programming language is important. Indeed, complex encodings into low level data structures should be avoided, in contrast to high level notions such as abstract types and transformation rules that should be provided.

It is now well established that the notions of term and rewrite rule are two universal abstractions well suited to model tree based data types and the transformations that can be done upon them. Over the last ten years we have developed a strong experience in designing and programming with rule based languages  [28] , [20] , [18] . We have introduced and studied the notion of strategy  [19] , which is a way to control how the rules should be applied. This provides the separation which is essential to isolate the logic and to make the rules reusable in different contexts.

To improve the quality of programs, it is also essential to have a clear description of their intended behaviors. For that, the semantics of the programming language should be formally specified.

There is still a lot of progress to be done in these directions. In particular, rule based programming can be made even more expressive by extending the existing matching algorithms to context-matching or to new data structures such as graphs or polygraphs. New algorithms and implementation techniques have to be found to improve the efficiency and make the rule based programming approach effective on large problems. Separating the rules from the control is very important. This is done by introducing a language for describing strategies. We still have to invent new formalisms and new strategy primitives which are both expressive enough and theoretically well grounded. A challenge is to find a good strategy language we can reason about, to prove termination properties for instance.

On the static analysis side, new formalized typing algorithms are needed to properly integrate rule based programming into already existing host languages such as Java. The notion of traversal strategy merits to be better studied in order to become more flexible and still provide a guarantee that the result of a transformation is correctly typed.