Incremental Attribute Evaluation for Multi-User Semantics-Based Editors

Micallef, Josephine

This thesis addresses two fundamental problems associated with performing incremental attribute evaluation in multi-user editors based on the attribute grammar formalism: (1) multiple asynchronous modifications of the attributed derivation tree, and (2) segmentation of the tree into separate modular units. Solutions to these problems make it possible to construct semantics-based editors for use by teams of programmers developing or maintaining large software systems. Multi-user semantics based editors improve software productivity by reducing communication costs and snafus. The objectives of an incremental attribute evaluation algorithm for multiple asynchronous changes are that (a) all attributes of the derivation tree have correct values when evaluation terminates, and (b) the cost of evaluating attributes necessary to reestablish a correctly attributed derivation tree is minimized. We present a family of algorithms that differ in how they balance the tradeoff between algorithm efficiency and expressiveness of the attribute grammar. This is important because multi-user editors seem a practical basis for many areas of computer-supported cooperative work, not just programming. Different application areas may have distinct definitions of efficiency, and may impose different requirements on the expressiveness of the attribute grammar. The characteristics of the application domain can then be used to select the most efficient strategy for each particular editor. To address the second problem, we define an extension of classical attribute grammars that allows the specification of interface consistency checking for programs composed of many modules. Classical attribute grammars can specify the static semantics of monolithic programs or modules, but not inter-module semantics; the latter was done in the past using ad hoc techniques. Extended attribute grammars support programming-in-the-large constructs found in real programming languages, including textual inclusion, multiple kinds of modular units and nested modular units. We discuss attribute evaluation in the context of programming-in-the-large, particularly the separation of concerns between the local evaluator for each modular unit and the global evaluator that propagates attribute flows across module boundaries. The result is a uniform approach to formal specification of both intra-module and inter-module static semantic properties, with the ability to use attribute evaluation algorithms to carry out a complete static semantic analysis of a multi-module program.



More About This Work

Academic Units
Computer Science
Department of Computer Science, Columbia University
Columbia University Computer Science Technical Reports, CUCS-023-91
Published Here
March 17, 2012