The Pan language-based editing system

Reviewer: Paul W. Abrahams

An editor is a programmer's most vital tool. Programmers have strong feelings about the editors they use and are sensitive to the quality of those editors. The Pan editor is designed for experienced programmers—software professionals proficient with their primary tools, intimately familiar with their programming languages, and skilled at creating programs in those languages. Judging by the description in this paper, Pan ought to be a programmer's delight. Pan is a syntax-recognizing editor—it combines the freedom and flexibility of a purely textual editor with the contextual awareness of a syntax-directed editor. Its linguistic knowledge is derived from a collection of language descriptions that include both syntactic and semantic information, so it is not inherently limited to a particular language. It can perform syntactic analysis even when the program text is syntactically ill-formed; it views such a program as merely a variant of a correct program, treating the variance as no more important than a spelling error and maintaining a syntactic view of those parts of the document that are not affected by the variance. It provides many of the same key bindings and text services as Emacs, including generalized undo, kill rings, text filling, customization, extension, and self-documentation. Even at the textual level, it has advantages over Emacs; for instance, it provides visual highlighting of selected regions. Lexical and syntactic analysis is handled by the Ladle (Language Description Language) component of Pan, while the checking of contextual semantic constraints is handled by the Colander (Constraint Language and Interpreter) component. Both components have two parts: an online part that is part of the editor itself and an offline part that generates the tables needed by the online part. An editing interface coordinates the analysis and makes derived information accessible to users and client programs. Ladle represents a language grammar as an abstract syntax augmented by additional productions that make it possible to derive the concrete syntax from the abstract syntax. The two syntaxes are related by grammatical abstraction, which ensures that a straightforward and reversible transformation exists between them. Grammatical abstraction is purely structural; it does not use semantic information to identify corresponding structures. Colander is based on the notion of logical constraint grammars, expressed in a Prolog-like language. Using the abstract syntax as a base, Colander provides for annotating production with goals (logical propositions) that must be satisfied if the document is semantically correct. An incremental evaluator monitors changes to the document and the derived information in Colander's database in order to maintain consistency between them. Users typically understand document structure in terms of structural components rather than syntax trees, naming those components using the specific terminology of the document's language. Pan therefore uses operand classes to designate kinds of document components. An operand class is an arbitrary collection of document components; operand classes may overlap. Operand classes define the textual units used for structure-oriented selection, navigation, highlighting, and editing. Expression and Statement are examples of language-specific operand classes. Examples of language-independent operand classes are Syntactic Error and Query Result . Text-oriented and structure-oriented operations combine naturally; for instance, you can move to the next expression and then make arbitrary textual changes to that expression. Despite my favorable impression of Pan, I did not find this paper easy reading. The description of Colander and its constraint grammars, which relies on a knowledge of Prolog, is particularly impenetrable; although I got the general idea of what was going on, I could not make sense of the details. Some of my difficulty was caused by the use of unexplained notations that are not standard Prolog. It is too bad that the authors were not more careful about defining their terminology and notation, since the content of the paper is so clearly worthwhile.