Conditional evaluation in simple multi-visit attribute-grammar evaluators

Attribute grammars are a formalism for specifying computations on context-free languages. Due to the non-strictness of the if constructs in attribution equations, it is possible to avoid evaluating certain attribute instances in a syntax tree. A dynamic evaluator can easily avoid such useless computations with a demand-driven approach. However, dynamic evaluators are not efficient because they need to keep the attribute dependence graph during evaluation, and they need to decide an evaluation order for each syntax tree. In contrast, a visit-oriented (static) evaluator can carefully re-arrange the evaluation order and still avoid unnecessary computations. We propose such a technique in this paper.