Coordinated incremental attribute evaluation on a DR-threaded tree

Abstract

Algorithms are presented for coordinated Reps-style incremental evaluation on a limited form of graph substrate, a DR-threaded tree. The substrate is essentially a tree with threads connecting known ‘divergence’ nodes to known ’receptor’ nodes. An arbitrary number of threads are connected to a divergence node, but only one is connected to each receptor node, which must be a leaf. Attribute propagation across the threads is restricted to copying in one direction (divergence to receptor). Substrates of this type arise in a syntax-directed editor that supports naming semantics, where declarations=divergence nodes and references=receptors.

Coordination is provided by constructing an interdependency covering graph, an attribute dependency graph consisting of fragments of actual dependency graphs connected by edges that represent approximate nonlocal dependencies. The ICG represents all possible dependency paths that can diverge from the modified attributes associated with a set of modification sites. ICG coordinated evaluation provides correct scheduling of attribute evaluations with a minimal amount of actual dependency analysis.