Minimizing LR(1) State Machines is NP-Hard

TL;DR

This paper proves that minimizing LR(1) state machines is NP-hard by reducing the node-coloring problem to the minimization task through an indirect and incremental reduction approach.

Contribution

It introduces a novel NP-hardness proof for LR(1) state machine minimization using indirect reduction and incremental construction techniques.

Findings

LR(1) minimization is NP-hard.

Reduction uses graph coloring problem.

Incremental grammar extension technique.

Abstract

LR(1) parsing was a focus of extensive research in the past 50 years. Though most fundamental mysteries have been resolved, a few remain hidden in the dark corners. The one we bumped into is the minimization of the LR(1) state machines, which we prove is NP-hard. It is the node-coloring problem that is reduced to the minimization puzzle. The reduction makes use of two technique: indirect reduction and incremental construction. Indirect reduction means the graph to be colored is not reduced to an LR(1) state machine directly. Instead, it is reduced to a context-free grammar from which an LR(1) state machine is derived. Furthermore, by considering the nodes in the graph to be colored one at a time, the context-free grammar is incrementally extended from a template context-free grammar that is for a two-node graph. The extension is done by adding new grammar symbols and rules. A minimized LR(1) machine can be used to recover a minimum coloring of the original graph.