Automatic layout of railroad diagrams

Abstract

Railroad diagrams (also called "syntax diagrams") are a common, intuitive visualization of grammars, but limited tooling and a lack of formal attention to their layout mostly confines them to hand-drawn documentation. We present the first formal treatment of railroad diagram layout along with a principled, practical implementation. We characterize the problem as compiling a *diagram language* (specifying conceptual components and how they connect and compose) to a *layout language* (specifying basic graphical shapes and their sizes and positions). We then implement a compiler that performs *line wrapping* to meet a target width, as well as vertical *alignment* and horizontal *justification per user-specified policies. We frame line wrapping as optimization, where we describe principled dimensions of optimality and implement corresponding heuristics. For front-end evaluation, we show that our diagram language is well-suited for common applications by describing how regular expressions and Backus-Naur form can be compiled to it. For back-end evaluation, we argue that our compiler is practical by comparing its output to diagrams laid out by hand and by other tools.