Type-directed diffing of structured data

Abstract

The Unix diff utility that compares lines of text is used pervasively by version control systems. Yet certain changes to a program may be difficult to describe accurately in terms of modifications to individual lines of code. As a result, observing changes at such a fixed granularity may lead to unnecessary conflicts between different edits. This paper presents a generic representation for describing transformations between algebraic data types and a non-deterministic algorithm for computing such representations. These representations can be used to give a more accurate account of modifications made to algebraic data structures - and the abstract syntax trees of computer programs in particular - as opposed to only considering modifications between their textual representations.