Blame Trees

Abstract

We consider the problem of merging individual text documents, motivated by the single-file merge algorithms of document-based version control systems. Abstracting away the merging of conflicting edits to an external conflict resolution function (possibly implemented by a human), we consider the efficient identification of conflicting regions. We show how to implement tree-based document representation to quickly answer a data structure inspired by the “blame” query of some version control systems. A “blame” query associates every line of a document with the revision in which it was last edited. Our tree uses this idea to quickly identify conflicting edits. We show how to perform a merge operation in time proportional to the sum of the logarithms of the shared regions of the documents, plus the cost of conflict resolution. Our data structure is functional and therefore confluently persistent, allowing arbitrary version DAGs as in real version-control systems. Our results rely on concurrent traversal of two trees with short circuiting when shared subtrees are encountered.