Peer to Peer Optimistic Collaborative Editing on XML-like trees

TL;DR

This paper investigates conflict resolution in peer-to-peer collaborative editing of XML-like trees, demonstrating the conditions under which convergence can be guaranteed and proposing a new set of operations that satisfy key correctness properties.

Contribution

It introduces a new set of operations for XML-like trees that satisfy convergence properties TP1 and TP2, advancing P2P collaborative editing without locks.

Findings

No transformation satisfying TP1 and TP2 exists for initial operations.

A richer set of operations can satisfy TP1 and TP2.

The approach can be combined with string editing processes for convergence.

Abstract

Collaborative editing consists in editing a common document shared by several independent sites. This may give rise to conficts when two different users perform simultaneous uncompatible operations. Centralized systems solve this problem by using locks that prevent some modifications to occur and leave the resolution of confict to users. On the contrary, peer to peer (P2P) editing doesn't allow locks and the optimistic approach uses a Integration Transformation IT that reconciliates the conficting operations and ensures convergence (all copies are identical on each site). Two properties TP1 and TP2, relating the set of allowed operations Op and the transformation IT, have been shown to ensure the correctness of the process. The choice of the set Op is crucial to define an integration operation that satisfies TP1 and TP2. Many existing algorithms don't satisfy these properties and are indeed incorrect i.e. convergence is not guaranteed. No algorithm enjoying both properties is known for strings and little work has been done for XML trees in a pure P2P framework (that doesn't use time-stamps for instance). We focus on editing unranked unordered labeled trees, so-called XML-like trees that are considered for instance in the Harmony pro ject. We show that no transformation satisfying TP1 and TP2 can exist for a first set of operations but we show that TP1 and TP2 hold for a richer set of operations. We show how to combine our approach with any convergent editing process on strings (not necessarily based on integration transformation) to get a convergent process.