This work addresses what we believe to be a central issue in the field of XML diff and merge computation: the mathematical modeling o-called editing deltas and the study of their formal abstract properties. We expect at least three outputs from this theoretical work: a common basis to compare performances of the various algorithms through a structural normalization of deltas, a universal and flexible patch application model and a clearer separation of patch and merge engine performance from delta generation performance. Moreover, this work could inspire technical approaches to combine heterogeneous engines thank to sound delta transformations. This short paper reports current results, discusses key points and outlines some perspectives.