Usage driven software tailoring

Overview & Synopsis

Applications manipulating tabular data models are ubiquitous, particularly in scientific domains such as agronomy. While flexible, these applications often rely on a clone-and-own evolution strategy, where models are duplicated to be adapted. Although this practice allows for rapid adaptation, it leads to model drift, significantly hindering traceability, maintenance, and evolution. Moreover, experimental processes, which evolve through iterative back-and-forth cycles and trial phases are sources of inconsistencies within models.This thesis proposes a high-level, domain-oriented modeling approach to represent tabular data. Coupled with an action model that captures end-user performed adaptation and cloning operations, this approach provides systematic support for the variability introduced during model evolution. Each modification to a tabular model is represented as a set of formalized actions, recorded in a trace. This trace captures the history of changes at a business level, enabling explicit reasoning over the differences between multiple versions of a model.Reasoning on the trace of model construction also facilitates the identification and correction of inconsistencies introduced through successive cloning and adaptation. These contributions were qualitatively validated through expert interviews and scenario-based evaluations using a prototype implementation. A quantitative evaluation on the scalability of the prototype complements this validation.Thus, this approach enables the control of variability in systems handling tabular data, while preserving the flexibility required by end-users.