Quality-Optimised Differencing Algorithms for Models

在现代基于模型的开发过程中，模型是操作软件的一部分。它们扮演源代码的角色，并且像源代码一样，经常进行版本控制。因此，需要强大的工具来比较、合并和修补模型。这些工具基于计算模型之间差异的算法。模型的差分算法必须考虑到每种模型类型的结构和典型编辑操作的特点。他们应该尽可能地解释模型之间的差异，换句话说，产生高质量的差异。然而，提高差异的质量会使已经很高的响应时间恶化，因此必须在质量和成本之间做出妥协。目前还缺乏确定差异质量的系统方法。因此，本项目旨在定义模型差异质量的概念和度量，并在运行时间和实现的差异质量方面优化差异算法。在资助的第一阶段，该项目开发了创建基准的方法和工具，可用于评估不同的算法。创建了大量的测试数据集，主要是类图，并且使用这些测试模型评估了算法的各种变体。与此同时，新的差分算法出现了，它通过使用重构和其他复杂的操作来报告变化，从而提高了差分的质量。在初步结果的基础上，计划在第三个项目年制定一个新的概念框架，根据各类差异工具的要求区分差异质量的各种变体。计划重新评估类图的差分算法，并首次评估BPMN模型的差分算法，这些算法显示出明显不同的特征，并且已经提出了许多重构。

In modern model-based development processes, models are part of the operating software. They assume the role of source code and they are, like source code, versioned frequently. Consequently, powerful tools for comparing, merging, and patching models are required. Such tools are based on algorithms which calculate the difference between models. Differencing algorithms for models must take the structure of each model type and the characteristics of typical edit operations into account. They should explain the difference between models as understandably as possible, in other words, produce differences of high quality. However, improving the quality of differences deteriorates the already high response times, so one has to compromise between quality and cost. Systematic approaches to defining the quality of differences are currently missing. This project thus aims at defining notions and measures for the quality of model differences and at optimizing differencing algorithms both in terms of runtime as well as the achieved quality of the differences. In the first period of funding, the project has developed methods and tools for creating benchmarks which can be used to evaluate differencing algorithms. A substantial set of test data, mainly class diagrams, was created and various variants of algorithms have been evaluated using these test models. In the meantime, new differencing algorithms became available which enhance the quality of differences by using refactorings and other complex operations to report changes. On the basis of the preliminary results the third project year is planned to develop a new conceptual framework which distinguishes variants of the quality of differences depending on the requirements of various classes of differencing tools. It is planned to re-evaluate differencing algorithms for class diagrams and to evaluate for the first time algorithms for differencing algorithms for BPMN models, which show significantly different characteristics and for which many refactorings have been proposed.