一句话：提出基于 BDD 的度量来比较 t-wise 采样质量，并给出无计数原子集算法以支持高效确定性 BDD 编译。
可配置系统测试中，故障常由多个特性交互引发，t-wise 采样技术通过生成配置集覆盖 t 个特性的有效交互，但现有技术难以扩展到约 18000 个特征的大规模系统，且不同采样技术产生的样本各异，难以选择。本文引入基于二元决策图（BDD）的度量来评估样本质量，并利用度量特征指导样本质量提升。作为 BDD 工作的基石，作者还提出一种无计数算法来计算原子集（约束总是被一起选择或取消选择的特征），作为预处理步骤，以实现高效且确定性的 BDD 编译。该工作旨在为大规模可配置系统的测试样本比较与选择提供依据。
Testing configurable systems is challenging, as faults often result from interactions between multiple features. Therefore, quality assurance also requires to test configurations including such interactions. T-wise sampling techniques generate sets of configurations to cover each valid interaction among t features. However, state-of-the-art techniques fail to scale to large systems, such as automotive02_v4 with ≈ 18, 000 features. Moreover, the diversity of sampling techniques results in a variety of different samples, raising the question of which sample to use for testing. We aim to tackle this question by introducing a Binary Decision Diagram (BDD)-based metric for assessing sample quality. The characteristics of the metric can further be used as a guide for ideas to enhance a sample’s quality. As a cornerstone of our BDD-based work, we introduce a counting free algorithm for computing the atomic sets—features that are constrained to always be (de)selected together—serving as a preprocessing step for efficient and deterministic BDD compilation.