In continuous testing, developers execute automated test cases once or even several times per day to ensure the quality of the integrated code. Although continuous testing helps ensure the quality of the code and reduces maintenance effort, it also significantly increases test execution overhead. In this paper, we empirically evaluate the effectiveness of test impact analysis from the perspective of code dependencies in the continuous testing setting. We first applied test impact analysis to one year of software development history in 11 large-scale open-source systems. We found that even though the number of changed files is small in daily commits (median ranges from 3 to 28 files), around 50 percent or more of the test cases are still impacted and need to be executed. Motivated by our finding, we further studied the code dependencies between source code files and test cases, and among test cases. We found that 1) test cases often focus on testing the integrated behaviour of the systems and 15 percent of the test cases have dependencies with more than 20 source code files; 2) 18 percent of the test cases have dependencies with other test cases, and test case inheritance is the most common cause of test case dependencies; and 3) we documented four dependency-related test smells that we uncovered in our manual study. Our study provides the first step towards studying and understanding the effectiveness of test impact analysis in the continuous testing setting and provides insights on improving test design and execution.

Autonomous Driving System (ADS) is one of the most promising and valuable large-scale machine learning (ML) powered systems. Hence, ADS has attracted much attention from academia and practitioners in recent years. Despite extensive study on ML models, it still lacks a comprehensive empirical study towards understanding the ML model roles, peculiar architecture, and complexity of ADS (i.e., various ML models and their relationship with non-trivial code logic). In this paper, we conduct an in-depth case study on Apollo, which is one of the state-of-the-art ADS, widely adopted by major automakers worldwide. We took the first step to reveal the integration of the underlying ML models and code logic in Apollo. In particular, we study the Apollo source code and present the underlying ML model system architecture. We present our findings on how the ML models interact with each other, and how the ML models are integrated with code logic to form a complex system. Finally, we inspect Apollo in a dynamic view and notice the heavy use of model-relevant components and the lack of adequate tests in general. Our study reveals potential maintenance challenges of complex ML-powered systems and identifies future directions to improve the quality assurance of ADS and general ML systems.