Navigation auf uzh.ch
Navigation auf uzh.ch
Speaker:
Prof. Dr. Carol Brandt, TU Darmstadt, Germany
Date: Thursday, 23 January 2025, 17:15
Location: room BIN 2.A.01 at the Department of Informatics (IfI), Binzmühlestrasse 14, 8050 Zürich (information here)
Details about the format of the talk shall be checked always just ahead of a certain presentation date: (information here)
Writing tests is a crucial part of ensuring that our software behaves as it should. However, it can be a tedious, boring and even painful activity for software developers. To alleviate this, research has developed a plethora of automatic test generation approaches over the years. While we create tools that are strong according to our metrics, their use in practice and industry is still limited.
One angle to address in automatic test generation is the usefulness and usability of these tests from the view of software developers. How much work is it for them to take over generated tests into their test suite? What changes do they still have to make? How many tests do they have to filter through?
In this talk, Dr. Carolin Brandt will introduce two research studies that take a deeper look at what developer still change when working with generated tests. In the first study [1], we submitted test generated through test amplification to 39 open source projects and analyzed the feedback from the project maintainers to understand what they would need to change before the tests are ready for their test suites. We observe that many of these changes greatly benefit from the developer’s understanding of the test intent and behavior, which is already a prerequisite for the tests to be accepted. In the second study [2], we empirically studied how usable tests generated by GitHub Copilot (a large language model) are for Python projects, including how developers could best formulate comments to obtain tests that require less manual changes. We learn that the large language model is strong in producing repetitive tests, whereas in most other cases additional fixes by the developer are necessary.
Dr. Carolin Brandt (https://carolin-brandt.de/) is an Assistant Professor at the Software Engineering Research Group (SERG) at TU Delft. She leads the research line on Human- and Developer-centered Software Engineering and investigates how to build automation tools that are better for the developers that create and use them.
Carolin received her PhD cum laude in 2024 from TU Delft. She is passionate about creating a healthy research environment through constructive reviewing and creating an inclusive neurodiverse community.
