Version Space Learning for Verification on Temporal Differentials
Configuration files provide users with the ability to quickly alter the behavior of their software system. Ensuring that a configuration file does not induce errors in the software is a complex verification issue. The types of errors can be easy to measure, such as an initialization failure of system boot, or more insidious such as performance degrading over time under heavy network loads. In order to warn a user of potential configuration errors ahead of time, we propose using version space learning specifications for configuration languages. We frame an existing tool, ConfigC, in terms of version space learning. We extend that algorithm to leverage the temporal structuring available in training sets scraped from versioning control systems. We plan to evaluate our system on a case study using TravisCI configuration files collected from Github.
Thu 13 Jul
|13:30 - 14:00|
Jenny HotzkowSaarland University
|14:00 - 14:30|
Deborah S. KatzCarnegie Mellon University
|14:30 - 15:00|