Any kind of smart testing technique must be very efficient to be competitive with random fuzz testing. State-of the-art test generators are largely inferior to random testing in real world applications. This work proposes to gather and evaluate lightweight analyses that can enable the creation of an efficient and sufficiently effective analysis-assisted fuzz tester. The analyses shall leverage information sources apart from the program under test itself, such as e.g. descriptions of the targeted input format in the form of extended context-free grammars, or hardware counters. As the main contributions, an efficient framework for building fuzzers around given analyses will be created, and with its help analyses will be identified and categorized according to their performance.
2017 IEEE/ACM 39th International Conference on Software Engineering Companion (ICSE-C)