My Master's Thesis

Fuzzing as Editor Feedback

You can download my Master's thesis here. This is the abstract:

Fuzzing – testing code with random inputs to see if it crashes – is a well-researched technique. In this work, we explore how fuzzing enables new editor tooling. We integrate fuzzing into the compiler pipeline and language tooling of Martinaise, a custom programming language. Our prototype can fuzz individual functions and show results in Visual Studio Code. Unlike other methods for testing code, this requires no extra effort from the developer. Our tool shows crashing inputs next to the function signature. This shortens the feedback loop compared to traditional fuzzing, as edge cases are shown while the code is written, thereby preventing bugs. Our tool also displays example inputs, chosen based on code coverage. Examples that execute different parts of a function can help understand the function’s behavior without reading the implementation. Examples can be filtered to ones reaching the cursor position. We evaluate the quality and performance of our prototype and for which kind of code fuzzing can give useful examples. We discuss how our approach can be applied to other programming languages. Fuzzing has the potential to serve as the basis for Babylonian Programming, visualizations, debugging sessions, and other tools that benefit from concrete examples.