An Empirical Study of Bugs in WebAssembly Compilers

WebAssembly is the newest programming language for the Web. It defines a portable bytecode format for use as a compilation target for programs developed in high-level languages such as C, C++, and Rust. As a result, WebAssembly binaries are generally created by WebAssembly compilers rather than being written manually. To port native code to the Web, WebAssembly compilers need to address the differences between the source and target languages and dissimilarities in their execution environments. A deep understanding of the bugs in WebAssembly compilers can help compiler developers determine where to focus development and testing efforts. In this paper, we conduct two empirical studies to understand the characteristics of the bugs found in WebAssembly compilers. First, we perform a qualitative analysis of bugs in Emscripten, the most widely-used WebAssembly compiler. We investigate 146 bug reports in Emscripten related to the unique challenges WebAssembly compilers encounter compared with traditional compilers. Second, we provide a quantitative analysis of 1,054 bugs in three open-source WebAssembly compilers, AssemblyScript, Emscripten, and Rustc/Wasm-Bindgen. We analyze these bugs along three dimensions: lifecycle, impact, and sizes of bug-inducing inputs and bug fixes. These studies deepen our understanding of WebAssembly compiler bugs. We hope that the findings of our study will shed light on opportunities to design practical tools for testing and debugging WebAssembly compilers.