Debugging non-deterministic programs on microcontrollers is notoriously challenging, especially when bugs manifest in unpredictable, input-dependent execution paths. A recent approach, called multiverse debugging, makes it easier to debug non-deterministic programs by allowing programmers to explore all potential execution paths. Current multiverse debuggers enable both forward and backward traversal of program paths, and some facilitate jumping to any previously visited states, potentially branching into alternative execution paths within the state space. Unfortunately, debugging programs that involve input/output operations using existing multiverse debuggers can reveal inaccessible program states, i.e. states which are not encountered during regular execution. This can significantly hinder the debugging process, as the programmer may spend substantial time exploring and examining inaccessible program states, or worse, may mistakenly assume a bug is present in the code, when in fact, the issue is caused by the debugger. This paper presents a novel approach to multiverse debugging, which can accommodate a broad spectrum of input/output operations. We provide the semantics of our approach and prove the correctness of our debugger, ensuring that despite having support for a wide range of input/output operations the debugger will only explore those program states which can be reached during regular execution. We have developed a prototype, called MIO, leveraging the WARDuino WebAssembly virtual machine to demonstrate the feasibility and efficiency of our techniques. As a demonstration of the approach we highlight a color dial built with a Lego Mindstorms motor, and color sensor, providing a tangible example of how our approach enables multiverse debugging for programs running on an STM32 microcontroller.

@article{paperbot2973,
  title = {MIO: Multiverse Debugging in the Face of Input/Output},
  author = {Tom Lauwaerts and Maarten Steevens and Christophe Scholliers},
  journal = {Proceedings of the ACM on Programming Languages},
  volume = {9},
  number = {OOPSLA2},
  year = {2025},
  doi = {10.1145/3763136}
}