> Well, if I am being honest, there is a bit of up-front knowledge here. I don’t think we can avoid spawning real threads here, unless we do something really cursed with inline assembly. When something calls that pause() function, and we want it to stay paused until further notice, that just has to happen in a thread which maintains a stack separate from the stack of our test.

Is there a reason we couldn't use some kind of async runtime? It seems like you're instrumenting atomic operations to achieve cooperative multitasking. Maybe I need to drink more coffee, but it seems simpler without threads.