I’ve seen one costly example in the wild - a problem domain was wildly oversimplified at an early stage of a project so one component was coded as a state machine. However, there were a bunch of background threads running independently of the state machine doing stuff and passing messages. And tracing code execution even within the state machine was a nightmare because a lot of what you’d think of as “call stacks” were coordinated by messaging passing between threads (mediated by the state machine controller) in a way that required you to do like 8 “find references” each time you wanted to see which function was actually getting called when a message was passed. Literally the entire thing could have just been a regular old init call making the same background threads, a regular request handler, and a regular handful of functions called if an error occurred but it was such a fucking mess we just started over in a new binary rather than trying to fix the existing one.