On: The free lunch is over: a fundamental turn toward concurrency in software (2005)

[Const-me]

> no way to efficiently replay concurrent programs in a deterministic fashion on multiple cores

OpenMP with static scheduler and hardcoded count of threads?

OpenMP ain’t a silver bullet, but for many practically useful scenarios it’s indeed as simple as calling a function.

[hderms → Const-me]

what about if you throw in async IO into the mix? seems like you'd have to have some kind of vmware-like virtualization where you record all IO interrupts into a linearalizable log

[Const-me → hderms]

When you throw async I/O in the mix it’s no longer just your application you’re debugging, it’s the whole OS with other processes, the OS kernel with these drivers and DMAs, and external hardware that’s not a code at all.

Can be tricky to debug or implement, but when I need that I normally use async-await in C#. The debugger in the IDE is multi-threaded, and does support these tasks.