I really love debugging stories like this as they provide an opportunity for learning about things, you'd normally won't touch, by providing relatable context.
When you start stack-swapping, at some point you're going to need to deal with unwinding the stack (exceptions).
What surprises me is that they said that, in most cases, throwing exceptions seemed to work OK!
I would have assumed that if it wasn't properly supported by the Boost Context object, it would have frequently been corrupting memory. Maybe they just never noticed the corruption.
The instant they told about null registers I though "someone has corrupted the stack".
From there enabling ASAN or perhaps even stack-protector (even in Release) should've caught it rather quick... (First context switch, likely resume with exception raised.)
Now seeing a crash in exception handler, that is something. It's rather sad they didn't fix Boost.Context, seems like it could always use more platform ifdefs.
It's crazy that the devs could do all this debugging in just 1 week. I think at least 99% of us programmers wouldn't be able to do it. I sometimes hunt for much simpler bugs for much longer time.
