On 24/10/2016 16:19, Markus Armbruster wrote: > cur_mon is semantically thread-local: it's non-null while we're > executing a monitor command. That's a property of the stack, thus the > thread. The fact that it's not actually thread-local now is a bug I > blame on inertia.
Or just not noticing it until now. > Further evidence: if a thread calls error_report(), it should honor > cur_mon *only* when *this* thread executes a monitor command. It should > not spew to some unrelated monitor just because some other thread > happens to execute a monitor command right now. Yes, that makes sense. I'll work on redoing David's patch on top of this one then. Paolo