What is the purpose of co_active_buffer?

[namandixit]

It seems that it is used only to get a unique address when calling co_active from a non-coroutine function. Why is it an array then, and not just a thread-local global variable?

[merryhime]

afaik, it's intended to be use as a thread's "default co_thread". since it can be used as an actual co_thread, it needs to be an array to store the required thread state.

[namandixit]

But the library doesn't seem to be storing anything in it, just using its pointer (which is unique for each thread) as a co_thread when we are NOT using a stack/coroutine created by co_create.

Also, what do you mean by "thread state"?

[merryhime]

The pointer can be used in other library functions as if it was a co_thread.

For example, if it was passed to co_switch, the new executing thread state would be read from the buffer.

If you follow the logic within the library internally uses it to store the non-co_thread thread state. (Hint: co_swap cannot have a nullptr argument and must swap the thread state with something)

[namandixit]

Oh wait, I think I get it: it's storing the registers in the array, instead of on the stack for the default thread of execution (created without using co-create). Is that right? (That's sneaky!!)

[merryhime]

Yes indeed!

[namandixit]

Thanks, I'm gonna close this one. Near's code is a tour de force :P