The big issue we have right now is that we're limited by the "slots" a counter has. Either users are stuck with the "base" 4 slots, or they have to dynamically allocate to get more. Which is also fragile for the user to know beforehand.
The core realization of this change is that any time a fiber is waiting (IE it needs to be in the counter's queue), it's "asleep", and we can guarantee the stack memory is valid. So instead of the counter allocating memory, we allocate memory on the stack for the wait, and use a linked list to store the "queue" of waiting fibers.
BREAKING CHANGE: This removes AtomicCounter, TaskCounter, and AtomicFlag. And replaces them with WaitGroup. WaitGroup functions very similarly to TaskCounter, but users no longer need to worry about how many "waiting fiber slots" they need. Fibtex is also restructured. It's no longer possible to configure the "lock behavior"
The big issue we have right now is that we're limited by the "slots" a counter has. Either users are stuck with the "base" 4 slots, or they have to dynamically allocate to get more. Which is also fragile for the user to know beforehand.
The core realization of this change is that any time a fiber is waiting (IE it needs to be in the counter's queue), it's "asleep", and we can guarantee the stack memory is valid. So instead of the counter allocating memory, we allocate memory on the stack for the wait, and use a linked list to store the "queue" of waiting fibers.
BREAKING CHANGE: This removes AtomicCounter, TaskCounter, and AtomicFlag. And replaces them with WaitGroup. WaitGroup functions very similarly to TaskCounter, but users no longer need to worry about how many "waiting fiber slots" they need. Fibtex is also restructured. It's no longer possible to configure the "lock behavior"