It would seem to be the logical thing to do?

That is, suppose two threads are sharing a resource. Thread A has it locked. B is "waiting". Is B in a loop burning cycles running in the background(regardless of thread.sleep, which only alleviates the problem) or does it yield completely and somehow inform the lock to resume it when A has unlocked the resources?

The first one burns cycles and can have timing problems. I.e., What if A locks and unlocks at the same rate that B checks? (I suppose a random sleep time would help with this) (

"Yielding", OTOH, has B burn no cycles waiting in a loop. This can lead to optimization and prioritization and all that(after an unlock, all the threads waiting can be called, but in what order).

Obviously yielding is more complex and requires the threads kept track of(an array for each lock/unlock pair) but far more efficient.

I'm hoping D does this, but not holding my breath.

On 9/12/15 10:48 PM, Prudence wrote:
> It would seem to be the logical thing to do?
>
> That is, suppose two threads are sharing a resource. Thread A has it
> locked. B is "waiting". Is B in a loop burning cycles running in the
> background(regardless of thread.sleep, which only alleviates the
> problem) or does it yield completely and somehow inform the lock to
> resume it when A has unlocked the resources?

If you are using D mutexes or synchronized statements, it uses the OS' mechanisms (e.g. pthreads). For all supported OSes, this means it is asleep and waiting for the OS to awaken it when it has locked the resource.

> The first one burns cycles and can have timing problems. I.e., What if A
> locks and unlocks at the same rate that B checks? (I suppose a random
> sleep time would help with this) (

You don't have to worry about this. If A unlocks a resource that B is waiting for the lock, it cannot lock it again before B gets it.

-Steve