On Thu, 14 Dec 2023, Thomas Rodgers wrote:

> I need to look at this a bit more (and not on my phone, at lunch).
> Ultimately, C++26 expects to add predicate waits and returning a
> ‘tri-state’ result isn’t something that’s been considered or likely to be
> approved.

Ok, then that seems to fit best with my second suggestion: the predicate 
should only test the value and do nothing else, and actually trying to 
decrement the semaphore is left up to the caller _M_acquire(), which must 
then retry __atomic_wait_address if the compare-exchange fails.

-- 
Nate Eldredge
nate@thatsmathematics.com