[vdr] Re: VDR developer version 1.3.18 with NTPL

[Klaus Schmidinger <Klaus.Schmidinger@cadsoft.de>]

Mattias Grönlund wrote:
> Hi,
>
> I must have missed something, but the implementation of:
> cCondWait::SleepMs(int TimeoutMs)
> {
>   cCondWait w;
>   w.Wait(TimeoutMS);
> }
>
> Will if I understand this right, create a new cCondWait instance and than
> wait for TimeoutMs milliseconds before returning, but no other threads know
> about this new cCondWait object and can therefore not signal it to break
> before the timeout has elapsed.
>
> If I understood the NTPL problem right, pthread_cond_timedwait will spin
> lock for up to two milliseconds (one this specific case), so it will not
> work as expected.
So does this mean that for values of TimeoutMS larger than two milliseconds
it would work as expected? In that case maybe simply changing to

    cCondWait::SleepMs(3); // this keeps the CPU load low

might help. The actual wait time isn't that critical here - there's nothing to
do, anyway. We could even do

cCondWait::SleepMs(int TimeoutMs)
{
  cCondWait w;
  w.Wait(max(TimeoutMS, 3));
}

to make sure no values less than 3 are used (and define this as SleepMs()'s
behaviour).

Klaus