[pulseaudio-discuss] [RFC PATCH] First draft implementation of better drain reporting

[David Henningsson]

On 03/02/2013 09:33 PM, Tanu Kaskinen wrote:
> On Sat, 2013-03-02 at 08:56 +0600, Dmitri Paduchikh wrote:
>> David Henningsson wrote:
>>
>>> Things remaining are:
>>>   - First, I'm not sure the calculations are entirely correct. I think I shouldn't need
>>>     to add process_usec to the wakeup, but if I don't I wake up a few samples before the
>>>     stream has finished playback.
>>
>> Sorry if it is a silly question but couldn't this happen because speed
>> of time inside sound card is somewhat different from that of system
>> time?
>
> That sounds plausible. Also, the timeout argument for ppoll() indicates
> the *maximum* sleep time. Interestingly, the timeout argument for poll()
> indicates the *minimum* sleep time. pa_rtpoll_run() may use either,
> depending on the system...
>

It could be, and there are also these comments later on:

/* Convert from the sound card time domain to the
                  * system time domain */
/* We don't trust the conversion, so we wake up whatever comes first */

...which doesn't exactly work in our favour here.

Anyway, I need to go back and check more deeply what causes us to wake 
up earlier than expected.

-- 
David Henningsson, Canonical Ltd.
https://launchpad.net/~diwic