[pulseaudio-discuss] Should corking be acknowledged by pa_sink_suspend()?

[Ville Sundell]

On 05.11.2014 11:57, Arun Raghavan wrote:
> On 5 November 2014 15:04, Ville Sundell <ville.sundell at nomovok.com> wrote:
>> Greetings everyone!
>> I am having some problems with corking: pulseaudio policy enforcer will cork
>> all the streams which belongs to a corked group. However, despite the fact
>> corking itself happens correctly (via pa_sink_input_cork() at
>> pulseaudio-policy-enforcement:src/policy-group.c) the stream will not end up
>> being corked, instead it will be set to PA_SINK_RUNNING.
> Not sure what code base this is, but this policy enforcement is not
> part of PulseAudio upstream. I guess this comes from Nemo or Tizen?
That is true, policy enforcer is implemented as a PulseAudio module, and 
is used by Nemo and other similar systems.
>
> Either way, there seems to be some confusion between concepts here.
> The idea of corking applies to sink inputs (pa_sink_input on the
> server side, pa_stream on the client side). The idea of being
> suspended applies to sinks (pa_sink). Could you explain your problem
> again, and what the symptoms are, keeping this in mind?
Thank you for clarifying this! The problem is as follows:
1) Program, which is part of a group which is already corked, opens a stream
2) Policy enforcer notices this, and corks the stream right away, like this:
     pa_sink_input_cork(si, group->corked);
3) PulseAudio corks the stream correctly, up to this point everything 
works as expected
4) However, pa_sink_suspend() gets called without a cause 
(s->suspend_cause), it tries to return it to a normal state 
(PA_SINK_RUNNING or PA_SINK_IDLE), but here the state should be and stay 
as SINK_INPUT_CORKED
5) This will lead to uncorking of the stream which should be corked 
(after applying the attached patch, this stream stays corked)

Thank you for your input :)

_Ville
>
>> After some tracing, it seems that it happens because of sink_set_state() is
>> called from pa_sink_suspend() (at pulseaudio:src/pulsecore/sink.c) right
>> after the corking. The end of the function is:
>>
>>      if (s->suspend_cause)
>>          return sink_set_state(s, PA_SINK_SUSPENDED);
>>      else
>>          return sink_set_state(s, pa_sink_used_by(s) ? PA_SINK_RUNNING :
>> PA_SINK_IDLE);
>> }
>>
>> Should it be something like:
>>
>>      if (s->suspend_cause)
>>          return sink_set_state(s, PA_SINK_SUSPENDED);
>>      else if (s->state != PA_SINK_INPUT_CORKED)
>>          return sink_set_state(s, pa_sink_used_by(s) ? PA_SINK_RUNNING :
>> PA_SINK_IDLE);
>>      else
>>          return 0;
>> }
> Just to reiterate, s->state will be from the pa_sink_state_t enum,
> whereas PA_SINK_INPUT_CORKED comes from the pa_sink_input_state_t enum.
>
> Cheers,
> Arun
> _______________________________________________
> pulseaudio-discuss mailing list
> pulseaudio-discuss at lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/pulseaudio-discuss