'Twas brillig, and Lu Guanqun at 24/08/11 01:29 did gyre and gimble:
> On Tue, Aug 23, 2011 at 05:00:58PM +0800, Colin Guthrie wrote:
>> Yes, the latest code is better but what I was meaning was something like:
>>
>> diff --git a/src/utils/pacat.c b/src/utils/pacat.c
>> index f687402..2568db5 100644
>> --- a/src/utils/pacat.c
>> +++ b/src/utils/pacat.c
>> @@ -91,6 +91,8 @@ static int32_t latency_msec = 0, process_time_msec = 0;
>> static pa_bool_t raw = TRUE;
>> static int file_format = -1;
>>
>> +static int cork_requests = 0;
>> +
>> /* A shortcut for terminating the application */
>> static void quit(int ret) {
>> pa_assert(mainloop_api);
>> @@ -408,6 +410,15 @@ static void stream_event_callback(pa_stream *s,
>> const char *name, pa_proplist *p
>>
>> t = pa_proplist_to_string_sep(pl, ", ");
>> pa_log("Got event '%s', properties '%s'", name, t);
>> +
>> + if (pa_streq(name, PA_STREAM_EVENT_REQUEST_CORK)) {
>> + if (!cork_requests)
>> + pa_operation_unref(pa_stream_cork(s, 1, NULL, NULL));
>> + cork_requests++;
>> + } else if (pa_streq(name, PA_STREAM_EVENT_REQUEST_UNCORK)) {
>> + cork_requests--;
>> + if (!cork_requests)
>> + pa_operation_unref(pa_stream_cork(s, 0, NULL, NULL));
>> + }
>> +
>> pa_xfree(t);
>> }
>>
>>
>>
>> (not a real patch!)
>>
>>
>> This way the actual operation is only sent at the transition of 0->1 and
>> not 1->2 etc. (and the opposite, it will only be sent when doing 1->0
>> and not from 2->1).
>>
>> I think this makes sense, but other opinions are welcomed too.
>
> I see your point of only doing the corking at 0->1 and vice versa.
OK, pushed this patch which should do the trick.
http://cgit.freedesktop.org/pulseaudio/pulseaudio/commit/?id=cef16430d0e307a3978dc6aa9a0d9b6940b0e0fa
>>
>>>> We may at some point have to change the semantics at the server side to
>>>> deal with the start_corked issue, and thus this is likely more robust in
>>>> those circumstances.
>
> BTW: Which semantics are we trying to change to? I may not have the
> context here...
You commented on the bug, but for reference one idea is to mandate
client side reference counting like the above, and to account for this
in the initial cork state handling too. But I'm not sure this is sufficient.
I've not got details just now but will do in a little while.
Col
--
Colin Guthrie
gmane(at)colin.guthr.ie
http://colin.guthr.ie/
Day Job:
Tribalogic Limited [http://www.tribalogic.net/]
Open Source:
Mageia Contributor [http://www.mageia.org/]
PulseAudio Hacker [http://www.pulseaudio.org/]
Trac Hacker [http://trac.edgewall.org/]
