On 02/20/2018 07:19 PM, Raman Shishniou wrote:
>>> + if (chunk.length) {
If the source is running or idle, the chunk.length always will be 0.
if (PA_SOURCE_IS_OPENED(u->source->thread_info.state)) {
...
chunk.length = 0;
}
guaranteed that.
>>> + /* We have a pending data, let's stop polling pipe.
>>> + * Setting up pollfd->events = 0 is not enough to stop
>>> + * POLLHUP spam if all writers are closed pipe.
>>> + * We need to stop polling pipe completely */
>>> + if (rtpoll_item) {
Here rtpoll_item will be freed only once after source was suspended.
It will stay free until the source will be resumed.
>>> + pa_rtpoll_item_free(rtpoll_item);
>>> + rtpoll_item = NULL;
>>> + }
>>> + } else {
>>> + /* We have no pending data, let's start polling pipe */
>>> + if (rtpoll_item == NULL) {
Here rtpoll_item will be allocated only once after source was resumed and
processed all pending data.
>>> + rtpoll_item = pa_rtpoll_item_new(u->rtpoll, PA_RTPOLL_NEVER, 1);
>>> + pollfd = pa_rtpoll_item_get_pollfd(rtpoll_item, NULL);
>>> + pollfd->events = POLLIN;
>>> + pollfd->fd = u->fd;
>>> + }
>>> + }
>>
>> Your code will allocate/deallocate the rtpoll_item on each
>> iteration. This is unnecessary and CPU intensive (I was told).
>> I would still prefer my approach and I only see disadvantages
>> with your way.
>>
>
> No, it's not like that.
>
> rtpoll_item will be allocated:
> - On thread start
> - We just processed all pending data, i.e. source was just resumed.
>
> rtpoll_item will be freed:
> - We were got any data from pipe, but source was just suspended.
>
> So there will be only one free per suspend and one allocate per resume.
>
>
--
Raman
