[pulseaudio-discuss] [PATCH 4/5] rtpoll: Implement pa_mainloop_api support

[Tanu Kaskinen]

On Tue, 2016-02-09 at 12:42 +0200, Tanu Kaskinen wrote:
> On Tue, 2016-02-09 at 15:47 +0530, Arun Raghavan wrote:
> > On Fri, 2015-01-02 at 15:04 +0200, Tanu Kaskinen wrote:
> > > +}
> > > +
> > > +static void find_expired_time_events(pa_rtpoll *rtpoll) {
> > > +    pa_usec_t now;
> > > +    pa_time_event *event;
> > > +    unsigned idx;
> > > +
> > > +    pa_assert(rtpoll);
> > > +    pa_assert(pa_dynarray_size(rtpoll->expired_time_events) == 0);
> > > +
> > > +    now = pa_rtclock_now();
> > > +
> > > +    PA_DYNARRAY_FOREACH(event, rtpoll->enabled_time_events, idx) {
> > > +        if (event->time <= now)
> > > +            pa_dynarray_append(rtpoll->expired_time_events, event);
> > > +    }
> > > +}
> > > +
> > > +static pa_time_event *find_next_time_event(pa_rtpoll *rtpoll) {
> > > +    pa_time_event *event;
> > > +    pa_time_event *result = NULL;
> > > +    unsigned idx;
> > > +
> > > +    pa_assert(rtpoll);
> > > +
> > > +    if (rtpoll->cached_next_time_event)
> > > +        return rtpoll->cached_next_time_event;
> > > +
> > > +    PA_DYNARRAY_FOREACH(event, rtpoll->enabled_time_events, idx) {
> > > +        if (!result || event->time < result->time)
> > > +            result = event;
> > > +    }
> > > +
> > > +    rtpoll->cached_next_time_event = result;
> > > +
> > > +    return result;
> > > +}
> > > +
> > >  static void reset_revents(pa_rtpoll_item *i) {
> > >      struct pollfd *f;
> > >      unsigned n;
> > > @@ -204,9 +632,14 @@ static void reset_all_revents(pa_rtpoll *p) {
> > >  }
> > >  
> > >  int pa_rtpoll_run(pa_rtpoll *p) {
> > > +    pa_defer_event *defer_event;
> > > +    pa_time_event *time_event;
> > >      pa_rtpoll_item *i;
> > >      int r = 0;
> > >      struct timeval timeout;
> > > +    pa_time_event *next_time_event;
> > > +    struct timeval next_time_event_elapse;
> > > +    bool timer_enabled;
> > >  
> > >      pa_assert(p);
> > >      pa_assert(!p->running);
> > > @@ -218,7 +651,28 @@ int pa_rtpoll_run(pa_rtpoll *p) {
> > >      p->running = true;
> > >      p->timer_elapsed = false;
> > >  
> > > -    /* First, let's do some work */
> > > +    /* Dispatch all enabled defer events. */
> > > +    while ((defer_event = pa_dynarray_last(p-
> > > > enabled_defer_events))) {
> > > +        if (p->quit)
> > > +            break;
> > > +
> > > +        defer_event->callback(&p->mainloop_api, defer_event,
> > > defer_event->userdata);
> > > +    }
> > 
> > Am I missing something, or is this an infinite loop if there are any
> > enabled defer events?
> 
> As discussed in IRC, I did this like this, because pa_mainloop_run() in
> practice behaves the same way. However, mainloop-api.h says that each
> defer event runs only once per mainloop iteration, so I'll have to
> change this.

I started looking into this, and realized that the "once per mainloop
iteration" clause makes no difference to pa_mainloop_api users. There
is no pa_rtpoll_iterate(), and pa_rtpoll_run() will behave the same way
in any case, so is it really worth the effort to artificially refactor
the code to have "iterations"?


-- 
Tanu