[Gnash-dev] Re: [Gnash-commit] [SCM] Gnash branch, master, updated. 6895d3b0542e1ac28a3e37dd150187dacb66195d

[Sandro Santilli]

On Tue, Aug 24, 2010 at 11:42:16AM -0700, John Gilmore wrote:
> > --- a/libmedia/MediaParser.cpp
> > +++ b/libmedia/MediaParser.cpp
> > @@ -411,6 +411,12 @@ MediaParser::parserLoop()
> >     while (!parserThreadKillRequested())
> >     {
> >             parseNextChunk();
> > +           gnashSleep(100); // thread switch 
> > +
> > +           // check for parsing complete
> > +           // TODO: have a setParsingComplete() function
> > +           //       exposed in base class for taking care
> > +           //       of this on appropriate time.
> >             boost::mutex::scoped_lock lock(_qMutex);
> >             waitIfNeeded(lock);
> >     }
> 
> If you wanted to revert the change, this won't do it, because we are
> still doing waitIfNeeded both in the main loop, and in each of the
> audio and video parseNextChunk's.

Didn't want a revert, just a sleep.
Note that waitIfNeeded is NOT being done in each of the audio 
and video parseNextChunk, unless something was really parsed.
Partial input frames won't push anything to buffers and thus
won't get to the waitIfNeeded.
I belive this was the problem you were originally encountering
(high CPU usage at end of parsing).

> There should be no need for this gnashSleep.  If we have our locks
> and semaphores set correctly, the system scheduler should do the job
> without being manually forced to switch tasks.

My tests here don't confirm that the scheduler is succeeding in that.

--strk;

  ()   Free GIS & Flash consultant/developer
  /\   http://strk.keybit.net/services.html