Re: [hatari-devel] SDL1 vs. SDL2 sound pausing

[David Savinkoff <dsavnkff@xxxxxxxxx>]

Thomas Huth wrote:
> Am Sat, 10 Sep 2016 18:32:05 +0200
> schrieb Nicolas Pomarède:
> 
> > Le 10/09/2016 à 17:45, Thomas Huth a écrit :
> > > Am Sat, 10 Sep 2016 17:03:16 +0200
> > > schrieb Jerome Vernet:
> > >  
> > >> Le 10/09/2016 à 16:09, Nicolas Pomarède a écrit :  
> > >>> printf ( "audio cbk %d %d\n" , nGeneratedSamples ,
> > >>> CompleteSndBufIdx )  
> > >> You're right, nGeneratedSamples == 0 while in menus (or in other
> > >> thing, before emulation can be paused), and CompleteSndBufIdx stay
> > >> on the same value until the release the menu.  
> > >
> > > I guess we should simply mute audio completely if nGeneratedSamples
> > > is 0. Jerome, could you please try a patch like this:
> > >
> > > diff -r b98a852e12f2 src/audio.c
> > > --- a/src/audio.c	Sat Sep 10 17:39:20 2016 +0200
> > > +++ b/src/audio.c	Sat Sep 10 17:43:52 2016 +0200
> > > @@ -87,7 +87,13 @@
> > >  		CompleteSndBufIdx += len;
> > >  		nGeneratedSamples -= len;
> > >  	}
> > > -	else  /* Not enough samples available: */
> > > +	else if (nGeneratedSamples == 0)
> > > +	{
> > > +		/* No samples - that could mean that the main
> > > thread has been
> > > +		 * halted. Simply mute audio to avoid annoying
> > > audio loops */
> > > +		memset(pBuffer, 0, len * 4);
> > > +	}
> > > +	else  /* Some, but not enough samples available: */
> > >  	{
> > >  		for (i = 0; i < nGeneratedSamples; i++)
> > >  		{  
> > 
> > Hi
> > 
> > I was thinking at a similar patch, but I think we should completely 
> > remove the case that mirror half of the data if we don't have enough 
> > sample ; I'm not sure that duplicating previous data will give better 
> > audio result if we don't fill Buffer fast enough, we should as well
> > just put 0 bytes in Buffer to complete it
> 
> Not sure, but in case there are just some few samples missing, I think
> the memset-to-zero will cause audible disturbing noises, while the
> duplication might hide the missing samples quite well...
> I guess this needs some tests to be done to see what's the best
> approach here... any volunteers?
> 
>  Thomas
> 
Hi,

Looking at audio.c, one can see that *pBuffer++ is not touched
when nGeneratedSamples is == 0.
This leaves old data to be repeated. Zeroing the data leaves
silence to be repeated.
However, repeated old data is better than chunks of silence
when Not paused. (note: repeated data is milliseconds long)

I think Thomas' patch is correct with a small modification:
+	else if (nGeneratedSamples == 0 && emulator_paused)