Continuity of the two FIFOs is maintained, as is consistency with the
input date. PTS discontinuity only seems to occur due to rounding
errors in the filters. It should be better not to drop samples, but
simply fix the buffer PTS and length.

Ideally, resamplers would use date_Increment() so they do not introduce
rounding errors. But first the block disconuity flag should be checked
for correctness during filtering, so that resamplers can reset the date
in case of hole.

...for good this time (yeah right).

(Logic for more than one reference was inverted)

libdbus has multiple thread-safety issues and using private connections
(ie: not shared with other vlc modules) helps avoiding some of them.

thx to j-b for the reminder

This happens almost all the time, and is unfortunately normal behavior.

This should fix the infrequent assertions failures at exit.

The decoder thread drains the output at end of stream. So there is
always an underflow between consequent audio inputs.
Thus this does not provide gap-less audio as is.

macosx: fail the video output if Quartz Extreme isn't supported, removed specific options, which are now handled by the core

QE is supported on virtually any Mac is running 10.5+ except for broken Hackintoshes

no functional change

This had no relations with the audio input anymore. In particular,
there is no need to reinitialize this when the input is restarted.

I assume this breaks with S/PDIF. And I do not see much use for this.

removed the broken fps display and added a field for EncodedBy

fixes #4710

This ensures that the variables will be available as soon as the aout
is externally visible, and that there value will persist across inputs.

This also saves a bunch of lookup at each new input.

Those variables are inherited and they have no callbacks.

(Lets assume that uintptr_t is big enough, i.e. 32-bits).

 - get the mode string directly from the callback (new) variable value,
 - reduce locking scope,
 - fix read overflow: do not apply peak protection in "max" mode.