It happens when pa_stream_write is called too late after a flush.
This fixes audio drop when the track has a positive delay.
Signed-off-by: Rémi Denis-Courmont <remi@remlab.net>

Some quicktime mov have invalid sample tables, requiring
quicktime "built-in" support, aka figure out.

The agsm atom must also not be used as codec as it is
apple GSM which has different block size and sample count
than the MS GSM codec (33,160/65,320).

refs samples/A-codecs/GSM/sample-gsm-8000.mov

This fixes minidlna browsing.
Fix #7690 #14333

One of the bounds of the loop were wrong, ending up in testing
formats that were not supported by the hardware.
Signed-off-by: Ricardo Ribalda Delgado <ricardo.ribalda@gmail.com>
Signed-off-by: Rémi Denis-Courmont <remi@remlab.net>

Fixes alaw regression.
/samples/A-codecs/aLaw/
Arbitrary limit still required for compressed format(AMR).
Now max to read 25ms of samples.

If defined, getTimestamp (since Android 4.4) or getPlaybackHeadPosition +
getLatency (since Android 4.3) will be used to get the AudioTrack position.

Since Android 4.3 there is a hidden method that retrieve the audio hardware
latency: AudioSystem.getOutputLatency() (that should not be used according to
the android comments).

This latency should not be used when we use AudioTrack.getTimestamp since it
already take into account the hardware latency. So why adding a new method ?
Because AudioTrack.getTimestamp don't work with Bluetooth (it returns false).

This fix lip sync with Bluetooth audio.

It improves delay precision for old devices.

Move AudioTimestamp variables into a separate struct.
Move AudioTimestamp calculation into a separate function.
Call getTimestamp every 500ms, keep the previous result in cache.

Since it only reset PlaybackHeadPosition.
Move i_samples_written initialisation out of this function.

It allows to play more than 13h30min (@44.1kHz).
Add more comments for the buggy getPlaybackHeadPosition method.

refs samples/A-codecs/Creative/

refs samples/A-codecs/CreativeADPCM8bit/

Some muxer is trowing incorrect rate/scale and then duration.
As we're not computing pts from fully decoded samples, we need
a correct pts for avcodec input. Fix is possible since ADPCM uses
a fixed bits/sample and header.

Was going into infinite loop in that case.
refs #14389

This mostly reverts commit f0380ff0 and
fixes it for Win32, but also Linux

Repicked pcr might be on a track that lacks
of pcr field.