This matches all the other networking access modules and solves performance issues when playing high bitrate content

This avoids spurious errors in the decoders (and logs).

This function will cancel the decoder picture pool.
Signed-off-by: Rémi Denis-Courmont <remi@remlab.net>

Signed-off-by: Rémi Denis-Courmont <remi@remlab.net>

This reverts commit 09bb0203.

This improves the compatibility with vlc_poll().
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

New profiles that might break packetization.
(ITU-T H264 2014/02)

Old profile, out of spec
ref: https://ffmpeg.org/pipermail/ffmpeg-devel/2012-October/132400.html

new sample:
samples/V-codecs/h264/lossless_timecode_HIGH444_IDC144.h264

chroma/scaling/matrix is on HIGH profile SPS
(as in 14496-10:2012)

Against further dynamic linker namespace pollution.

Also optimize it away where applicable.

(see comments)

The ES output never calls FrameNext on non-video tracks.

Pruning by time was removed a while ago, but
pruning by segment number only can't work for pruning the whole
playlist as that number can be inconsistent across representations
and streams/adaptation sets.

Pruning by number should only occur merging/at representations level.

was not working with live streams as not starting from zero

And use targetduration hint for scheduling HLS updates

segment could become unavailable before we could fully download it.

In case of memory error, the functions will abort (table macros). Even
if it did not, callback deletion would later abort failing to find the
callback instead. (The API could be redesigned not to require memory
allocation if necessary.)

In case of nonexistent variable, log an error and fail safe as before.
This lead to deadlocks of misbehaviours as before - no changes here.

If the callback does not exist, the program aborts anyway, so the
error value was mostly useless, and indeed always ignored.