GitLab

was regression with AAC seek since rewrite of buffering

and add a no PCR reset condition

we can no longer rely on codecs to guess format

Required to correctly switch to another period,
as we disable es on period end.

also fixes a mishandled exception.
let late fail on regular access creation.

otherwise creates out of sync multiple streams with mp4

as child demuxers are not created yet when demux
wants seekable capability, CAN_SEEK was always false

Due to hls we we waiting for demuxer to start, which would
load the demuxer source (sub playlist then) and from that
file we would have the format and change the demux if
required.

The problem is that any unsupported format (like split subtitles)
is now being to be demuxed and gets probed and demuxed.

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.

only handling them should differ. adds debugging

And add implicit discontinuity on format change.
Use only tracker events to change state (will need to move currentchunk away)

And avoids creating stats from non http chunks (smooth)

because we'll need fake chunks

Can now flush buffers on demux restart.
Do align pcr after sending to decoders instead of always incrementing
by the target value (avoid dropping blocks across segments on restart).
Always issue a fakees to demuxer, then recycle on execution.
Avoids double deletion with duplicate Del commands (demuxer 0..n+self n).
Can now handle HLS discontinuities.
Drops the streamoutput layer.