The maximum resolution encouraged in the help text for the qtcapture
arguments is 1280x480, which doesn't make sense. The patch author
intended to imply 720p but appears to have forgotten the height
dimension. In addition, clean up the help text a bit.

--qtcapture-height works with 720, anyway, so it's just a doc issue.
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

All ES packets are gathered if there's no access control/dvb
filtering, adding lots of memcopy/realloc operations for packets
which will be discarded later by ES out.

This happend when multiple audio devices with the same name are
connected.

see https://code.google.com/p/android/issues/detail?id=65807
see https://code.google.com/p/android/issues/detail?id=70877Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

Less CPU usage.
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

Don't postpone delay when a command is received.
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

Create the thread after p_aout->sys is set.
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

It is maybe miss used, but the delay reported by this function is not coherent.
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

There is a new write method that can be non blocking and that can use a direct
ByteBuffer (no memcpy between java and jni).
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

- All JNIThread commands are now executed in a locked state.

- The delay (audiotrack delay + queue delay) is now calculated inside JNIThread.

- Write is now non-blocking and return almost immediately (it takes between 5us
  and 100us). Indeed when audiotrack internal buffer was full, write method was
  taking way too much time, therefore others commands were not processed in
  time (like TIME_GET).
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

and missing extra es

From 56 to 40 bytes on 64bit systems, saves 128KB

And drop all our patches :)

initWithFrame: is not always the designated initializer, thus the
object got improperly initialized. Use default initialization and
simplify code.

* Fully track pid parenting
* Simplify tables updates by ref counting
* No longer allows pid collisions by funky or forged tables
* Kills the 0..8192 loops
* Removes the 8192 stream_Control on close
* Remove unnecessary members and arrays
* Removes the nonsense multiple program by pmt
* Avoids by number/pid lookups
* Restarts ES on format change

Close #14053

Close #12135

We cannot assume that timeval.tv_usec (i.e. suseconds_t) is long.
Reported-by: Ian Chamberlain <ian.h.chamberlain@gmail.com>