Note: cc_pts/dts are still set to 0, I can't tell if this value is the
starting timestamp, or if it means the timestamps are not valid.

Better understanding on how pf_get_cc and packetizers are called in
src/input would clarify (fenrir?)

To build the old version, you have to pass the --force-deprecated option to the setup.py script.

* set a meta (without changing the file)
* save all meta to the file

Note, the opengl output is still broken on Mac OS X, in the way that no picture are rendered.

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

This issue was pointed out by Sébastien Zwickert.

We ensure this by two things:
- Retaining it by the media player and ensuring we are effectively stopped when releasing it. (ie, no one will use the drawable from now on).
- Retaining it during the life span of the vout. After a stop, the drawable might still be in the progress of receiving the notification of the vout removal, so we need it not to be freed during this period of time.
An alternative would be to create a protocol between drawable->vout to unregister the drawable.

A weirdness was introduced somehow. We were relying on the media state. (!)

Also, make sure that we always require stop() to be called when the media_player is destroyed. This may prevent some unwanted bad usage, whose consequence might be that _release() will block on input thread destruction.

macosx/framework: Print minutes and seconds when there is less than 5 mins remaining in -[VLCTime verboseStringDescription]

It will be used by opengl vout_displays.

This reverts commit ce7a4746.
Signed-off-by: Pierre Ynard <linkfanel@yahoo.fr>

This reverts commit b377e1af.
Signed-off-by: Pierre Ynard <linkfanel@yahoo.fr>

Currently, we never read from RTP and RTCP sockets, so we might as well
set their RX buffer size to zero.

It does more harm than good, since some clients (RealPlayer) will
just throw out a "not enough bandwidth" error and fail to open the
stream.