This more closely matches the usual open/read/seek/close pattern.

Pointers cannot be inherited as normal configuration item for somewhat
obvious reasons. For output parameters, this is not much of an issue as
LibVLC sets the pointer values directly on its media player object. But
for input-related parameters, LibVLC would need to store pointer values
in the media item. Thus this change to input items.

Note that pointers are considered intrinsically unsafe, so there are no
flags for the time being.

Call timestamp_FifoGet before releasing Output Buffers.
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

it uses its own system, so we must give it the proper value with --disable-option
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

This issue is easily reproducible with VDPAU activated, with a 60fps ts file,
see #14199.

With this sample, The video_output Thread is looping in ThreadDisplayPicture
way more often than with other videos. Consequently, vout_ManageWrapper and
ThreadControl are not called enough. As a result, subtitles are processed too
late, the mouse isn't able to hide/unhide, we can be stuck in fullscreen mode,
or second click on video to pop up the menu doesn't work.

To fix this issue: don't loop in ThreadDisplayPicture and don't wait in
vout_control_Pop if a picture was previously displayed.

Fixes #14199
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

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

this time it's working properly with a SMPTE color pattern
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

Differentiate between BT.601 (SD) and BT.709 (HD) for YUV pixel conversion
use the same precision values as the openGL vout
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

Some muxes, e.g. MP4, will send multiple header blocks. These are
merged by the HTTP server to a single header which is sent out to
the beginning of each client. However, they are _also_ sent out
directly on the wire to any client that connected before the first
block. In this case, we would send two separate Metacube header
blocks, which would have Cubemap (correctly) discard the first and
set only the second as header.

This would make us send only part of the MP4 header when sending
through Cubemap, if Cubemap connected before the first non-header
block, which would in turn confuse Chrome on Android (although not
Chrome on e.g. Linux). As Cubemap is pretty aggressive about
reconnecting (trying every 200 ms), this could easily happen in practice.
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

Signed-off-by: Hugo Beauzée-Luyssen <hugo@beauzee.fr>

Fixes read overflows.

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

libvlc_MetaRequest comes from the user, so we can increment i_preparse_depth to
force a preparsing.
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

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

Points to PMT's IOD

This reverts commit 6d35a852.

Added a boolean Playlist preference called start-pause.

Setting the preference causes the src/input/input.c Run function to call the
ControlPause function after Init finishes successfully.
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

Since it's not used anymore.
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

Get the type of the media.
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

Add i_preparse_depth in input_item to handle how many level of sub items can be
parsed.

The "recursive" option is now moved from access/file to the playlist category.

You can now abort a long local directory opening.

NET items won't be parsed recursively since playlist_preparser_Push is not
called with the META_REQUEST_OPTION_SCOPE_NETWORK argument.

Fixes #13850
Fixes #11921
Fixes #13872
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

Set b_net to true.
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

Set b_net to true, and set ITEM_TYPE_DIRECTORY if the item is a share of a
directory.
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

Set b_net to true, and set ITEM_TYPE_DIRECTORY if the item is a share of a
directory.
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

When an item was a NET/STREAM type, there was no way to distinguish if the item
was a file, a playlist or directory.

Add a new variable, b_net in addition to i_type. This variable is automatically
set when creating a new Item (set to true for ITEM_TYPE_STREAM). It can be
overridden via the new input_item_NewWithTypeExt function. This new function
will allow accesses to create FILE, or DIRECTORY items with b_net set to true.

Unless forced, the preparser won't parse items with b_net == true.
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

This type is set for streams (rtsp, httplive, mms...).
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

It allows to use dvdnav via a VLC access using new dvdnav stream callbacks.

When using a VLC access, dvdnav readahead flag is disabled in order to read
small chunk of data from pf_demux (that is 2kB).
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

Few PID in the whole range are really used.
Minus pointers and alignment, should save ~320KBytes
in most cases.

list of possible formats https://msdn.microsoft.com/en-us/library/windows/desktop/bb173059%28v=vs.85%29.aspx

the ones marked BROKEN_PIXEL are not writing pixels properly, not knowing much
about pixel shaders, I'm not the best qualified to handle this

the ones marked UNTESTED means my Win8 machine doesn't support them
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>