Also deinitalizes both objects. This should fix the refcount
crash on exit.

VLCIntf now points to a static getter for the interface thread
pointer.

This fixes several problems: During initialization of
VLCMain, [VLCMain sharedInstance] is not ready yet when all other
objects are initialized inside VLCMains constructor. Due to the
way ARC works, the same applies to dealloc (the main shared instance
is nil already). But in both situations, we need VLCIntf for
callback (un)registration and potential logging.

The mac interface relies on static data in any case and does not
support multiple instantiations.

Only self should be used as callback parameters. Otherwise,
depending on the destruction order the original object might be
deinitialized already.

The window controller ensures that all top-level objects of the
mainwindow xib are properly released.

This removes the last occurence of VLCMain in xib files, thus
awakeFromNib will not be called for this object anymore. The code
has been moved to initThis removes the last occurence of VLCMain in xib files, thus
awakeFromNib will not be called for this object anymore. The code
has been moved to init.

This improves separation between main menu and the unrelated goto
time window.

Besides that, note that each time bindings are used with the file
owner, this owner needs to be an NSWindowController class. Only in
this case retain cycles are avoided (bindings have a strong
reference to the object used).

Thus, this commit helps to properly deinitialize the VLCMain object.

By deinitializing VLCMain, all connected objects are supposed to
be deallocated automatically. If [VLCMain sharedInstance] is gone,
the intf is deinitalized (this was previously checked with a
nil value of VLCIntf).

Fixes some retain cycles in input manager and playlist code which
prevented these objects to be deinitalized.

This should make it clearer what a l10n string and a IBOutlet is.

Also fixes behaviour for eq profile list

The former delegate approach had some drawbacks as one class could
be only delegate for one panel. This commit replaces it with a
completion handler block which moves the result handling code
closer to the actual creation of the dialog.

Also fixes crash for eq panels.

Libav/ffmpeg seems to store littlebit different values in framerate, so
there is #if to check that. Not sure if it's good idea at all to do that
like this.

Splits PopupPanel and TextfieldPanel into two xib files and
creates NSWindowController subclasses for each panel.

Each user of the panel gets its own window controller instance.
This is needed as the same kind of window can be shown multiple
times simultaneously (e.g. both in video and audio effects).

This also did not work with the previous implementation of the
2.2 branch and fixes broken behaviour.

if codec has list of supported framerates, select highest one
unless user has provided one. If we end up using CLOCK_FREQ in
time_base, it should be quite ok in most cases. Tested with ts/mp4/webm
and mp2v/mp4v/VP80. For some cases it can select higher fps in
mpeg-codecs (60fps) in case of input fps isn't yet known in that point
or is fubared.

Also add special case for MP4V as it doesn't like CLOCK_FREQ in
time_base/frame_rates.

Otherwise bitrate control is quite fobar.

Fixes #11538

Pointed out by Sesse

This should not adversely affect LibVLC: more errors cases will lead to
error code returns instead of abnormal termination. It is not strictly
semantically correct though (update to Win7 or above if you care).

MMDevice is not available on older Windows versions anyway. Failing to
load the plugin there is fine. It is even arguably better: it prevents
impossible choices showing in the preferences.

video/mpeg is underspecified, could be ES, PS or TS.
video/MP1S, video/MP2P and video/MP2T are more specific/better
specified (but should only be used internally or in SDPs).

Where the search paths list is not changed, using AC_PATH_PROG(S) has
no useful effects and makes manual override more difficult.

(Compare to Qt code using different search paths.)

Otherwise, it demuxes at max speed without allowing
user to change program

Some streams send clear padded data (probably
due to insufficient remaining data for encryption)

We'll only test on payload start then.

Since we can swap formats, some doesn't provide any or same values