xspf writer: forget to increment the source pointer in [22665], that would result in file://file://file:///path/to/input entries in ml.xspf

telepathy: register the callbacks only when we are connected to the bus, to avoid the module being unloaded, while the callbacks are still active

Fixed use of uninitialized data for CC detection.

We don't do that on other URIs because those characters may be used as delimiters, and must not be percent encoded.

Now produces valid XSPF-1 files.

* don't use the text archive format for nibs, as the binary version is faster to load and smaller to store. Additionally, it's a lot better for diffing on trac, as humans should't understand the text format anyway.

playlist_CreateNode(): add an argument to specify an input_item_t to be linked with the node, rather than creating a new input.
Setting that argument to NULL will make playlist_CreateNode() behave like previously.

That way we can create only one input per pair of node (for local playlist, media library, and service discovery)
Previous behaviour was to create 2 inputs with the same i_id member, but we would store both input in a binary search tree (sorted by i_id), and that kind of tree MUST NOT have 2 items with the same key, else we will get some bad memory corruption when the wrong input is removed from the tree (the other being left in the tree while the memory referred by it was freed).
Note that this breaks ABI

* implemented the generic VLC media player Help window on OSX to match the QT4 interface. Additionally, I modified the HTML source a bit, so potential umlauts won't be displayed as UTF-16 on OSX, which would lead to German umlauts being Chinese characters, etc. BTW. Withthis commit, we also link against the WebKit framework..

Please reports broken CC decoding (with a sample to test).

is used. (Only dvb/atsc CC support is working)

are dynamically added when detected.
(Decoder/Packetizer support not yet commited)

The rollup and paint on mode is not really good (a lot of flicker) but
I don't see how to fix it (the subtitle is updated as soon as a new
character is added).

Always check the certificates chain - it does not make any sense, is counter-intuitive and insecure to use x509 but not validate certs.
The whole security and point of TLS rely on this...

(this would be a bug in whatever is using the TLS object anyway)

playlist_NodeDelete(): we want to delete a specific playlist_item_t*, not a playlist_item_t* associated to a specific input_item_t*

svn: contrib/ffmpeg: Work around the fact that there is not enough register available on Mac OS X/x86 for certain optimization (most likely a gcc bug).