This fixes a corner case:
A: rdlock -> OK
B: wrlock -> sleeps
A: rdlock -> dead lock

Signed-off-by: Laurent Aimar <fenrir@videolan.org>

(cherry picked from commit 3aa5a9f8cd7005aacbdb9211c17fae4570e065a1)

This reverts commit 3d0ef016.

As agreed, the information loss is too important, even for sanitizing the ABI. And there are alternative solutions available.

Signed-off-by: Jean-Paul Saman <jean-paul.saman@m2x.nl>
(cherry picked from commit 2c1ccdbdd8070a7f0881619be181b9a2906a7dd8)

Most if not all subscribers are using a fast asynchronous callback
internally anyway (at least Qt4 and libvlc do that). This was
redumdant and wasteful.

Parsing of a format string will anyway stop at the first nul, so there
is no point in having two nuls.

(cherry picked from commit 4054dd1fe53dac91138281fc3341bcbbc50f1fe8)

(cherry picked from commit 3709fb1eea9ec601981791c0e77a5d9d30dca38d)

The code (with or without this patch) looks quite fishy though.
Fortunately, I do not understand Objective-C, so I can blindly assume
it is correct.

This gets rid of the blocking queued connection for non-modal errors,
which was causing deadlocks in the audio output

Pointed-out-by: Laurent Aimar

Signed-off-by: Rémi Duraffort <ivoire@videolan.org>

In HandleMessage(), don't let the 0xF0 case fallthrough. Otherwise setting datalen would be pointless.
Signed-off-by: Rémi Duraffort <ivoire@videolan.org>

Signed-off-by: Rémi Duraffort <ivoire@videolan.org>

Signed-off-by: Rémi Duraffort <ivoire@videolan.org>

This is probably not enough to fix all the fontconfig issues.
(cherry picked from commit f25ad01c)
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

Thanks a lot, French CSA for using it with almost noone able to decode it...