Signed-off-by: Christophe Mutricy <xtophe@videolan.org>

Since at least vlc 1.0.2, compiling vlc with a static version of a
recent version of the live555media library (the version released on
2009/11/12 for instance) results in the following warning while loading
the liblive555_plugin.so module:

[0x1901090] main libvlc warning: cannot load module `/usr/local/lib/vlc/demux/liblive555_plugin.so' (/usr/local/lib/vlc/demux/liblive555_plugin.so: undefined symbol: _ZTI9HashTable)

This warning is due to an error in the link order of the
live555_plugin, fixed by the following patch.
Signed-off-by: Christophe Mutricy <xtophe@videolan.org>

It seems that .mtv files don't work though

This reverts commit 59b025e3.

Those functions automatically abort if allocation fails (which is not
quite the same as calling assert()). Avoid these functions in new code.

Those are the same as the Linux kernel macros, and probably a bunch of
other projects. Usage:

    if (likely(condition))
        branch_likely_taken();

    if (unlikely(condition))
        branch_unlikely_taken();

Attention: those two macros convert the predicate to a boolean value.
Therefore you generally cannot use them for assignments like this:
    void *p_sys = likely (malloc (sizeof (*p_sys)));
    if (!p_sys)
        return VLC_ENOMEM;
Instead you can do this:
    void *p_sys = malloc (sizeof (*p_sys));
    if (unlikely (!p_sys))
        return VLC_ENOMEM;

Replace argument = realloc( argument, size ); with realloc_or_free() in modules/*, and while at it add assert( argument ) to mark unhandled ENOMEM conditions, also for malloc().

Introduce realloc_or_free() to src/*, and add assert() to mark unhandled ENOMEM error conditions. Allocation shrinking or otherwise handled allocations don't need realloc_or_free.

Introduce realloc_or_free(), which does what realloc() does but will free the argument if the reallocation fails. This is useful in some, but not all, use cases of realloc().

Skins2: Add x11 transparency support. Hope it sets the property on the right Window; didn't have a known transparent skin to test with.

Skins2: Cosmetics and drop const from string return as there really is no point: We're returning a fresh object.

Skins2: Move method bodies with inline tags to before where they're needed. Drop const in favour of static on power() and instead of recursing(!) make it use powf().

skins2: I fail to see why Subject (not Observer) needs to have virtual methods. If you do find a reason do tell. Observer should probably be defined inside Subject instead.

Skins2: Mostly cosmetics, some C++ style casting, factor out a scroll() in anticipation of figuring out where it belongs.

Skins2: Add (and use) template parameter typedefs so that they might be used, say, as function return types. The next commit needs this.

Skins2: Factor out duplicate code (mod), and provide a const accessor to the keymap: std::map<>::operator[] will create empty nodes if they don't exist already.