GetWindowDC paints in nonclient area (not recommended on msdn)
and not needed for skins. Also, it doesn't take WS_CLIPCHILDREN
into account and forced to add a lengthy hack which can now be removed.

This patch hopefully could also fix a refresh issue found only on
Vista/Win7 (to be tested) (http://forum.videolan.org/viewtopic.php?f=15&t=68891)

...without reference. With a direct connection, the call stack ensures
that the pointer remains valid until the slot returns. This is not the
case for asynchronous ("queued" in Qt4 documentation) signals.

Obviously, this means we have to emit the signals from the Qt4 main loop
thread, *or* alternatively not touch any Qt4 UI stuff fromt he slot.
Hopefully, any such bug would be easier to find this ways.

This reverts commit dcd59284.

This bug is so pervasive within the Qt4 interface that removing this
instance won't make a difference.

Ref #3332
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

Patch by ale5000, on IRC. Close #3336

...unless proper reference counting si involved,
or the signal is blocking queued (but that is deadlock-prone).

Ref: LP#528285

This should fix LP#465560 - segmentation fault in strstr()

We only use getopt_long(). Also we don't care about pre-C89 compilers.

(We still need -lgnugetopt for cachegen)

This mostly reverts commit 37de4458.

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

Fix the bug where caching would stay all the time when stopping