vlc_cancel() terminates the thread. There is a massive leak though.

This reverts commit a11c642f.

This reverts commit 4dc0b22b.

The channels count can trivially be computed with popcount().

It should be called before exit can happen.
This restores reverted commit 4d636fa3.

In my opinion, amplifying the audio that gets recorded to a file may do
more harm than good. Indeed, there is no direct feedback to detect and
correct saturation or distortion.

The proper reciprocal formula to that of VLC to PA conversion must be
used. Otherwise, we get an inconsistent value in the "volume" variable.

Now it should be usable within own libvlc applications

As a side note, this fixes a valgrind heap read overflow, but I do not
see why. I suspect a conflict between valgrind and some optimization in
libc or gcc here.

The total size of the buffer (i_size) was initialized with the whole
allocated size for the block. This fooled block_Realloc() in the case of
resizing to slightly larger, with the extra size in range from
32 to 32 + 80 bytes. block_Realloc() assumed it had enough space left in
the buffer padding to avoid reallocating memory.

Consequently, the block ended up with a i_buffer field with a value
larger than the allocated memory around p_buffer.

In the end, this could cause memory corruptions in all sorts of cases.
In my case, vlc was crashing while encoutering a corrupted mp3 file.
Signed-off-by: Rémi Denis-Courmont <remi@remlab.net>

This makes no sense. RPM needs the package specification as a separate
file, where the source tarballs are indicated. Chicken-and-egg...

Besides those files are out-of-date.

left out for now: cmake, ipkg and most of win32/UAC