Shorten build time.

On x86_64 linux build:

LIBTOOL make -j2  60,82s user 13,98s system 112% cpu 1:06,59 total
        make -j2  56,83s user 12,72s system 110% cpu 1:03,20 total
DOLT    make -j2  44,32s user 11,02s system 108% cpu 51,215 total
        make -j2  42,15s user 11,04s system 106% cpu 50,155 total
Signed-off-by: Rafaël Carré <funman@videolan.org>

Signed-off-by: Diego Elio Pettenò <flameeyes@flameeyes.eu>
Signed-off-by: Rafaël Carré <funman@videolan.org>

As agreed last week with upstream, that's basicailly the least insane
way to deal with the ever changing and expanding V4L2 API. This allows
compiling VLC with old Linux headers and run it well on a new Linux
kernel. This also avoids a pile of backward compatibility hacks.

A few other userspace tool already copy videodev2.h like that.

CROPCAP is no longer mandatory, and many drivers failed to provide it.
If there is no CROPCAP, then there is no cropping.

Some formats can only be converted to FI32, and then
a FI32->FL32 converter can be used.

If we need to use pts for timestamps, but pts isn't set (the
avi demuxer seems to set it to 0 for cases where it isn't known, while
dts has a nonzero value), use dts instead.

This fixes things on OMAP4 (the only platform where b_use_pts is
set at the moment), when playing back files that have only dts,
no pts.
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

Bored as hell of those crashes...

Both conversion and streaming output wizard are covered.
However, demuxdump is _not_ covered.

To avoid disrupting existing workflows, the dialog is only shown when
overwrite is disabled. If there is no UI, then it will be assumed that
the file is not overwritten.

Signed-off-by: Rémi Denis-Courmont <remi@remlab.net>

This stops rejecting 0-byte packets, adds checks everywhere we
access the first byte of a packet, updates the th_decode_packetin()
return value checking so that duplicate frames aren't thrown away,
and updates video pts generation so the resulting frames are
displayed correctly,

Fixes bug #3416.
Signed-off-by: Rafaël Carré <funman@videolan.org>

This patch uses the information passed by the demuxer in block_t's
i_nb_samples and i_length to properly handle pre-skip, seeking
pre-roll, and end-trim. Multi-channel decoding should work. It also
adds .opus to the list of supported formats.
Signed-off-by: Rafaël Carré <funman@videolan.org>

This properly handles pre-skip, seeking pre-roll, and end-trim. It
uses the i_nb_samples field of block_t to signal to the decoder how
many samples should be returned after (possibly) skipping some at
the start of the block (for both pre-skip and pre-roll). In
addition, it abuses the i_length field of block_t to signal to the
decoder how many samples to trim from the end of the packet (for
end-trimming).

This patch does not compute correct timestamps for streams which
start at a non-zero offset (e.g., live streams joined in the
middle), nor does it correctly compute the length of a stream. But
neither of those things work for Vorbis, either. I'm leaving them
for a follow-up patch.
Signed-off-by: Rafaël Carré <funman@videolan.org>

Signed-off-by: Felix Paul Kühne <fkuehne@videolan.org>

Signed-off-by: Felix Paul Kühne <fkuehne@videolan.org>

Signed-off-by: Felix Paul Kühne <fkuehne@videolan.org>

If --no-sout-file-overwrite is specified, the access output will refuse
to overwrite an existing file. Overwrite is enabled by default for
backward compatibility with existing command lines.

While less precise, ticks count should be more stable than performance
counters.