It can even crash Qt interface

 * Enumerate frame sizes once rather than twice.
 * Do not enumerate frame rates if not supported.
 * Get actual frame rate from the device driverr.
 * Get exact fractional frame rate rather than round to single precision
   floating point.

Notes:
 * --v4l2-fps becomes totally redumdant. It should probably be redefined
   to select a maximum capture frame rate.
 * --v4l2-width and --v4l2-height are ignored. This is a regression.
   Maybe they should be redefined as maxima as well as --v4l2-fps.

Probably more recent address.
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

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

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

This decoder reorders timestamps according to the frame reordering,
and thus should be given pts instead of dts. Other decoders don't
do this reordering and should be given dts.

Still to be investigated how android/stagefright handles this,
since it always seems pass pts, not dts, to decoders.
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

CVE-2011-3048 (serious vulnerability)

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

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

Integer values outside a C enumeration are not defined (this is a bug in
V4L2 API really), so this would not work anyway.

"capabilities" counter-intuitively specifies the overall capabilities of
all device nodes provided by the given instance of the device driver.
"device_caps" specifies the capabilities of the opened device node,
if the V4L2_CAP_DEVICE_CAPS bit is set in "capabilities" (phew!).
Those two sets of capabilities are different if the hardware has
multiple functions, e.g. both video and VBI capture.

VLC cares about the fact that the specific device node supports video
capture or not, so lets use "device_caps" when available. Unfortatunely,
this requires kernel version 3.4. In practice, this would only cause an
actual failure if V4L2_CAP_STREAMING is set even though the current node
does not support streaming I/O, I think.

Unlike POLLERR, POLLHUP, and POLLNVAL, POLLRDHUP is not set by default.
See http://man7.org/linux/man-pages/man2/poll.2.htmlSigned-off-by: Rémi Denis-Courmont <remi@remlab.net>

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

Only do it for the right events.
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

This makes buffer allocation not fail.
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

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

This interpretation of nTop isn't too intuitive to me, but this is what
frameworks/base/media/libstagefright/colorconversion/ColorConverter.cpp
does for this pixel format.
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

As a side effect, this also makes the output correct on Samsung
Galaxy S II on ICS.
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

Support the OMX_IndexConfigCommonOutputCrop config struct, too.
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

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

A lot of releases have subtitles in subs/ subfolder

This was not correctly saving the last opened folder with some encoding

- add a win64 file
- do not repeatedly define the constant part
- use the test URL on every platform using --enable-debug, not only win32

This is completely untested, but it seems logical.

Also remove stray httpd_t.

Reported by Marcin 'Icewall' Noga from Hispasec
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

Reported by Marcin 'Icewall' Noga from Hispasec

Reported by Marcin 'Icewall' Noga from Hispasec

I wrote this as a quick hack after Gildas made the HTTP server bind to
all addresses instead of only loopback by default. It turns out it
lasted a bit longer :D