platform/upstream/gstreamer.git
7 years agov4l2bufferpool: Don't stop streaming when pool is flushing
Nicolas Dufresne [Sat, 5 Aug 2017 16:23:30 +0000 (12:23 -0400)]
v4l2bufferpool: Don't stop streaming when pool is flushing

The purpose of being able to flush the buffer pool is only to
unlock any blocked operation. Doing streamoff/streamon had the
side effect of turning off and on the camera. As we do a flush_start
/ flush_stop sequence when shutting down, that would cause a really
quick sequence of streamoff/streamon/streamoff/close which was
causing some cameras to stop working.

https://bugzilla.gnome.org/show_bug.cgi?id=783945

7 years agortpjitterbuffer: implement basic chain_list function
Tim-Philipp Müller [Sun, 17 Sep 2017 15:18:48 +0000 (16:18 +0100)]
rtpjitterbuffer: implement basic chain_list function

Doesn't do anything fancy yet, but still avoids lots of
unnecessary locking/unlocking that would happen if the
default chain_list fallback function in GstPad got invoked.

7 years agomultifilesink: use new gst_buffer_list_calculate_size()
Tim-Philipp Müller [Sun, 17 Sep 2017 11:50:30 +0000 (12:50 +0100)]
multifilesink: use new gst_buffer_list_calculate_size()

7 years agortpbin: add option for sanity checking timestamp offset
Patrick Radizi [Thu, 14 Sep 2017 11:00:56 +0000 (13:00 +0200)]
rtpbin: add option for sanity checking timestamp offset

Timestamp offsets needs to be checked to detect unrealistic values
caused for example by NTP clocks not in sync. The new parameter
max-ts-offset lets the user decide an upper offset limit. There
are two different cases for checking the offset based on if
ntp-sync is used or not:
1) ntp-sync enabled
   Only negative offsest are allowed since a positive offset would
   mean that the sender and receiver clocks are not in sync.
   Default vaule of max-ts-offset = 0 (disabled)
2) ntp-sync disabled
   Both positive and negative offsets are allowed.
   Default vaule of max-ts-offset = 3000000000
The reason for different default values is to be backwards
compatible.

https://bugzilla.gnome.org/show_bug.cgi?id=785733

7 years agortpbin: add option for increasing ts_offset gradually
Patrick Radizi [Thu, 14 Sep 2017 09:20:17 +0000 (11:20 +0200)]
rtpbin: add option for increasing ts_offset gradually

Instant large changes to ts_offset may cause timestamps to move
backwards and also cause visible effects in media playback. The new
option max-ts-offset-adjustment lets the application control the rate to
apply changes to ts_offset.

https://bugzilla.gnome.org/show_bug.cgi?id=784002

7 years agov4l2src: Ensure all caps a fixated
Nicolas Dufresne [Wed, 13 Sep 2017 20:05:08 +0000 (16:05 -0400)]
v4l2src: Ensure all caps a fixated

The code relied on the list compare function to fixate the caps
but if the caps only has one structure, the compare function will
never get called. Capture device for which there is only one
structure in the caps would then get some assertion and later
fail badly.

Instead, fixate before inserting into the list and split the reading
and the fixation of the structures.

7 years agov4l2object: Don't leak the par value
Nicolas Dufresne [Wed, 13 Sep 2017 15:52:09 +0000 (11:52 -0400)]
v4l2object: Don't leak the par value

7 years agov4l2-renegotiate: Don't leak the option context
Nicolas Dufresne [Wed, 13 Sep 2017 15:38:44 +0000 (11:38 -0400)]
v4l2-renegotiate: Don't leak the option context

7 years agov4l2src-renegotiate: Don't leak pipeline desc string
Nicolas Dufresne [Wed, 13 Sep 2017 15:33:33 +0000 (11:33 -0400)]
v4l2src-renegotiate: Don't leak pipeline desc string

7 years agov4l2-renegotiate: Change --enable-dmabuf into --io-mode=
Nicolas Dufresne [Wed, 13 Sep 2017 15:32:09 +0000 (11:32 -0400)]
v4l2-renegotiate: Change --enable-dmabuf into --io-mode=

This gives allow testing dmabuf importation but also exportation buy
letting user pick anything from the io-mode property on v4l2src.

7 years agomatroskademux: search_cluster should find preceding cluster before target
Mark Nauwelaerts [Mon, 11 Sep 2017 18:24:27 +0000 (20:24 +0200)]
matroskademux: search_cluster should find preceding cluster before target

... since failing this constraint takes search_pos by surprise which might
then end up in an infinite loop.

Fixes https://bugzilla.gnome.org/show_bug.cgi?id=787538

7 years agortprtx{send,receive}: improve the debug messages
George Kiagiadakis [Thu, 7 Sep 2017 11:33:57 +0000 (14:33 +0300)]
rtprtx{send,receive}: improve the debug messages

* use INFO/DEBUG/LOG/TRACE equaly and meaningfully;
  previously rtprtxsend:LOG and rtprtxreceive:LOG would generate
  a totally different amount of log traffic and sometimes it was
  impossible to see the information you wanted without useless
  spam being printed around
* improve the wording, give a reasonable and self-explanatory
  amount of information
* print SSRCs in hex
* avoid G_FOO_FORMAT for readability (we are just printing integers)

7 years agov4l2object: Handle BT2020 for colorspace and transfer
Ian Jamison [Tue, 5 Sep 2017 14:30:40 +0000 (15:30 +0100)]
v4l2object: Handle BT2020 for colorspace and transfer

This was not fully handled in switches and
ub gst_v4l2_object_get_colorspace();

https://bugzilla.gnome.org/show_bug.cgi?id=787313

7 years agov4l2object: Fix colorimetry transfer lookup for 4K video
Ian Jamison [Tue, 5 Sep 2017 14:29:24 +0000 (15:29 +0100)]
v4l2object: Fix colorimetry transfer lookup for 4K video

https://bugzilla.gnome.org/show_bug.cgi?id=787160

7 years agov4l2src: Check if caps have changed after try_fmt
Nicolas Dufresne [Wed, 6 Sep 2017 15:25:53 +0000 (11:25 -0400)]
v4l2src: Check if caps have changed after try_fmt

try_fmt will update the caps colorimetry and interlace-mode. Before this
call, those field are missing. The caps equality check was always
failing when a spurious reconfigure event was received.

7 years agoqtmux: Allow MPEG layer 1/2, AC3 and Opus in qtmux
Jan Schmidt [Wed, 6 Sep 2017 13:55:38 +0000 (23:55 +1000)]
qtmux: Allow MPEG layer 1/2, AC3 and Opus in qtmux

qtmux is supposed to be the muxer that allows all formats,
with others (mp4mux and friends) being profile-restricted.

7 years agortph265depay: fix keyunit detection
Tim-Philipp Müller [Tue, 5 Sep 2017 11:56:44 +0000 (12:56 +0100)]
rtph265depay: fix keyunit detection

https://bugzilla.gnome.org/show_bug.cgi?id=787254

7 years agojpegdec: Fix decoding of streams that don't signal exactly twice the height
Sebastian Dröge [Tue, 5 Sep 2017 12:42:17 +0000 (15:42 +0300)]
jpegdec: Fix decoding of streams that don't signal exactly twice the height

... and also progressive streams.

7 years agojpegdec: Handle interlaced MJPEG streams
Sebastian Dröge [Tue, 5 Sep 2017 10:28:16 +0000 (13:28 +0300)]
jpegdec: Handle interlaced MJPEG streams

These come with two JPEG images per buffer of half height than signalled
in the container.

Changes based on Tim-Philipp Müller's 0.10 branch:
https://cgit.freedesktop.org/~tpm/gst-plugins-good/log/?h=jpegdec-interlaced

https://bugzilla.gnome.org/show_bug.cgi?id=568555

7 years agodcaparse: Really fix "usage before unmap"
Edward Hervey [Mon, 4 Sep 2017 15:06:39 +0000 (17:06 +0200)]
dcaparse: Really fix "usage before unmap"

Previous patch would try to unref a buffer that was pushed downstream.

Instead only unref when/if needed and keep usage of the cleanup: goto
block

7 years agodcaparse: Don't unmap buffer before accessing data from it
Arun Raghavan [Sun, 3 Sep 2017 09:53:10 +0000 (15:23 +0530)]
dcaparse: Don't unmap buffer before accessing data from it

The previous patch added a check for a substream header after
gst_buffer_unmap(), which is incorrect.

7 years agodcaparse: preserve DTS HD substream
Matej Knopp [Sat, 24 Jun 2017 16:47:14 +0000 (18:47 +0200)]
dcaparse: preserve DTS HD substream

7 years agortpsbcpay: Fix some tabs that crept in somehow
Arun Raghavan [Mon, 14 Aug 2017 12:48:07 +0000 (18:18 +0530)]
rtpsbcpay: Fix some tabs that crept in somehow

7 years agortpbin: Also log local and SR RTP running times when doing ntp-sync=true
Sebastian Dröge [Tue, 29 Aug 2017 16:13:58 +0000 (19:13 +0300)]
rtpbin: Also log local and SR RTP running times when doing ntp-sync=true

7 years agortpbin: also create session when creating the send_rtcp_src_%u pad
Matthew Waters [Thu, 24 Aug 2017 07:06:38 +0000 (17:06 +1000)]
rtpbin: also create session when creating the send_rtcp_src_%u pad

If one requests the send_rtcp_src_%u pad before a recv_rtcp_sink_%u pad,
the session/pad would never be created and NULL was returned.
Switching the request order would work.

https://bugzilla.gnome.org/show_bug.cgi?id=786718

7 years agotests: mpg123audiodec: add files needed by unit tests
Tim-Philipp Müller [Sat, 26 Aug 2017 11:59:35 +0000 (12:59 +0100)]
tests: mpg123audiodec: add files needed by unit tests

7 years agotests: add basic unit test for twolame as well
Tim-Philipp Müller [Sat, 26 Aug 2017 09:10:19 +0000 (10:10 +0100)]
tests: add basic unit test for twolame as well

7 years agotests: lame: fix build
Tim-Philipp Müller [Sat, 26 Aug 2017 08:59:22 +0000 (09:59 +0100)]
tests: lame: fix build

7 years agotests: ignore another binary
Tim-Philipp Müller [Sat, 26 Aug 2017 08:52:33 +0000 (09:52 +0100)]
tests: ignore another binary

7 years agotwolame: hook up to build system
Tim-Philipp Müller [Sat, 26 Aug 2017 08:41:13 +0000 (09:41 +0100)]
twolame: hook up to build system

https://bugzilla.gnome.org/show_bug.cgi?id=774252

7 years agoMoving twolame mp2 encoder plugin from -ugly
Tim-Philipp Müller [Sat, 26 Aug 2017 08:21:44 +0000 (09:21 +0100)]
Moving twolame mp2 encoder plugin from -ugly

https://bugzilla.gnome.org/show_bug.cgi?id=774252

7 years agolame: hook up to build system
Tim-Philipp Müller [Sat, 26 Aug 2017 08:03:08 +0000 (09:03 +0100)]
lame: hook up to build system

https://bugzilla.gnome.org/show_bug.cgi?id=774252

7 years agoMoving lame mp3 encoder plugin from -ugly
Tim-Philipp Müller [Fri, 25 Aug 2017 20:13:58 +0000 (21:13 +0100)]
Moving lame mp3 encoder plugin from -ugly

https://bugzilla.gnome.org/show_bug.cgi?id=774252

7 years agomeson: Link mpeg123audiodec test against gstfft
Thibault Saunier [Sun, 20 Aug 2017 23:41:19 +0000 (20:41 -0300)]
meson: Link mpeg123audiodec test against gstfft

Fixing build error:

/run/build/gst-plugins-good/_flatpak_build/../tests/check/elements/mpg123audiodec.c:150: undefined reference to `gst_fft_s32_new'
/run/build/gst-plugins-good/_flatpak_build/../tests/check/elements/mpg123audiodec.c:151: undefined reference to `gst_fft_s32_window'
/run/build/gst-plugins-good/_flatpak_build/../tests/check/elements/mpg123audiodec.c:151: undefined reference to `gst_fft_s32_fft'
/run/build/gst-plugins-good/_flatpak_build/../tests/check/elements/mpg123audiodec.c:147: undefined reference to `gst_fft_s32_free'

7 years agotests: tagschecking: remove gst-check-xmp-* temp files when done
Tim-Philipp Müller [Sun, 20 Aug 2017 16:15:33 +0000 (17:15 +0100)]
tests: tagschecking: remove gst-check-xmp-* temp files when done

Also fix temp file creation a bit.

7 years agodocs: update for changes in git
Tim-Philipp Müller [Sun, 20 Aug 2017 14:49:12 +0000 (15:49 +0100)]
docs: update for changes in git

7 years agompg123: add to docs
Tim-Philipp Müller [Sun, 20 Aug 2017 14:48:24 +0000 (15:48 +0100)]
mpg123: add to docs

7 years agompg123: hook up to build system
Tim-Philipp Müller [Sun, 20 Aug 2017 12:56:19 +0000 (13:56 +0100)]
mpg123: hook up to build system

https://bugzilla.gnome.org/show_bug.cgi?id=774252

7 years agoMoving mpg123 plugin from -ugly
Tim-Philipp Müller [Sun, 20 Aug 2017 12:48:48 +0000 (13:48 +0100)]
Moving mpg123 plugin from -ugly

7 years agoAutomatic update of common submodule
Tim-Philipp Müller [Thu, 17 Aug 2017 11:23:25 +0000 (12:23 +0100)]
Automatic update of common submodule

From 48a5d85 to 3f4aa96

7 years agoac3parse: fix not-linked handling causing glitches when selecting stream
Sky Juan [Mon, 14 Aug 2017 07:28:22 +0000 (15:28 +0800)]
ac3parse: fix not-linked handling causing glitches when selecting stream

Fix chain function not handling not-linked from baseparse.

When an input data is separated into 2 buffers, the second buffer
would not be pushed into the adapter if baseparse returns not-linked
for first buffer.

This caused glitches when switching streams and selecting
a stream that was previously unselected.

https://bugzilla.gnome.org/show_bug.cgi?id=786268

7 years agogoom2k1: Convert source files to UTF-8
Jan Alexander Steffens (heftig) [Wed, 16 Aug 2017 11:57:50 +0000 (13:57 +0200)]
goom2k1: Convert source files to UTF-8

Causes problems with the new gtk-doc 1.26 otherwise,
but is a good idea in any case.

https://bugzilla.gnome.org/show_bug.cgi?id=786364

7 years agowavparse: Add support for growing WAV files
Eduard Sinelnikov [Mon, 14 Aug 2017 08:08:41 +0000 (03:08 -0500)]
wavparse: Add support for growing WAV files

With some fixes by me.

7 years agortpsbcpay: Fix compile error
Arun Raghavan [Mon, 14 Aug 2017 12:09:15 +0000 (17:39 +0530)]
rtpsbcpay: Fix compile error

7 years agortpsbcpay: fix if buffer size exceeds MTU
Jochen Henneberg [Mon, 14 Aug 2017 10:36:56 +0000 (10:36 +0000)]
rtpsbcpay: fix if buffer size exceeds MTU

The plugin queued buffer data if not all buffer data fit
into a single RTP packet. Now RTP packets are pushed as long
as enough data is available.

7 years agovpxenc: discard frames that have been dropped by libvpx
George Kiagiadakis [Thu, 27 Jul 2017 14:21:48 +0000 (17:21 +0300)]
vpxenc: discard frames that have been dropped by libvpx

This fixes a memory leak. When dropframe-threshold has been set,
libvpx may output less frames than the input ones, which causes
some GstVideoCodecFrames to queue up in GstVideoEncoder's internal
frame queue with no chance of ever being all released. And because
the frames keep references to the input buffers, the input buffer
pool keeps allocating new buffers and memory usage grows very fast.
For example the following pipeline's memory usage grows at a rate
of about 1GB per minute!

videotestsrc ! capsfilter caps=video/x-raw,width=1920,height=1080,framerate=30/1,format=I420 ! \
  vp8enc target-bitrate=1000000 end-usage=cbr dropframe-threshold=95 ! fakesink

https://bugzilla.gnome.org/show_bug.cgi?id=783086

7 years agortpstats: fix unsigned integer comparisons.
Mathieu Duponchelle [Tue, 8 Aug 2017 11:11:58 +0000 (13:11 +0200)]
rtpstats: fix unsigned integer comparisons.

Callers of the API (rtpsource, rtpjitterbuffer) pass clock_rate
as a signed integer, and the comparison "<= 0" is used against
it, leading me to think the intention was to have the field
be typed as gint32, not guint32.

This led to situations where we could call scale_int with
a MAX_UINT32 (-1) guint32 as the denom, thus raising an
assertion.

https://bugzilla.gnome.org/show_bug.cgi?id=785991

7 years agotaglib: use -fvisibility=hidden with this C++ plugin in meson too
Tim-Philipp Müller [Thu, 10 Aug 2017 13:44:35 +0000 (14:44 +0100)]
taglib: use -fvisibility=hidden with this C++ plugin in meson too

Also pass args as cpp_args.

7 years agoqtdemux: allow larger files
Michael Olbrich [Wed, 22 Mar 2017 14:25:17 +0000 (15:25 +0100)]
qtdemux: allow larger files

For really long files such as contiguous recordings of a whole day, the
50MB limit is not sufficient.

https://bugzilla.gnome.org/show_bug.cgi?id=781458

7 years agoqtdemux: Fix offsets for reading lpcm specific fields
Sebastian Dröge [Thu, 10 Aug 2017 13:08:06 +0000 (16:08 +0300)]
qtdemux: Fix offsets for reading lpcm specific fields

We were reading at the completely wrong positions, 16 bytes later in the
data.

Also add support for high-aligned samples.

7 years agomeson: don't export symbols by default
Tim-Philipp Müller [Thu, 10 Aug 2017 13:01:09 +0000 (14:01 +0100)]
meson: don't export symbols by default

Only plugin entry points should be exported.

Currently plugins might export more symbols with
the meson build, as we don't have the exports
regexp there that we pass to libtool.

7 years agowavparse: Fix memory leak in wavparse element
Deepak Srivastava [Thu, 10 Aug 2017 09:44:31 +0000 (15:14 +0530)]
wavparse: Fix memory leak in wavparse element

Fixing of leaking the text field of the GstWavParseNote and
GstWavParseLabl structure.

https://bugzilla.gnome.org/show_bug.cgi?id=785429

7 years agov4l2bufferpool: Don't mark jpeg frames as deltas
Cyril Lashkevich [Tue, 8 Aug 2017 10:37:12 +0000 (10:37 +0000)]
v4l2bufferpool: Don't mark jpeg frames as deltas

JPEG formats are encoded, but they never have keyframe flag. But in
fact they are keyframes

https://bugzilla.gnome.org/show_bug.cgi?id=785990

7 years agoosxvideo: rename library according to the plugin name
Philippe Normand [Sun, 6 Aug 2017 12:06:45 +0000 (13:06 +0100)]
osxvideo: rename library according to the plugin name

https://bugzilla.gnome.org/show_bug.cgi?id=785880

7 years agov4l2bufferpool: Don't drop buffer ref on qbuf
Nicolas Dufresne [Wed, 2 Aug 2017 21:16:21 +0000 (17:16 -0400)]
v4l2bufferpool: Don't drop buffer ref on qbuf

This function no longer take ownership of the buffer.

CID 1414800

7 years agov4l2: Enable VP9 format
Nicolas Dufresne [Wed, 2 Aug 2017 21:13:55 +0000 (17:13 -0400)]
v4l2: Enable VP9 format

This was missing, preventing the encoder and decoder to work
properly. This also adds support for camera that would produce
VP9 (if that exists).

7 years agov4l2: Remove spurious CATEGORY_EXTERN
Nicolas Dufresne [Wed, 2 Aug 2017 16:28:38 +0000 (12:28 -0400)]
v4l2: Remove spurious CATEGORY_EXTERN

These have been copy pasted all over the place and are not used anymore.
All object have it's own category now. This fixes build warning since
the VP9 decoder had vp8 category declared.

7 years agov4l2videoenc: Move the profile/level negotation in the base class
Nicolas Dufresne [Wed, 2 Aug 2017 14:39:46 +0000 (10:39 -0400)]
v4l2videoenc: Move the profile/level negotation in the base class

This removes duplicated code across different codec.

7 years agov4l2videoenc: Turn gst_v4l2_is_video_enc into a helper
Nicolas Dufresne [Wed, 2 Aug 2017 13:36:08 +0000 (09:36 -0400)]
v4l2videoenc: Turn gst_v4l2_is_video_enc into a helper

This reduces the amount of code needed in each codec class.

7 years agov4l2: Add VP8/9 encoder support
Nicolas Dufresne [Tue, 1 Aug 2017 20:01:11 +0000 (16:01 -0400)]
v4l2: Add VP8/9 encoder support

7 years agov4l2object: Use mmap64 to match libv4l2 signature
Nicolas Dufresne [Mon, 31 Jul 2017 15:56:05 +0000 (11:56 -0400)]
v4l2object: Use mmap64 to match libv4l2 signature

https://bugzilla.gnome.org/show_bug.cgi?id=785628

7 years agov4l2bufferpool: Copy flags and timestamp when importing
Nicolas Dufresne [Tue, 1 Aug 2017 13:22:43 +0000 (09:22 -0400)]
v4l2bufferpool: Copy flags and timestamp when importing

Whenever we import from downstream pool (userptr or dmabuf-import), we
should copy over the flags and timestamp, otherwise downstream will not
get proper synchronization or will not be able to notice frames that has
corruption in it.

https://bugzilla.gnome.org/show_bug.cgi?id=785680

7 years agov4l2: Add H263 Encoder support
Nicolas Dufresne [Mon, 31 Jul 2017 20:09:30 +0000 (16:09 -0400)]
v4l2: Add H263 Encoder support

7 years agov4l2: Add missing no-inst header
Nicolas Dufresne [Thu, 27 Jul 2017 17:51:25 +0000 (13:51 -0400)]
v4l2: Add missing no-inst header

7 years agov4l2: Add interface for MPEG4 encoding
Nicolas Dufresne [Wed, 26 Jul 2017 19:18:01 +0000 (15:18 -0400)]
v4l2: Add interface for MPEG4 encoding

7 years agov4l2: Ignore register issue and keep probing
Nicolas Dufresne [Thu, 27 Jul 2017 14:51:07 +0000 (10:51 -0400)]
v4l2: Ignore register issue and keep probing

Don't stop registering the other dynamic plugins if one registration
fails.

7 years agomulawdec: Unmap input buffer if failing to map the output buffer
Sebastian Dröge [Thu, 27 Jul 2017 11:21:34 +0000 (14:21 +0300)]
mulawdec: Unmap input buffer if failing to map the output buffer

7 years agoalawdec: Fix Memory leak in error case
Satya Prakash Gupta [Thu, 27 Jul 2017 03:52:25 +0000 (09:22 +0530)]
alawdec: Fix Memory leak in error case

https://bugzilla.gnome.org/show_bug.cgi?id=785435

7 years agov4l2: Update external files with latest
Nicolas Dufresne [Thu, 27 Jul 2017 00:36:15 +0000 (20:36 -0400)]
v4l2: Update external files with latest

This is copied from the linux kernel with only some include changes so
it works outside the kernel headers.

7 years agomatroskamux: For audio tracks, take the default duration from the first buffer
Sebastian Dröge [Tue, 18 Jul 2017 07:41:40 +0000 (10:41 +0300)]
matroskamux: For audio tracks, take the default duration from the first buffer

... if we don't have any better idea from the caps. This allows writing
SimpleBlocks for a majority of audio streams where the duration of
frames is usually fixed. And as a side effect, allows VLC to play
streams with Opus as it only works with SimpleBlocks currently:
  https://trac.videolan.org/vlc/ticket/18545

https://bugzilla.gnome.org/show_bug.cgi?id=784969

7 years agov4l2: Fix compilation without libv4l2
Nicolas Dufresne [Mon, 24 Jul 2017 20:45:40 +0000 (16:45 -0400)]
v4l2: Fix compilation without libv4l2

7 years agov4l2: Keep ref to element in allocator/pool
Nicolas Dufresne [Mon, 24 Jul 2017 20:13:56 +0000 (16:13 -0400)]
v4l2: Keep ref to element in allocator/pool

Removes the FIXME/Question in the buffer pool and add a ref to the
element in the GstAllocator too. This ref is strictly required to keep
the GstV4l2Object structure around.

7 years agov4l2object: Removed unused members
Nicolas Dufresne [Mon, 24 Jul 2017 18:27:05 +0000 (14:27 -0400)]
v4l2object: Removed unused members

7 years agov4l2: Add run-time environment to enable libv4l2
Nicolas Dufresne [Mon, 24 Jul 2017 18:19:02 +0000 (14:19 -0400)]
v4l2: Add run-time environment to enable libv4l2

The library has started preventing a lot of interesting use cases,
like CREATE_BUFS, DMABuf, usage of TRY_FMT. As the libv4l2 is totally
inactive and not maintained, we decided to disable it. As a convenience
we added a run-time environment that let you enable it for testing.

  GST_V4L2_USE_LIBV4L2=1

This of course only works if you have enabled libv4l2 at build time.

7 years agojpegenc: declare quality property changeable in PLAYING state
Nicola Murino [Mon, 17 Jul 2017 08:04:02 +0000 (10:04 +0200)]
jpegenc: declare quality property changeable in PLAYING state

https://bugzilla.gnome.org/show_bug.cgi?id=785012

7 years agov4l2object: Fix colorimetry validation
Nicolas Dufresne [Sat, 22 Jul 2017 03:34:59 +0000 (23:34 -0400)]
v4l2object: Fix colorimetry validation

While not documented, gst_video_colorimetry_matches() only accepts well
known names. Looking at the code and unit test, this seems to be on
purpose, so fixing by parsing the string and compating the colorimetry
structures.

7 years agov4l2encoder: Fix negotiation error handling
Nicolas Dufresne [Fri, 21 Jul 2017 19:40:24 +0000 (15:40 -0400)]
v4l2encoder: Fix negotiation error handling

The subclass negotiated function will call set_format, if that fails the
pool will not be created. We ended up with an assertion.

GStreamer-CRITICAL **: gst_buffer_pool_set_active: assertion 'GST_IS_BUFFER_POOL (pool)' failed

7 years agov4l2src: Speedup camera startup by skipping try_fmt
Nicolas Dufresne [Thu, 20 Jul 2017 02:25:49 +0000 (22:25 -0400)]
v4l2src: Speedup camera startup by skipping try_fmt

In this commit, we enabled skip_try_fmt_probes quirk in order to speed
up the start which is known to be disastrously slow with certain USB
cameras.

This has the side effect that we needed to rewrite the entire
negotiation process in a way that we iterate over the possible caps
until we find one that works.

The new negotiation method consist of extracting a preferred structure
from the peer caps and using this to fixate and sort the caps. To
reflect the old behaviour, we sort all resolution strictly bigger
to the preferred one with the closes one first. The rest is appended,
keeping the same order. We then normalize the caps in case there was
some list of interlace-mode or colorimetry left. We finally iterate
over all fixed caps and try it. 99% of the time, the first or the
second one should work, whit the result of a single S_FMT being issues.

From there, it will be relatively easy to introduce new negotiation
algorithm. The current algorithm is made for optimal image quality
with a scaling sink that sets it's window resolution as preference.
This the case if for:

  v4l2src ! videoconvert ! videoscale ! ximagesink

Other strategy would be needed to optimize for non-scaling sink like
ximagesink or kmssink when the driver does not scale.

https://bugzilla.gnome.org/show_bug.cgi?id=785156

7 years agov4l2object: Introduce quirk to skip slow probes
Nicolas Dufresne [Thu, 20 Jul 2017 02:09:38 +0000 (22:09 -0400)]
v4l2object: Introduce quirk to skip slow probes

skip_try_fmt_probes quirk is set, V4L2 object will not probe for
interlace-mode and colorimetry to avoid relying on try_fmt. This quirk
will be used by v4l2src to avoid desastrous startup time with slow
USB webcams.

When this quirk is enabled, caller will have to iterate over the
negotiated caps as it may contains unsupported formats. If the peer
didn't choose a specific interlace-mode, or colorimetry, the value
chosen by the driver is set into the caps. For this reason, when this
mode is enabled, gst_v4l2_object_set_format() will require writable
caps.

https://bugzilla.gnome.org/show_bug.cgi?id=785156

7 years agov4l2object: always set the GstV4l2Error on error
Nicolas Dufresne [Thu, 20 Jul 2017 02:07:32 +0000 (22:07 -0400)]
v4l2object: always set the GstV4l2Error on error

Some of the error case were conditional to using try_fmt or not.
This is slightly unexpected, always set the error so the caller
can decide.

https://bugzilla.gnome.org/show_bug.cgi?id=785156

7 years agov4l2object: Minor style fix and useful trace
Nicolas Dufresne [Thu, 20 Jul 2017 02:05:49 +0000 (22:05 -0400)]
v4l2object: Minor style fix and useful trace

https://bugzilla.gnome.org/show_bug.cgi?id=785156

7 years agov4l2object: Fix try/s_fmt errors
Nicolas Dufresne [Thu, 20 Jul 2017 02:03:29 +0000 (22:03 -0400)]
v4l2object: Fix try/s_fmt errors

According to the spec,TRY_FMT cannot return EBUSY, though it can
return EINVAL if it was not possible to update the format to
something supported.

https://bugzilla.gnome.org/show_bug.cgi?id=785156

7 years agov4l2object: Validate colorimetry in S/TRY_FMT
Nicolas Dufresne [Thu, 20 Jul 2017 02:01:26 +0000 (22:01 -0400)]
v4l2object: Validate colorimetry in S/TRY_FMT

This is in preparation for removing slow TRY_FMT probes for
colorimetry. As we won't have tried that colorimetry we cannot
assume the driver will accept it.

https://bugzilla.gnome.org/show_bug.cgi?id=785156

7 years agov4l2object: Validate field in S/TRY_FMT
Nicolas Dufresne [Thu, 20 Jul 2017 01:56:14 +0000 (21:56 -0400)]
v4l2object: Validate field in S/TRY_FMT

This is in preparation from removing the slow TRY_FMT probes for
interlacing. As we won't have tried that interlace-mode already
we need to validate that the driver isn't refusing it.

https://bugzilla.gnome.org/show_bug.cgi?id=785156

7 years agotests: icles: fix build
Tim-Philipp Müller [Fri, 21 Jul 2017 18:01:19 +0000 (19:01 +0100)]
tests: icles: fix build

Can't do additions/subtractions on void* pointers.

7 years agotests:icles: Fix previous patch by implementing our memmem
Thibault Saunier [Fri, 21 Jul 2017 15:04:17 +0000 (11:04 -0400)]
tests:icles: Fix previous patch by implementing our memmem

Using the string version of it will fail on '\0'.

7 years agotests:icles: Do not use memmem GNU extension function
Thibault Saunier [Fri, 21 Jul 2017 14:17:00 +0000 (10:17 -0400)]
tests:icles: Do not use memmem GNU extension function

As it is not avalaible on windows/msvc and we can use pure GLib for that

7 years agodirectsound: Fix .c file name in Makefile
Nicolas Dufresne [Thu, 20 Jul 2017 21:21:05 +0000 (17:21 -0400)]
directsound: Fix .c file name in Makefile

This was broken by accident, bad search and replace.

7 years agowaveform: Fix DLL name to match plugin name
Nicolas Dufresne [Thu, 20 Jul 2017 15:02:10 +0000 (11:02 -0400)]
waveform: Fix DLL name to match plugin name

https://bugzilla.gnome.org/show_bug.cgi?id=785168

7 years agodirectsound: Fix DLL name to match plugin name
Nicolas Dufresne [Thu, 20 Jul 2017 14:38:32 +0000 (10:38 -0400)]
directsound: Fix DLL name to match plugin name

https://bugzilla.gnome.org/show_bug.cgi?id=785168

7 years agoqtdemux: preferably send open-ended segment rather than repeated segment events
Mark Nauwelaerts [Wed, 19 Jul 2017 10:38:03 +0000 (12:38 +0200)]
qtdemux: preferably send open-ended segment rather than repeated segment events

7 years agoqtdemux: fix seeking in fragmented file without mfra random access info
Mark Nauwelaerts [Wed, 19 Jul 2017 09:27:32 +0000 (11:27 +0200)]
qtdemux: fix seeking in fragmented file without mfra random access info

... which no longer worked due to unconditionally clearing sample info and
ending up in inconsistent state.  Let's tread a bit more carefully and also
allow for the old seek handling that resorts to scanning if no mfra info
is available.

7 years agov4l2videodec: add some useful debug messages
Nicolas Dechesne [Wed, 19 Jul 2017 08:42:46 +0000 (10:42 +0200)]
v4l2videodec: add some useful debug messages

Add a couple of useful debug traces , they happened to be useful to
debug/investigate a 4K video playback issue with v4l2, so let's make these
changes more permanent.

Signed-off-by: Nicolas Dechesne <nicolas.dechesne@linaro.org>
https://bugzilla.gnome.org/show_bug.cgi?id=785109

7 years agov4l2: Fix 4K colorimetry
Nicolas Dufresne [Tue, 18 Jul 2017 15:28:37 +0000 (11:28 -0400)]
v4l2: Fix 4K colorimetry

Since 1.6, the transfer function for BT2020 has been changed from BT709
to BT2020_12. It's the same function, but with more precision. As a side
effect, the V4L2 colorpsace didn't match GStreamer colorspace. When
GStreamer ended up making a guess, it would not match anything supported
by V4L2 anymore. This this by using BT2020_12 for BT2020 colorspace and
BT2020 transfer function in replacement of BT709 whenever a 4K
resolution is detected.

7 years agov4l2object: Only check CROPCAP for par once
Nicolas Dufresne [Fri, 14 Jul 2017 20:21:38 +0000 (16:21 -0400)]
v4l2object: Only check CROPCAP for par once

The pixel aspect ratio is documented to not change unless the TV
Standard is changed. So this mean that this will be uniform across all
possible format and resolutions.

https://bugzilla.gnome.org/show_bug.cgi?id=784674

7 years agoRevert "matroskamux: adjust unit test to modified behaviour"
Sebastian Dröge [Tue, 18 Jul 2017 07:01:13 +0000 (10:01 +0300)]
Revert "matroskamux: adjust unit test to modified behaviour"

This reverts commit 8fe478c8a7746cd2c63f20d23e97e26e1a0e6192.

We're back to previous behaviour

7 years agomatroskamux: add properties to control cluster duration
Nicola Murino [Mon, 17 Jul 2017 22:26:11 +0000 (00:26 +0200)]
matroskamux: add properties to control cluster duration

https://bugzilla.gnome.org/show_bug.cgi?id=784971

7 years agov4l2: UVC driver is named uvcvideo these days
Nicolas Dufresne [Tue, 18 Jul 2017 00:47:26 +0000 (20:47 -0400)]
v4l2: UVC driver is named uvcvideo these days

The quirk to avoid probing interlacing didn't work anymore as the driver
is now name uvcvideo. This should slightly speed up camera startup.

7 years agov4l2object: Remove unused defines
Nicolas Dufresne [Thu, 13 Jul 2017 01:02:39 +0000 (21:02 -0400)]
v4l2object: Remove unused defines

7 years agov4l2: Make gst_v4l2_get_capabilities static
Nicolas Dufresne [Thu, 13 Jul 2017 00:53:51 +0000 (20:53 -0400)]
v4l2: Make gst_v4l2_get_capabilities static

It's not used outside of v4l2_calls.c