rtspsrc: proxy new "add-reference-timestamp-meta" property from rtpjitterbuffer

When syncing to an RFC7273 clock this will add the original
reconstructed reference clock timestamp to buffers in form
of a GstReferenceTimestampMeta.

This is useful when we want to process or analyse data based
on the original timestamps untainted by any local adjustments,
for example reconstruct AES67 audio streams with sample accuracy.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1964>

rtpbin: proxy new "add-reference-timestamp-meta" property from rtpjitterbuffer

When syncing to an RFC7273 clock this will add the original
reconstructed reference clock timestamp to buffers in form
of a GstReferenceTimestampMeta.

This is useful when we want to process or analyse data based
on the original timestamps untainted by any local adjustments,
for example reconstruct AES67 audio streams with sample accuracy.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1964>

rtpjitterbuffer: add "add-reference-timestamp-meta" property

When syncing to an RFC7273 clock this will add the original
reconstructed reference clock timestamp to buffers in form
of a GstReferenceTimestampMeta.

This is useful when we want to process or analyse data based
on the original timestamps untainted by any local adjustments,
for example reconstruct AES67 audio streams with sample accuracy.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1964>

wpe: Reintroduce persistent WebContext

A WebContext leak was introduced in MR
https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2252.
If we wanted one WebContext per WebView we should also unref the
WebKitWebContext when destroying the WebView.

This patch reintroduces the persistent WebContext, initially part of
https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1484.

Fixes #1084

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1933>

opusenc: change default bitrate-type from cbr to constrained-vbr

Which is the default in libopus itself as well, with a comment
that constrained-vbr is considered "safer for real-time use".

Unclear why CBR was the default in the first place.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1451>

webrtcbin: fix deadlock when setting up FEC encoder

We bind transceivers' fec_percentage property to the FEC encoder
percentage property, and with the binding bidirectional a deadlock
was introduced by the latest changes from !1762:

We take hold of the transceiver's object lock, then add the binding
and set the property to its initial value on the encoder, which causes
set_property to deadlock in the transceiver when the binding kicks in.

Changing the binding type to DEFAULT (source to target) is enough
to address the deadlock and still serves the original intent.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1967>

doc: handy-elements: fix audiotestsrc description

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1966>

webrtcbin: Check data channel transport for notifying 'ice-gathering-state'

Previously, it did not care about data channel's. It is fixed by adding
some conditions.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1957>

rtpbuffer: The out args for rtp extension data are optional

The code checks that these are != NULL before dereferencing them.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1962>

v4l2videodec: safely retrun from video_dec_loop with stream unlock

This is to avoid decoder hang when doing trick play between
different resolutions.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1960>

rtpjitterbuffer: Improve accuracy of RFC7273 clock time calculations

Previously the result of the calculations included inaccuracies caused
by the NTP clock estimation, which caused the timestamps to jitter
+/- 1/clockrate.

By reorganizing the calculations it is possible to get rid of this
inaccuracy and calculate deterministic and exact packet timestamps based
on the actual NTP clock as long as the estimation is not off by more
than 2**31 clockrate units.

The only remaining inaccuracy that is introduced now is caused by the
conversion from the NTP clock to the pipeline clock.

Also split up debug output, demote many messages to the trace debug
level and output more intermediate results.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1955>

twcc: Add some logging to debug TWCC feedback

This should allow people to debug when TWCC feedback is not enabled
because they haven't set the extmap in the caps.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1952>

twcc: Note that packet-loss-pct can count reordering as loss

This is difficult to encounter in ordinary networks, but is
encountered when using tc-netem to add random delays to packets, and
also when your UDP stream is bonded over multiple links with varying
characteristics.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1952>

nvcodec: Move CUDA <-> GL, D3D11, NVMM copy function to utils

This method can be used in other elements as well.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1945>

nvh265sldec: Add support for delayed output

Delay 4 frames in case of non-live to improve throughput

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1937>

h265decoder: Add support for delayed output

Functionally identical to the other decoder baseclasses.
Delayed output can improve throughput depending on decoding APIs.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1937>

h265decoder: Update documentation

Sync up with other baseclasses

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1937>

h265decoder: Drain decoder on new_sequence()

Holding previously decoded but not outputted pictures even after
new_sequence is not a safe approach in various aspect.
However, we cannot drain out DPB on new_sequence() unconditionally,
because there is a case where decoder should drop decoded pictures
if NoOutputOfPriorPicsFlag is set.

To detect NoOutputOfPriorPicsFlag before the new_sequence() call,
this patch splits decoding process into two path, one for nal unit parsing
in order to detect NoOutputOfPriorPicsFlag and then each nal unit
will be decoded.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1937>

h265decoder: Remove unused pts variable

We can know timestamp from associated GstVideoCodecFrame

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1937>

convertframe: Add support for GL-memory backend GstFrame input

Add "gldownload" early in the pipeline so that GL-memory backed raw
frames can be downloaded and processed on the CPU.

Closes: #1073
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1916>

rtprtxsend: don't require clock-rate in caps

For multiplexing, the rtpstreams you are multiplexing might not use
the same clock-rate.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1881>

rtprtxsend: don't start the task unless we are doing rtx

The rtxsend element can do pass-through when not enabled (no pt-map set)
and in those cases there is no point in starting an additional task
that does absolutely nothing.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1880>

Bump Meson requirement to >=0.60.0

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1934>

Meson: Set install_tag on some files

Meson tries to guess the tag (runtime, devel, etc) for every installed
file, but it cannot guess them all. There is a list at the end of
meson-log.txt of files we need to tag manually.

See https://mesonbuild.com/Installing.html#installation-tags.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1934>

rtprtxreceive: add ssrc-map property

Mirroring the rtxsend, this allows the application to "pre-map" the
retransmission-ssrcs to the "real" ssrc, if this information is known.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1878>

mpg123: Add gapless playback support

Co-authored-by: Sebastian Dröge <sebastian@centricular.com>
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1028>

avaudenc: Add support for AV_PKT_DATA_SKIP_SAMPLES side data

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1028>

avauddec: Add clipping meta support for gapless playback

Co-authored-by: Sebastian Dröge <sebastian@centricular.com>
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1028>

mpegaudioparse: Support gapless playback

Gapless playback is handled by adjusting buffer timestamps & durations
and by adding GstAudioClippingMeta.

Support for "Frankenstein" streams (= poorly stitched together streams)
is also added, so that gapless playback support doesn't prevent those
from being properly played.

Co-authored-by: Sebastian Dröge <sebastian@centricular.com>
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1028>

deinterlace: scalerbob: Reduce latency to 0

We only need the current field, just like `linear`.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1926>

yadif: Fix CHECK macro for YUY2 format

Used to make comb artifacts for videotestsrc pattern=ball for YUY2
format only (not AYUV).

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1938>

nvenc: Fix deadlock because of too strict buffer pool size

The pool size might need to be larger than encoding surface pool size.
Also, because we always copy input frame into internal CUDA memory,
there's no reason to restrict max size of buffer pool.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1939>

va: Fix a regression because of "Invert video codec frame dependency".

1. Always set the according GstVaH264EncFrame pointer when GstVideoCodecFrame
   pointer is assigned, which can make the logic safe.
2. Fix the forgotten change in _sort_by_frame_num. Its input pointer now is
   GstVideoCodecFrame type.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1935>

codecs: Rename picture clear functions

Our convention for clear method is gst_clear_foo_bar().

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1897>

windows: increase the timeout to create the windows docker image

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1912>

ci: force a new windows image to update cerbero's bootstrap

See:https://gitlab.freedesktop.org/gstreamer/cerbero/-/merge_requests/743#note_1270646

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1912>

nvh265sldec: Always fill SPS/PPS related parameters

Address compare was not a valid approach since it works
only if SPS/PPS id are changed. Otherwise it will always point to
the same address of member variables of h265parser.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1931>

doc: New cropping parameters added to v4l2src

v4l2src add several new parameters to control cropping of
the captured video stream.  Update the doc cache to reflect
this.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1089>

examples: v4l2: Add v4l2src crop example

Add a simple utility to illustrate how to set input cropping on v4l2src.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1089>

v4l2src: Add support for cropping at capture source input

Add properties to control input cropping in the V4L2 device.
The input cropping is applied before composing the result to the
capture buffer.  By default the capture size will be set to the same
size as the crop region, but it can be scaled to a different output
frame size if supported by the V4L2 device.
If scaling is not supported, the cropped image will
be composed as is into the top-left corner of the capture buffer.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1089>

v4l2object: Add function to get crop regions from device

Get the current crop bounding region from the V4L2 device so
that it can be provided to applications and used to validate
crop settings. Also make the default crop region available so
that it can be used to reset the crop when appropriate.

Uses the selection API when available with fallback to the crop
API for older kernels.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1089>

v4l2object: rename crop function to reflect its usage

The gst_v4l2_object_set_crop() is used for removing buffer
alignment padding. Give it a name that better reflects
that usage.  This helps to distinguish from cropping of the
input image (e.g. cropping at the image sensor on a captre
device), which can be  unrelated to the memory buffer padding,
especially if scaling is involved.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1089>

playbin3: Remove stale code

Remove now-unused get_stream_type_for_event() function.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1900>

decodebin3: Reset parsebin when new caps arrive

Check if parsebin can handle the new caps, and if not reset it so that it can
reconfigure itself for the new stream format.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1900>

decodebin3: Convert checks to assertions

"decodebin.input" is never resetted and should always be present, therefore make
it an assertion check

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1900>

parsebin: Implement ACCEPT_CAPS handling

The default query handler would go through typefind, which by default accepts
any CAPS. But once configured, parsebin can't reconfigure itself, it should
therefore pass through the ACCEPT_CAPS query to the first element after
typefind (if any).

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1900>

playbin3: Hold playbin lock on pad-added

Take the playbin lock when accessing the combiner
to add a new pad to link to. Fixes races against
streams-selected messages triggering reconfiguration.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1900>

playbin3: Reconfigure on streams-selected message.

Don't reconfigure outputs when the select-streams
event is sent from the app, as the selection may
not take effect for some time. Instead, wait
for the pipeline to confirm the new set of
selected streams when it sends the message.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1900>

playsink: Fix reconfiguration after removing text_sink

If we previously had subtitles coming in, the video
may be chained through a text overlay block. Before,
the code would end up trying to link pads that were
already linked and video would not get reconnected
properly.

To fix that, make sure that the candidate
pads are actually unlinked first. If a textoverlay
is present and no longer needed, it will be cleaned
up later in the reconfiguration sequence.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1900>

playsink: Complete reconfiguration on pad release.

Requesting a new pad can start a reconfiguration cycle, where
playsink will block all input pads and wait for data on them
before doing internal reconfiguration. If a pad is released,
that reconfiguration might never trigger because it's now waiting
for a pad that doesn't exist any more.

In that case, complete the reconfiguration on pad release.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1180>

h264decoder: Fix invalid memory access

gst_h264_dpb_needs_bump() can be called with null picture
in case of live

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1928>

rtponviftimestamp: add support for using reference timestamps

Make it posible to configure the element to obtain the timestamps from
reference timestamp meta data instead of using the ntp-offset property,
or estimating its own offset. Currently the only time format supported
is "timestamp/x-unix", i.e. UTC time expressed in the unix time epoch.

In addition the custom event GstNtpOffset has been renamed to
GstOnvifTimestamp, to reflect that it is not necessarily used to convey
the ntp-offset. As a consequence we had to modify a couple of files in
the rtsp-server as well.

Fixes #984

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1683>

pbutils: Add function to parse RFC 6381 codecs field

This is the opposite of `gst_codec_utils_caps_get_mime_codec()`, which allows
elements to get the `GstCaps`

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1894>

sdpdemux: add media attributes to caps to fix ptp clock handling

Those are needed by rtpjitterbuffer to do the right thing, e.g.

a=ts-refclk:ptp=IEEE1588-2008:00-**-**-**-**-**-**-**:0
a=mediaclk:direct=1266592257

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1924>

rtpvp8depay: Fix crash when making 'GstRTPPacketLost' custom event

This patch fixes a seg.fault in gst_structure_new() with warnings as below.

GLib-GObject-WARNING **:
 ../gobject/gtype.c:4330: type id '0' is invalid
GLib-GObject-WARNING **:
 can't peek value table for type '<invalid>' which is not currently referenced

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1918>

cudamemorycopy: add D3D11 resource support

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1807>

cudacontext: find associated DXGI Adapter LUID

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1807>

cudautils: add support of D3D11 resource as Cuda graphics resource type

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1807>

cudaloader: add D3D11 API

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1807>

mpegts: Handle glib < 2.58

By using a workaround to the lack of g_ptr_array_steal_index.

Fixes #1078

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1917>

webrtcbin: fix ulpfecenc passthrough pt

ulpfecenc uses a value of pt=255 for passthrough.

Fixes: https://gitlab.freedesktop.org/gstreamer/gstreamer/-/issues/1075
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1914>

nvh264dec,nvh265dec: Fix broken key-unit trick and reverse playback

On GstVideoDecoder::{drain,flush}, we send null packet with
CUVID_PKT_ENDOFSTREAM flag to drain out decoder. Which will
reset CUVID parser as well.
To continue decoding after the drain, the next input buffer
should include sequence headers otherwise CUVID parser will
not report any decodeable frame.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1911>

rtpbasepayload: Remove dead twcc code

This feature was removed in 7a53fbad68d702ca4905f7da223f4d2884548edc,
but this code was left behind.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1902>

rtpbin: allow FEC elements with Always pads

This patch enable picking up FEC decoder or enocder that have
static repair packets pad.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1860>

soup: Load the runtime library, not the development library

libsoup-2.4.so / libsoup-3.0.so are symlinks installed by development
packages, they are not available at runtime.

Also eliminate G_MODULE_SUFFIX since it's not useful for us, and is
actually incorrect on macOS anyway.

Fixes https://gitlab.freedesktop.org/gstreamer/gstreamer/-/issues/1071

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1899>

subparse: Handle GAP events before buffers

Make sure we did initial negotiation and segment pushing if we get GAP events
before buffers.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1903>

tagdemux: Properly propagate sequence numbers

If we received a time segment from upstream, we need to make sure we propagate
it downstream with the same sequence number.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1903>

tsbase: Handle more program updates

There could be a case where the new program has the same program number as the
previous one ... but is actually located on a PID previously used for elementary
stream. In that case the program is guaranteed to not be an update of the
previous program but a completely new one.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1893>

mpegtsbase: Use an array to track programs

We need to be able to look for programs by their PID also. Using a hash table
was a bit sub-par (and overkill) for storing a range of programs.

This is needed because there could potentially be two programs with the same
program id but different PMT PID (while one is being deactivated the new one
would "exist").

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1893>

multiqueue: Fix interleave calculation for data before segment start

This commit modifies the interleave calculation to allow growing when incoming
data is before the segment start.

The rationale is that there is no requirement whatsoever for data before the
segment start to be "coherent" on all streams.

For example, a demuxer could rightfully send data from the video stream from the
previous keyframe (potentially quite a bit before the segment start) and the
audio from just before the segment start.

This will activate the same logic as growing the interleave when some streams
haven't received buffers yet.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1892>

multiqueue: Improve interleave calculation at startup and EOS

* When a stream receives EOS, it will no longer change, we shouldn't take that
  stream into account for interleave calculation.

* When streams (re)appear, we do not want to grow the initial interleave values
  to excessive values. Instead of setting it to a default of 5s, progressively
  grow it to that maximum.

* When the status of input streams change (i.e. going to/from "some haven't
received data yet" and "all have received data"), update the interleave
immediately instead of waiting for (potentially) 5s of data before updating
it.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1892>

gst-python: gstmodule.c: fix build with 3.11

https://docs.python.org/fr/3.10/whatsnew/3.10.html

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1639>

qtdemux: Propagate stick events downstream when creating pads

If upstream provided a stream collection event before any pads were created,
make sure it's propagated downstream when pads are created.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1891>

rtprtxsend: if no rtx is present, don't expose a rtx-ssrc in caps

The point here is that rtpsession will create a new rtpsource when
the field "rtx-ssrc" is present, and when not doing rtx, that means
a random ssrc will create a new rtpsource that will be included in RTCP
messages for the current session.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1882>

vah264enc: Invert video codec frame dependency.

Instead of using GstMiniObject to hold H264 frame, now it uses a plain
structure. Besides, instead of holding a reference to
GstVideoCodecFrame, the H264 frame structure is set as a
GstVideoCodecFrame user data.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1856>

vaapih265dec: fixed st_rps_bits setting in h265 decoding.

According to va_dec_hevc.h, pic_param->st_rps_bits should be set
for accelorater to skip parsing the *short_term_ref_pic_set
(num_short_term_ref_pic_sets) structure.
Also modified fill_picture to get parser info as a parameter,
in order to get slide_hdr->short_term_ref_pic_set_size.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1886>

rtprtxsend: don't process or warn if no map is set

This makes it more gentle when doing "pass-through"

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1879>

cudamemorycopy: Fix GL resource leak

Clear GL resources on stop()

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1896>

playbin3: Add lock to protect buffering messages

Fix a small race where a group can receive stream-start
and post a pending buffering message just as another
thread posts a different buffering message, causing them
to be received by the application out of order. In the
worst case, this leads the application receiving a
stale 99% buffering message and going back to buffering
right after the 100% buffering message.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1840>

gstreamer-sharp: Add test checking AppSrc and AppSink constructors work properly

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1885>

gstreamer-sharp: Fix App{Src,Sink} constructors

Apparently GtkSharp expects each object has only one ToggleRef at any
time. Assigning element.Handle into Raw has a consequence that second
ToggleRef attempts to get created but fails on g_object_unref () that
breaks a GObject assertion:

  toggle_refs_notify: assertion failed: (tstack.n_toggle_refs == 1)

This is because toggle references should be removed with
g_object_remove_toggle_ref(), not a simple unref().

In order to avoid duplicate toggle references, introduce
ElementFactory.MakeRaw(), which creates a GstElement without its
accompanying C# object. The returned raw pointer can be assigned into
another GLib.Object without trouble.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1885>

mpegtsmux: Start last_ts with GST_CLOCK_TIME_NONE

And use the output segment position for the outgoing timestamp while it
is. This is needed to delay the calculation of `output_ts_offset` until
we actually have a usable timestamp, as tsmux will output a few initial
packets while `last_ts` is still unset.

Without this, the calculation would use the initial `0` value, which did
not have the intended effect of making VBR mode behave like CBR mode,
but always calculated an offset equal to the selected start time.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1884>

mpegtsmux: Use GST_CLOCK_STIME_NONE for output_ts_offset

It's a GstClockTimeDiff, thus GST_CLOCK_TIME_NONE isn't appropriate.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1884>

decklink: Update SDK version to 12.2.2

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1883>

rtprtxreceive: fix crash when RTX payload has zero length

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1875>

check: Fix valgrind suppression for debug function list

Fix suppression to support release and debug builds.

Here is the debug build call stack:
```
==10707==    by 0x48B5520: g_malloc (gmem.c:106)
==10707==    by 0x48D19DC: g_slice_alloc (gslice.c:1069)
==10707==    by 0x48D3947: g_slist_copy_deep (gslist.c:619)
==10707==    by 0x48D38B8: g_slist_copy (gslist.c:567)
==10707==    by 0x4ADC90B: gst_debug_remove_with_compare_func (gstinfo.c:1504)
```

In release build `g_slist_copy (gslist.c:567)` got inlined:
```
==15419==    by 0x48963E0: g_malloc (gmem.c:106)
==15419==    by 0x48AA382: g_slice_alloc (gslice.c:1069)
==15419==    by 0x48AB732: g_slist_copy_deep (gslist.c:619)
==15419==    by 0x4A39B8F: gst_debug_remove_with_compare_func (gstinfo.c:1504)
```

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1814>

rtprtxreceive: allow passthrough and non-rtp buffers

To avoid mapping rtp buffers when RTX is not in use, and to not
do a full error on receiving a non-rtp buffer, since you have no control
of what a rouge sender might send you.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1874>

rtprtx: don't access type-system per buffer

When doing only a single stream of audio/video this hardly matters,
but when doing many at the same time, the fact that you have to get
a hold of the glib global type-system lock every time you process a buffer,
means that there is a limit to how many streams you can process in
parallel.

Luckily the fix is very simple, by doing a cast rather than a full
type-check.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1873>

rtprtx: signed/unsigned and style fixes

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1872>

cudamemorycopy: Remove texture-target caps field

It's GL specific field, and we can remove it unconditionally

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1865>

cudaupload,cudadownload: Add support for dGPU NVMM

Implement NVMM <-> CUDA, GL, SYSTEM memory conversion. Jetson is
not supported yet. Note that NVMM <-> GL interop on Jetson platform
is supported by GstGL

Some example pipelines are:
- Convert NVMM to GstGL memory
  nvv4l2decoder ! "video/x-raw(memory:NVMM)" ! cudadownload ! "video/x-raw(memory:GLMemory)" ! glimagesink

- Upload system memory to NVMM and encode
  video/x-raw,format=NV12 ! cudaupload ! "video/x-raw(memory:NVMM)" ! nvv4l2h264enc

- Convert NVMM to GstCUDA memory and encode
  nvvideoconvert ! "video/x-raw(memory:NVMM)" ! cudaupload ! "video/x-raw(memory:CUDAMemory)" ! nvh264enc

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1865>

Update wrap files from latest wrapdb version

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1877>

va: encoder: Don't preallocate reconstruct buffers.

It's not required by VA to register the reconstruct buffers at context
creation, just as in decoders.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1857>

v4l2bufferpool: Fix race condition between qbuf and pool streamoff

There is a chance that pool->buffers[index] sets BUFFER_STATE_QUEUED, but
it has not been queued yet which makes pool->buffers[index] still NULL.
At this time, if pool_streamff release all buffers with BUFFER_STATE_QUEUED
state regardless of whether the buffer is NULL or not, it will cause segfault.

To fix this, also check buffer when streamoff release buffer.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1842>

gst-rtsp-server: Plug a few memory leaks in tests

Found and fixed a few memory leaks in the gst_rtspserver, gst_onvif and
gst_stream tests by running the tests in valgrind.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1742>

flvmux: Add protection when unref GstFlvMuxPad

This is to avoid gst_object_unref: assertion 'object != NULL' failed.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1843>

gstplay: tests: Keep track of errors/warnings

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1869>

gstplay: Do not error out on message parsing failures

Specially when parsing errors and warnings, the details field can be NULL and
the gst_structure_get() call would return FALSE in such cases, triggering false
positive errors.

Follow-up for #1063

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1869>

gstplay: Fix warning parsing API

The GError is an out parameter, so should be a ** parameter, like the details
parameter.

See also #1063

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1869>

cudamemorycopy: Fix build when gl is disabled

Fixes: https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/issues/1707
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1868>