sctpenc: Fix potential shutdown deadlock

When transitioning from state PAUSED to READY, the sctpenc element
could previously be stuck in an endless loop trying to resend data
in case the underlying sctp stream was in the process of
resetting. usrsctp_sendv() would repeatedly return EAGAIN with the
result that 0 bytes were sent and then sctpenc would retry forever.

To bring sctpenc out of the resend loop we just need to inform the
sink pad that it is flushing, which is already done for the associated
data queue, but we also need to set the bools associated with the
sinkpads that are used as the loop criterion.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4636>

webrtc: do not tear down data channel before data is flushed

Current implementation can in some cases detect
that all data is sent but in reality it is not,
leading to a push to an unlinked pad.
This is a race between the probe used to track data sent and a
call to close.

This patch sends an EOS before starting the close procedure
and then waits for the EOS event to come through to the
src pad before commencing with tear down.
This ensures that any queued data before EOS is flushed.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4633>

osxvideosink: fix deadlock upon closing output window

Invoking gst_osx_video_sink_osxwindow_destroy() can currently cause a deadlock
because showFrame() keeps trying to get the same lock as well. Moving the lock
closer to where it's actually needed seems to be enough to fix the issue for now.

Reported-by: Alexande B <abobrikovich@gmail.com>
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4627>

avviddec: Temporarily unlock stream lock while flushing buffers

This can call into the decoder again from other threads and try to take
the stream lock from there, which would cause a deadlock.

Fixes https://gitlab.freedesktop.org/gstreamer/gstreamer/-/issues/2558

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4625>

parse/grammar: fix missing unref of looked up child

the target parameter of gst_child_proxy_lookup() is (transfer full)

Fixes: https://gitlab.freedesktop.org/gstreamer/gstreamer/-/issues/2560
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4624>

mpegpsdemux: Rework gap sending

Take the gap logic from mpegtsdemux, and don't
send gap events on a stream that's outputting buffers with
no timestamps. Time isn't advancing, but the stream has
buffers - so it's not sparse.

Fixes #2374

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4623>

uridecodebin3: Ensure atomic urisourcebin state change

When dynamically adding and synchronizing the state of urisourcebin, we need to
ensure that no-one else attempts to change the state in case of failures

Fixes https://gitlab.freedesktop.org/gstreamer/gstreamer/-/issues/1803

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4620>

meson: Install viv-fb GL headers, needed by i.MX

Needed by qmlglsink at build time to allocate a viv-fb display.

Without this, the GL fastpath doesn't work, and performance is really
bad.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4618>

gstutils: Add category and object to most logging messages

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4615>

vavp8dec: Fix return type of decode_picture()

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4614>

vajpegdec: Hide gst_jpeg_decoder_get_type() symbol

It's not a public symbol yet

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4614>

vabaseenc: Fix return type of encode_frame vfunc

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4614>

va: Fix struct empty initialization syntax

"struct Foo bar; bar = {};" is not a valid syntax. Also remove use
of __typeof__ which is GCC specific

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4614>

qtdemux: add unit test for edit list regression

File is the mp4 file from #2549 with the mdat atom
zeroed out and compressed. We compress twice because
apparently compressing 5MB of zeroes effectively in
one run is too difficult for gzip.

https://gitlab.freedesktop.org/gstreamer/gstreamer/-/issues/2549

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4605>

Revert "qtdemux: fix conditions for end of segment in reverse playback"

This reverts commit 9deb3c27acd4161f810cd782f03bcdaccf2643c7.

The test case that was described in the associated MR
(https://gitlab.freedesktop.org/gstreamer/gst-plugins-good/-/merge_requests/262)
remains adequately fixed by a related MR that was merged later
(https://gitlab.freedesktop.org/gstreamer/gst-plugins-good/-/merge_requests/275).

It introduced incorrect logic that broke edit lists as described in
https://gitlab.freedesktop.org/gstreamer/gstreamer/-/issues/2549

Fixes: https://gitlab.freedesktop.org/gstreamer/gstreamer/-/issues/2549
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4605>

d3d11videosink: Don't clear prepared buffer on unlock_stop()

That can be called between prepare() and render() which results in
unexpected error flow return

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4596>

h264decoder: Drop nonexisting picture silently without error

If end_picture() was not successful, we do drop corresponding
GstVideoCodecFrame and therefore gst_video_decoder_get_frame()
will return nullptr which is expected behavior.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4586>

rtpmanager/rtsession: data race leading to critical warnings

This is a fix for a data race leading to:

> GLib-CRITICAL: g_hash_table_foreach:
>   assertion 'version == hash_table->version' failed

Identified sequence:

* `rtp_session_on_timeout` acquires the lock on `session` and proceeds with its
  processing.
* `rtp_session_process_rtcp` is called (debug log : received RTCP packet) and
  attempts to acquire the lock on `session`, which is still held by
  `rtp_session_on_timeout`.
* as part of an hash table iterator, `rtp_session_on_timeout` transitively
  invokes `source_caps` which releases the lock on `session` so as to call
  `session->callbacks.caps`.
* Since `rtp_session_process_rtcp` was waiting for the lock to be released, it
  succeeds in acquiring it and proceeds with `rtp_session_process_rr` which
  transitively calls `g_hash_table_insert` via `add_source`.
* After `source_caps` re-acquires the lock and gives the control flow back to
  `rtp_session_on_timeout`, the hash table iterator is changed, resulting in the
  assertion failure.

This commits copies `sess->ssrcs[sess->mask_idx]` and iterates on the copy so
the iterator is not affected by a concurrent change due to the lock being
released in the `source_caps` callback.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4585>

rtpdtmfdepay: Classify as RTP element

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4584>

rtpdtmfsrc: Classify as RTP source

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4584>

v4l2: device provider: Fix GMainLoop leak

On very quick start/stop, the mainloop may never be run. As a side
effect, our idle stop function is not really being ran, so we can't rely
on that to free the main loop. Simply unref the mainloop when the
thread have completely stop.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4539>

webrtcdatachannel: Bind to parent webrtcbin using a weak reference

The previous approach of using a simple pointer could lead to a use-after-free
in case a data-channel was created and its parent webrtcbin was disposed soon
after.

Fixes #2103

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4580>

webrtc/nice: support consent-freshness RFC7675

As is supported by libwebrtc already.  This allows ICE components to
transition to failed if consent to send from the peer is revoked or if
multiple consent packets are lost.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4575>

d3d11memory: Don't clear wrapped texture memory

The external texture may hold already rendered scene and therefore
it should not be cleared in alloc method

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4558>

qtdemux: emit no-more-pads after pruning old pads

If we don't do that, clients can rely on this signal to see the final pad
topology but it won't be the real one as some of them will disappear after
emitting that signal. This can happen after injecting a different init segment.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4557>

videoflip: fix setting of method property at construction time

Since c2f890ab, element properties are gathered from the parse-launch
line and passed at object construction.

This caused the following issue to happen in videoflip:

* videoflip installed a CONSTRUCT property named method, now deprecated
* videoflip now also overrides that property with a video-direction
  property

GObject construction causes method to be set first at construct time,
with the user-provided value, then video-direction with the default
value.

The user-provided value was thus overridden, causing a regression.

Fix by not installing the properties as CONSTRUCT, and explicitly
implementing constructed() instead in order to ensure that we do still
call gst_video_flip_set_method() at least once during construction.

Fixes https://gitlab.freedesktop.org/gstreamer/gstreamer/-/issues/2529

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4551>

mfvideoenc: Allow only even resolution numbers

Some H/W vendors support odd resolution if D3D11 texture is used
or via IMF2DBuffer, but not all vendors support it.
Also software MFT does not allow odd resolution.

Fixes: https://gitlab.freedesktop.org/gstreamer/gstreamer/-/issues/1165
Fixes: https://gitlab.freedesktop.org/gstreamer/gstreamer/-/issues/2537
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4548>

webrtcbin: Fix potential deadlock when closing before any data was sent

A blocking pad probe is added on new sink pads, it's usually removed after the
caps have been negotiated or the signaling state switched to stable, but if that
never happens and the pad is released we kept the pad probe active, leaving the
pad blocked, preventing clean disposal.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4533>

rtpmanager/rtsession: race conditions leading to critical warnings

While testing the [implementation for insertable streams] in `webrtcsink` &
`webrtcsrc`, I encountered critical warnings, which turned out to result from
two race conditions in `rtpsession`. Both race conditions produce:

> GLib-CRITICAL: g_hash_table_foreach:
>   assertion 'version == hash_table->version' failed

This commit fixes one of the race conditions observed.

In its simplest form, the test consists in 2 pipelines and a Signalling server:

* pipelines_sink: audiotestsrc ! webrtcsink
* pipelines_src: webrtcsrc ! appsrc

1. Set `pipelines_sink` to `Playing`.
2. The Signalling server delivers the `producer_id`.
3. Initialize `pipelines_src` to establish a session with `producer_id`.
4. Set `pipelines_src` to `Playing`.
5. Wait for a buffer to be received by the `appsrc`.
6. Set `pipelines_src` to `Null`.
7. Set `pipelines_sink` to `Null`.

The race condition happens in the following sequence:

* `webrtcsink` runs a task to periodically retrieve statistics from `webrtcbin`.
  This transitively ends up executing `rtp_session_create_stats`.
* `pipelines_sink` is set to `Null`.
* In `Paused` to `Ready`, `gst_rtp_session_change_state()` calls
  `rtp_session_reset()`.
* The assertion failure occurs when `rtp_session_reset` is called while
  `rtp_session_create_stats` is executing.

This is because `rtp_session_create_stats` acquires the lock on `session` prior
to calling `g_hash_table_foreach`, but `rtp_session_reset` doesn't acquire the
lock before calling `g_hash_table_remove_all`.

Acquiring the lock in `rtp_session_reset` fixes the issue.

[implementing insertable streams support]: https://gitlab.freedesktop.org/gstreamer/gst-plugins-rs/-/merge_requests/1176

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4532>

v4l2: pool: Flush events on capture queue

Unfortunately streamoff does not flush the events, and this can cause all
sort of issues. Flush events on capture queue. We also return
GST_V4L2_FLOW_RESOLUTION_CHANGE in case a resolution change was seen.
This allow skipping streamon(capture) on flush, which could lead to a
configuration miss-match, or failure if the buffers aren't of the right
size.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4525>

v4l2: videodec: Detect flushes while setting up the capture

As we missed the fact we were flushing, we could create and activate
that buffer pool, and wait on it, causing a hang. We detect that we
are flushing by checking the related pad state.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4525>

v4l2: bufferpool: Don't copy buffer when flushing

Threshold handling can race with flushing operation. This can lead to
avoidable buffer copies. Simply check and return the flushing status.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4525>

v4l2: videodec: Don't forcibly drain on resolution changes

Let the driver detects the change and reconfigure the capture side
transparently from there. This avoid reallocation of the output buffers,
and eliminates the need to stop and restart the capture task. This is
only happening if the driver have support for this, otherwise the old
behaviour is maintained.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4525>

v4l2: videodec: Remove the spurious srccaps probe

We don't need to probe the srccaps in set_format() anymore, this
handled already in the capture thread while setting up the capture
queue.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4525>

v4l2: videodec: Improve few logs

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4525>

v4l2: videodec: Only warn of incomplete drain on success

We may have hit an error, or just flushing in order to stop the thread,
in which case, not having drain everything is expected and not a
driver bug.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4525>

v4l2: bufferpool: Don't assert when orphaning is not needed

This may happen when shutting down and should not cause
any harm. This removes the associated assert when shutting
down the pipeline, notably with CTRL+C.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4525>

v4l2: videodec: Wait for source change event

Stop doing capture buffer allocation based on guesses
and wait for the source change event when available.
Unlike stateless decoder, the stateful decoder is not aware of
the coded resolution, and this may lead to the wrong result
even when using TRY_FMT.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4525>

v4l2: object: Move the GstPoll into v4l2object

Moves the GstPoll from the buffer pool into v4l2object. This will be
needed to poll for events before the pool has been created.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4525>

v4l2: object: Fix bogus debug objects pointers

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4525>

v4l2: videodec: Move the capture setup into the processing loop

In previous implementation that job was split between handle_frame and
the processing loop and it wasn't clear if this mechanism was race
free. The capture setup would also be tried for every buffer, which was
not necessary.

This also simplify the handling of SRC_CH event, dropping the unneeded
atomic boolean.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4525>

v4l2: videodec: Ensure object is inactive on failure

Sprinkle stop() calls in error case to guaranty that the capture object
is inactive on failure. Not doing so could allow some code to be called
in unexpected (and possibly undefined) conditions.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4525>

ci: Always run the windows build image job

Much like the linux counterparts, this either build the image
once the tags change or quickly exits if that has been done
already.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4474>

dtlstransport: Keep strong ref of dtls encoder/decoder

Otherwise get_property() calls for the client, certificate and/or
remote-certificate properties might access moved objects, since the encoder and
decoder are added to the transportsendbin.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4526>

gstcodectimestamper: remove PC file generation from plugin's own meson.build

The file generated here is incomplete; it is generated for all plugins in a loop at an upper level.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4520>

wasapi2: Allows process loopback capture on Windows 10

As per MS documentation[1], it requires Windows 10 Build 20348
but it seems to be supported by old versions too

[1] https://learn.microsoft.com/en-us/windows/win32/api/audioclientactivationparams/
ns-audioclientactivationparams-audioclient_process_loopback_params

Fixes: https://gitlab.freedesktop.org/gstreamer/gstreamer/-/issues/2524
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4517>

Revert "splitmuxsink: Avoid assertion when WAITING_GOP_COLLECT on reference context"

This reverts commit f29c19be5821d26a6b7682d0c69ff31d0815e072. If this is
called for the reference context then we would run into an infinite
loop, which is not really better than an assertion.

By fixing up DTS to never be ahead of the PTS in the previous commit
this situation should be impossible to hit now.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4515>

splitmuxsink: Catch invalid DTS to avoid running into problems later

DTS > PTS makes no sense, so we clamp DTS to the PTS. Also if there's a
PTS but no DTS, then assume that PTS=DTS to make sure we're not working
with a much older DTS.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4515>

cudamemory: Fix for semi planar YUV memory size decision

UV plan of the semi planar format requires only half of Y plane size

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4506>

rtspsrc: Fix handling of `*` control path

Regression introduced by 7f9d689572843ff9e0b8a92128034a8fc4a14d96.
Thanks to Tristan Matthews for reporting this.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4503>

decodebin3: Handle streams without CAPS or TIME segment

decodebin3 will do its best to figure out whether a parsebin is required to
process the incoming stream.

The problem is that for push-based stream it could happen that the stream would
not provide any caps, resulting in nothing being linked internally.

Furthermore, there is the possibility that a stream *with* caps would not be
using a TIME segment, which is required for multiqueue to properly work.

In order to fix those two issues, we force the usage of parsebin on push-based
streams:
* When the pad is linked, if upstream can't provide any caps
* When we get a non-TIME segment

Fixes #2521

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4499>

d3d11videosink: Fix for ignored initial render rectangle

Application can set target render rect before internal HWND
configuration

Fixes: https://gitlab.freedesktop.org/gstreamer/gstreamer/-/issues/2518
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4496>

bad: disable dtls test if openssl is not present

On MacOS with homebrew, the openssl library is not
properly detected with pkg-config.
So disable the test compilation if openssl
is not properly detected along with libcrypto.

libcrypto is detected but it uses the system one
which leads to the error:

your binary is not an allowed client of /usr/lib/libcrypto.dylib for
architecture x86_64

See more details from Apple:

https://developer.apple.com/forums/thread/124782

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4483>

webrtc: Plug leaks of resolved ICE addresses

The addresses we get from `resolve_host_finish()` (via
`resolve_host_async()`, `resolve_host_main_cb()`, `on_resolve_host()`,
`g_resolver_lookup_by_name_finish()`) must be freed. Otherwise we leak
memory.

Leak found and confirmed fixed with GCC AddressSanitizer.

Change-Id: If32d24452d626234f01b253b77a7d6d16eac1cee
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4482>

audiotestsrc: Initialize all samples in wave=ticks mode

Previously samples were only initialized in 2 out of 3 cases.

Probably fixes https://gitlab.freedesktop.org/gstreamer/gst-plugins-rs/-/issues/337

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4476>

d3d11compositor: Skip zero alpha input

Blending such input is a waste of resource since nothing will
be rendered

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4456>

playsink: Fix volume leak

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4463>

ccconverter: reintroduce frame count reset on cycle completion

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4453>

tsdemux: Set number of channels to 2 for dual mono Opus

Instead of leaving it at 0, which will then cause caps creation to fail.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4446>

ges: base-xml-formatter: Don't pass non-GObject pointers to GST_DEBUG_OBJECT

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4444>

dash: mpdclient: fix divide by 0 if segment has no duration

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4441>

d3d11compositor: Reconfigure resource only when output caps is changed

GstD3D11Converter setup is heavy operation since it requires
shader compile, GPU resource allocation, some math, mutex, etc.
We can avoid it if negotiated caps is not changed.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4440>

qtdemux: Fix av1C parsing

This is a regression introduced by
https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3882

The av1c codec configuration parsing would always fail due to an off-by-one
error, the content of an atom starting at offset 8 (i.e. the 9th byte) and not
9 (the 10th byte).

Also introduce a break in order to not get stray warnings

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4435>

multifile: error out if no filename was set

Fixes #2483

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4429>

qtwindow: unref caps in destructor

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4418>

v4l2: Fix use after free of fmtdesc part 2

Add missing code in merge commit e890e6e8d8bd
("v4l2: Fix use after free of fmtdesc"). The v4l2object code was
missing.

Related to https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4317

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4426>

v4l2: Fix use after free of fmtdesc

The decoder needs to force another enumeration of the format. For
this it was clearing the v4l2object insternal list, leaving a fmtdesc
pointer pointing to freed memory. This patch clears the fmtdesc pointer
that has just been free. It also makes sure the probe function does not
use the cached formats list. The probe function will restore the current
fmtdesc pointer based on the currently configured pixelformat.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4426>

v4l2: videodec: Prefer acquired caps over anything downstream

As we don't have anything smart in the fixation process, we may endup with
a format that has a lower bitdepth, even if downstream can handle higher
depth. it is notably the case when negotiating with deinterlace, which places
is non-passthrough caps before its passthrough one. This makes the generic
fixation prefer the formats natively supported by deinterlace element over
the HW 10bit format. As some HW can downscale 10bit to 8bit, this can break
10bit decoding.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4426>

v4l2: videodec: Remove leading space in comment

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4426>

tests: allocators: Fix fdmem test with recent GLib

The test failed with recent GLib, where `g_close` emits a critical
warning on EBADF. Remove the `g_close` check from `test_fdmem` and add
another version that tests `GST_FD_MEMORY_FLAG_DONT_CLOSE`.

We will depend on the Valgrind test run to warn us about leaked FDs.

Fixes: https://gitlab.freedesktop.org/gstreamer/gstreamer/-/issues/2480
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4422>

imagesequencesrc: Properly set default location

Noticed this because the generic_states test kept segfaulting at random.
GLibC 2.37 can crash when NULL is supplied as a format string.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4425>

gst: tracer: Initialize tracing infrastructure even if the debug system is not compiled in

There is a separate #define for the tracing infrastructure.

Fixes https://gitlab.freedesktop.org/gstreamer/gstreamer/-/issues/2467

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4424>

srtpdec: fix "srtp-key" check

The original code was:

if (!gst_structure_get (s, "srtp-key", GST_TYPE_BUFFER, &buf, NULL) || !buf) {
  goto error;
} else {
  stream->key = buf;
}

So use "srtp-key" if it is set so a non NULL buffer. The condition was
incorrectly inverted in ad7ffe64a66ab48d81671651031c449149db4973 to:

if (gst_structure_get (s, "srtp-key", GST_TYPE_BUFFER, &buf, NULL) || !buf) {
  stream->key = buf;
} ...

Fix the condition so it works as originally intended and avoid accessing
'buf' uninitialised.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4423>

adaptivedemux2: fix critical when using an unsupported URI

adaptivedemux2 only supports http(s), trying to use it with, say,
file:// was raising a CRITICAL in libsoup.

Fix #2476

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4421>

glvideoflip: fix leaked caps

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4420>

glcontext_wgl: fix missing unref

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4419>

d3d11videosink: fix race conditions in win32 window

One race condition is the fact that the window object
can be destroyed while running some routine in the UI
thread (such as resizing). To avoid that situation we make
UI thread hold a reference on the window object while it's
running.
Other probpematic case is when the window handle is reused:
if we stop and start the pipeline very fast,
so the sink creates a new window object that is going to use
the same window handle as the previous one.
And finally the case when the pipeline is stopped immediatelly
right after starting, this one is also handled in this commit.

NOTE: a unit test that reproduces this cases have been added
in the previous commit.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4417>

d3d11videosink: add gstcheck test for win32 window

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4417>

d3d11videosink: Enhancement for initial window size decision

Use AdjustWindowRect() method to calculate window size so that
video scene can be rendered on client area without black borders

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4416>

Back to development

Release 1.22.2

gst-plugins-base: update translations

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4390>

gstreamer: update translations

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4390>

avmux: fix element leak

Fixes #2473

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4385>

avdeinterlace: fix element leak

Fixes #2473

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4385>

h264decoder: Enable low-latency bumping in case of pic_order_cnt_type 2

In case of POC type 2, output order is equal to decoding order
(no frame reordering)

Fixes: https://gitlab.freedesktop.org/gstreamer/gstreamer/-/issues/2447
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4383>

h264decoder: Fix for latency report

The minimum latency answered by an element should be the maximum
latency from the element's perspective. Also consider max_reorder_frames
update as a sequence change

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4383>

h264decoder: Ignore invalid max_num_reorder_frames in VUI

It's not fatal and can be ignored

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4383>

h265decoder: Fix for latency report when src caps is not configured

Depending on subclass, negotiation might not happen on new_sequence()

Fixing regression introduced by the commit
4a4823b9728fbb944e05d89e0fee681ab5b33e25

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4383>

h264decoder: Fix for latency report when src caps is not configured

Depending on subclass, negotiation might not happen on new_sequence()

Fixing regression introduced by the commit
4a4823b9728fbb944e05d89e0fee681ab5b33e25

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4383>

gstd3d11window: fix memory leak

GstStructure may leak when using external HWND

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4382>

meson: add missing source_filename in the dav1d wrap

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4379>

debugqroverlay: fix string leak

g_string_free(.., FALSE) gives us ownership of the string
already, no need to duplicate that again with g_strdup(),
and doing so will leak the string returned by g_string_free()
here. Caught by compiler warnings in newer GLib versions.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4380>

twcc: Better handle duplicate packets

The previous code would only check if two packets in a row were duplicates. If
not (i.e. a packet is a duplicate of a packet received slightly before) the code
would generate completely bogus FCI because it assumes there were no duplicates
present in the array.

In order to be efficient, just store all received packets and remove the
duplicates just before the FCI is generated once the array of observations have
been sorted by seqnum.

Fixes TWCC usage with moderate to high packet duplication.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4378>

decklinkaudiosink: Fix playback when video caps is configured before audio

Scheduled playback starts on videosink's state change or on caps
but it's possible that audiosink is configure without caps yet.
Try start scheduled playback on audiosink's caps event as well

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4368>

gst-plugins-base: gl: wayland: cleanup on close

The proxy and queue are created in the gst_gl_window_wayland_egl_open()
function and will be recreated on open. This leaks both objects, the
wayland client documentation mentions that they should be destroyed
using the appropriate destroy functions.

Found during valgrind memory leak testing, these blocks were marked as
definitely lost.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4355>

osxvideosink: fix broken aspect ration and frame drawing region

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4346>

rtsp-server: media: First set state to PLAYING again temporarily, then send EOS

Sending the EOS event while the pipeline is PAUSED can deadlock on the
stream lock if a sink is currently blocked because of pre-rolling.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4345>

sdp: Skip source-specific caps fields when creating an SDP media from caps

Regression from https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2132

We parse the source-specific fields from the SDP and put it into the
caps, but when converting caps into an SDP again this would need special
handling. By default it would end up as part of the fmtp field, which is
simply wrong.

Automatically putting it into the caps and SDP will need some more work.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4344>

decklink: fix 10 bit RGB (r210) format auto detection

Fixes https://gitlab.freedesktop.org/gstreamer/gstreamer/-/issues/2391

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4342>