vaapi: libs: Use unsized arrays as parameters.

Instead of defining a sized array for function signature, use it
unsized (a pointer alias, basically). In this way clang warning is
silenced:

warning: ‘fill_profiles’ accessing 64 bytes in a region of size 12 [-Wstringop-overflow=]

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1357>

vaapi: meson: Fail if static build.

And initialize the plugin array.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1357>

env: Fix deprecations from python 3.10

distutils is now deprecated and strtobool is simple enough for us to just
vendor.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1358>

python: Avoid treating float as int

Since python 3.10 implicit conversion to integers using `__int__` as
been completely removed (was deprecated behavior in 3.9) so we need
to cleanly handle it now.

See https://gitlab.gnome.org/GNOME/pitivi/-/issues/2589

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1358>

vaapipostproc: fix mising metadata when using system

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1325>

qmlglsink: fix another resize case

If qt asks us to redraw before we have both set a buffer and caps we
would attempt to use the new caps with the old buffer which could result
in bad things happening.

Only update caps from new_caps once the buffer has actually been set.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1346>

wpevideosrc: Fix frame stuttering in GL rendering path

Make sure the EGLImage we're rendering to the GL memory stays alive long enough,
until the the GL memory has been destroyed.

This change fixes tearing and black flashes artefacts that were happening
because the EGLImage was sometimes destroyed before the sink actually rendered
the associated texture.

Fixes #889

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1354>

wpevideosrc: Run through gst-indent

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1354>

gstinfo: Add gst_debug_log_literal() function

This takes a plain message string and not a format string, and as a
result doesn't have to be passed through vasprintf() and lead to further
unnecessary allocations. It can also contain literal `%` because of
that.

The new function is mostly useful for bindings that would have to pass a
full string to GStreamer anyway and would do formatting themselves with
language-specific functionality.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1356>

v4l2: Update fmt if padded height is greater than fmt height

If padded height is greater, buffer bytesused could be larger than plane length,
and cause VIDIOC_QBUF failure.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1355>

uridecodebin3: Nullify current item after all play items are freed.

There's a potential race condition with this sort of pipelines on
certain systems (depends on the processing load):

GST_DEBUG_DUMP_DOT_DIR=/tmp \
gst-launch-1.0 uridecodebin3 uri=file://stream.mp4 ! glupload ! \
glimagesink --gst-debug=*:4

Right after the pipeline passes from PAUSED to READY, bin_to_dot_file
dumps uridecodebin3 properties, but current uri and suburi might be
already freed, causing a potential use-after-freed.

This patch makes NULL the current item right after all the play items
are freed.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1353>

gst: Fix license headers and add SPDX

Fixes https://gitlab.freedesktop.org/gstreamer/gst-python/-/issues/57

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1351>

caps: fix type of return value if string is null in gst_caps_from_string

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1348>

rtpfunnel: don't enforce twcc during upstream negotiation

A previous patch has caused rtpfunnel to output twcc-related
information downstream, however this leaked into upstream
negotiation (through funnel->srccaps), causing payloader to
negotiate twcc caps even when not prompted to do so by the user.

Fix this by only enforcing that upstream sends us application/x-rtp
caps as was the case originally.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1278>

rtptwcc: complete bufferlist fix

When dealing with bufferlists, we need to store one "SentPacket"
structure per buffer, not one per buffer list!

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1278>

qtdemux: Log cslg_shift that was determined

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1294>

qtdemux: Use a composition time offset of 0 for "no decode samples" for the time being

This needs codec-specific handling, but using 0 instead of G_MININT32 at
least gives somewhat reasonable behaviour.

See https://gitlab.freedesktop.org/gstreamer/gstreamer/-/issues/883

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1294>

qtdemux: Always check ctts for unreasonably large offsets

If this happens then ignore the whole ctts. Previously we only did this
if the PTS/DTS shift was determined from the ctts instead of the cslg.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1294>

qtdemux: Dump composition time offsets in trun as signed integers

Just like we do for ctts without regard of the version of the box.
Huge offsets are interpreted as negative offsets by qtdemux so this
works.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1294>

qtdemux: Add a comment why only positive cslg shifts are considered

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1294>

qtdemux: Only adjust segment.stop by cslg_shift if stop is not -1

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1294>

qtdemux: Handle negative composition offsets in the trun box the same way as for non-fragmented streams

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1294>

qtdemux: Parse ctts version

Negative composition time offsets are only allowed with version 1 of the
box, however we parse it as a signed value also for version 0 boxes as
unfortunately there are such files out there and it's unlikely to have
(valid) huge composition offsets.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1294>

qtdemux: Add support for version 1 cslg boxes

They use 64 bit fields instead of 32 bit.

Also parse offset as a signed integer (in both versions) and clamp it to
a positive value as negative values don't really interest us here.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1294>

qtdemux: Don't free cslg data that we don't own on corrupt files

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1294>

audio-resampler: Fix segfault when we can't output any frames

Sometimes the resampler has enough space to store all the incoming
samples without outputting anything. When this happens,
gst_audio_resampler_get_out_frames() returns 0.

In that case, the resampler should consume samples and just return.
Otherwise, we get a segfault when gst_audio_resampler_resample() tries
to resample into a NULL 'out' pointer.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1343>

matroska: Ref index table when updating track info

Track index table array was being lost during track info update.
Ref it over to updated info, so it can be used for finding
nearest seek points.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1203>

matroska: Use g_array_unref everywhere

Instead of using g_array_free which is not thread safe use g_array_unref instead

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1203>

docs: fix unnecessary ampersand, < and > escaping in code blocks

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1340>

st2022-1-fecdec: fix packet trimming

g_sequence_remove_range's end iter is exclusive, so if one
wants to remove that item as well, it should be called with
the next iter.

This could in theory fix an issue where:

* The sequence isn't entirely trimmed, with an old item lingering

* Following FEC packets are immediately discarded because they
  arrived later than corresponding media packets, long enough for
  seqnums to wrap around

* We now try to reconstruct a media packet with a completely obsolete
  FEC packet, chaos ensues.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1341>

qmlsink: support caps changes better

We need to hold onto the last buffer until the next buffer arrives.
Before, if a caps change comes we would remove the currently rendering
buffer.  if Qt asks use to render something, we would render the dummy
black texture.

Fixes a period of black output when upstream is e.g. changing resolution
as in hls adaptive bitrate scenarios.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1338>

v4l2: Record buffer states in pool to fix dequeue race

The `gst_v4l2_buffer_pool_dqbuf` function contains this ominous comment:

    /* get our GstBuffer with that index from the pool, if the buffer was
     * outstanding we have a serious problem.
     */
    outbuf = pool->buffers[group->buffer.index];

Unfortunately it is common for buffers in _output_ buffer pools to be
both queued and outstanding at the same time. This can happen if the
upstream element keeps a reference to the buffer, or in an encoder
element itself when it keeps a reference to the input buffer for each
frame.

Since the current code doesn't handle this case properly we can end up
with crashes in other elements such as:

    (gst-launch-1.0:32559): CRITICAL **: 17:33:35.740: gst_video_frame_map_id: assertion 'GST_IS_BUFFER (buffer)' failed

and:

    (gst-launch-1.0:231): GStreamer-CRITICAL **: 00:16:20.882: write map requested on non-writable buffer

Both these crashes are caused by a race condition related to releasing
the same buffer twice from two different threads. If a buffer is queued
and outstanding this situation is possible:

**Thread 1**
- Calls `gst_buffer_unref` decrementing the reference count to zero.
- The core GstBufferPool object marks the buffer non-outstanding.
- Calls the V4L2 release buffer function.
- If the buffer is _not_ queued:
  - Release it back to the free pool (containing non-queued buffers).

**Thread 2**
- Dequeues the queued output buffer.
  - Marks the buffer as not queued.
- If the buffer is _not_ outstanding:
  - Calls the V4L2 release buffer function.
  - Release it back to the free pool (containing non-queued buffers).

If both of these threads run at exactly the same time there is a small
window where the buffer is marked both not outstanding and not queued
but before it has been released. In this case the buffer will be freed
twice causing the above crashes.

Unfortunately the variable recording whether a buffer is outstanding is
part of the core `GstBuffer` object and is managed by `GstBufferPool` so
it's not as straightforward as adding a mutex. Instead we can fix this
by additionally recording the buffer state in `GstV4l2BufferPool`, and
handle "internal" and "external" buffer release separately so we can
detect when a buffer becomes not outstanding.

In the new solution:
- The "external" buffer pool release and the "dqbuf" functions
  atomically update the buffer state and determine if a buffer is still
  queued or outstanding.
- Subsequent code and a new
  `gst_v4l2_buffer_pool_complete_release_buffer` function can proceed to
  release (or not) a buffer knowing that it's not racing with another
  thread.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1010>

gst-python: Add option to disable python plugin

It is not always needed, at least Ubuntu package it separately and don't
install it by default. Also when doing a static build there is an
unavoidable warning otherwise.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1337>

avtp: crf: Process also local CRF streams

Without this patch locally generated CRF streams will be ignored.
Therefore the same network interface could not be CRF talker and
CRF listener.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1074>

webrtc janus rust: Update extra dependencies

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1332>

webrtc multiparty rust: Upgrade all other deps

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1332>

webrtc sendrecv rust: Upgrade all other deps

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1332>

webrtc multiparty rust: Port to bindings 0.17 version

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1332>

webrtc sendrecv rust: Port to bindings 0.17 version

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1332>

glcontext/egl: add missing unref

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1328>

rtpbin: separate out the two fec decoder locations

The pipeline flow for receiving looks like this:

rtpsession ! rtpssrcdemux ! session_fec_decoder ! rtpjitterbuffer ! \
  rtpptdemux ! stream_fec_decoder ! ...

There are two places where a fec decoder could be placed.
1. As requested from the 'request-fec-decoder' signal: after rtpptdemux
   for each ssrc/pt produced
2. after rtpssrcdemux but before rtpjitterbuffer: added for the
   rtpst2022-1-fecenc/dec elements,

However, there was some cross-contamination of the elements involved and
the request-fec-decoder signal was also being used to request the fec
decoder for the session_fec_decoder which would then be cached and
re-used for subsequent fec decoder requests.  This would cause the same
element to be attempted to be linked to multiple elements in different
places in the pipeline.  This would fail and cause all kinds of havoc
usually resulting in a not-linked error being returned upstream and an
error message being posted by the source.

Fix by not using the request-fec-decoder signal for requesting the
session_fec_decoder and instead solely rely on the added properties for
that case.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1300>

id3tag: Map GST_TAG_MUSICBRAINZ_RELEASETRACKID

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1331>

id3tag: Map GST_TAG_MUSICBRAINZ_RELEASEGROUPID

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1331>

id3tag: Remove trailing whitespace

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1331>

qtdemux: Fix can not demux Opus track made by qtmux

Opus stream info is read from dOps box [1]. The offset of dOps box in Opus box is different in mp4a version 1 and 0 [2]. Calculate the offset of dOps box according to mp4a version.

[1] https://opus-codec.org/docs/opus_in_isobmff.html

[2] subprojects/gst-plugins-good/gst/isomp4/atoms.c:sample_entry_mp4a_copy_data:2146

Fixed: https://gitlab.freedesktop.org/gstreamer/gst-plugins-good/-/issues/918
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1329>

docs: app-dev: events: seeking: use CLOCK_TIME_NONE instead of -1 and fix parameter names

to match the parameter names in the gst_element_seek() declaration.

Closes https://gitlab.freedesktop.org/gstreamer/gst-docs/-/merge_requests/34/

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1324>

docs: installing-on-macos: flesh out instructions what to download

Closes https://gitlab.freedesktop.org/gstreamer/gst-docs/-/merge_requests/106/

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1324>

plugin-development: basics-boilerplate: Remove unneeded meson.build edit step

Closes https://gitlab.freedesktop.org/gstreamer/gst-docs/-/merge_requests/157/

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1324>

docs: installing-on-windows: mention packages to download

Closes https://gitlab.freedesktop.org/gstreamer/gst-docs/-/merge_requests/158/

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1324>

docs: tutorials: add link to Rust version of the tutorials

Closes https://gitlab.freedesktop.org/gstreamer/gst-docs/-/merge_requests/159/

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1324>

docs: tutorials: playback-3: flesh out build instructions

Add information about how to compile this file. The code in this
tutorial also requires `gstreamer-audio-1.0`, so we cannot use
the commands from the previous tutorial.

Closes https://gitlab.freedesktop.org/gstreamer/gst-docs/-/merge_requests/164/

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1324>

docs: plugin-dev: basics-boilerplate: fix typo in variable name

Closes https://gitlab.freedesktop.org/gstreamer/gst-docs/-/merge_requests/166/

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1324>

avtpsrc: Use correct size for provided buffers

Without this patch the following pipeline would send packets containing
garbage in the data section.
$ gst-launch-1.0 avtpsrc ! avtpsink

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1077>

appsrc: log when segment changes

We were logging when it does not change but not when it does, which is
confusing when reading logs.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1327>

avtp: crfsync: Warn when CRF package not yet received

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

v4l2bufferpool: set video alignment of video meta

need apply video alignment info on video meta, downstream
element can get buffer alignment from video meta

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1326>

test: avtp: crf: Check for rounding errors

on average period calculation.

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

avtp: crf: Use double for average period calculation

to also support CRF intervals like every 1,333,333ns 64 events

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

tests: avtp: crf: Test for timestamp_interval > 1

in case of CRF AVTPDUs with single CRF timestamp.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1076>

avtp: crf: Properly handling one timestamp per PDU

The average_period should always represent the time between two
events. The specification defines the event time as the time
between audio samples, video frame sync, video line sync, etc.
In case of one timestamp per PDU the timestamp_interval identifies
the amount of events between the timestamp of one PDU and the
timestamp of the next PDU.
As described in IEEE 1722-2016 chapter
"10.4.12 timestamp_interval field" timestamp_interval shall be
nonzero.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1076>

shm: NetBSD build fix

shm_unlink() and friends live in librt on NetBSD. Adapt build system.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1323>

bin: Don't check twice for adding a bin to itself or removing it from itself

This is already covered by a `g_return_val_if_fail()` in the calling
function.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1320>

bin: Switch `g_warning()`s to `GST_WARNING_OBJECT()`s when adding/removing an element to a bin fails

The failure conditions can be overidden by subclasses, and a boolean
return value is provided to the caller whether adding/removing the child
element has actually worked. The caller can then handle this
accordingly but flooding stderr with this is not very useful.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1320>

vpxdec: Skip check of key frame when open GOP

Valid stream [1] might has no key frame.

Fixed: https://gitlab.freedesktop.org/gstreamer/gst-plugins-good/-/issues/890

[1] https://storage.googleapis.com/downloads.webmproject.org/test_data/libvpx/vp90-2-16-intra-only.webm

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1316>

gst-integration-tests: medias: update git module to latest commit

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1184>

osxaudio: Increase max channels to 64

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1059>

gstinfo: Fix leak in generate_backtrace_trace

Spotted by Laurent Pinchart.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/901>

aom: Set fixed_qp_offsets to a deactivated value

aom only uses fixed_qp_offsets with the
Constant Quality (Q) Rate Control mode,
previously this was locking any usage
with another Rate Control mode.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1198>

meson:ges: Fix typo adding ges gir to libs list

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1322>

gst-ptp-helper: Do not disable multicast loopback

Otherwise we cannot run gst-ptp-helper if the PTP master is on the
same device.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1306>

tests:ges: Take TIMEOUT_FACTOR into account

This env var is set in the CI so we grow the timeout as required
(when running in valgrind for example).

See https://gitlab.freedesktop.org/gstreamer/gstreamer/-/jobs/15449209

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1315>

magicleap: update lumin_rt libraries names to the latest official version

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1314>

codecs: h264dec: Calculate the latency by its bump mode.

The current latency calculation just uses the num_reorder_frames,
which is not very precise. We should consider the bump mode of the
DPB, the faster it bumps, the lower latency we will have.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1046>

doc: Update nvdec documentation

Fixes https://gitlab.freedesktop.org/gstreamer/gstreamer/-/issues/870

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1309>

imagefreeze: Only set caps if they do not match current caps

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1311>

h264parse: Don't insert extra AUD if exists in bitstream already

AUD nalu in packetized format is completely valid and therefore we should not
assume that we should insert AUD for packetized -> bytestream
conversion.

Fixes: https://gitlab.freedesktop.org/gstreamer/gstreamer/-/issues/862
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1296>

tests: h264parse: Add test for AUD insertion

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1296>

v4l2: Support bools for setting extra-controls

They're just mapped to 0 / 1 and can already be set as integers, but
being able to set them as booleans seems useful.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1308>

wasapi2ringbuffer: Fix client object leak

Check whether ringbuffer is holding client object already since
open_device() may be called multiple times

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1307>

webrtcbin: Clear errors from finding codec preferences before the next iteration

The media is just skipped and the error is not propagated to the caller,
so keeping it around here would cause assertions a bit later when trying
to set a new error over the old one.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1291>

webrtcbin: Move addition of attributes to the caps after making sure they're not empty or any

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1291>

webrtcbin: Don't require fixed caps when querying caps for a transceiver pad to match it with a media

Upstream caps might for example be
  application/x-rtp,media=audio,encoding-name={OPUS, X-GST-OPUS-DRAFT-SPITTKA-00, multiopus}
and while that is not fixed caps it is enough to match it with a media.

Only caps structures that have the correct structure name and that have
the media and encoding-name field are preserved, but if both are present
then these caps are used as "codec preferences".

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1291>

typefind: fix reading file extension from URI

Currently reading extension relies on the fact that everything after the
last"." character is a file extension. Whereas that works fine for most
of the cases, it breaks when the URI contains a query part.

E.g.: `http://url.com/file.mp4?param=value` returns `mp4?param=value`
instead of `mp4`.

In this commit we use URI parser to read the path of the URI (in the example
above, that is `/file.mp4`) and read extension from that path.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1305>

Back to development

Release 1.19.3

Update ChangeLogs for 1.19.3

ci: Do not clone in the trigger job

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1298>

multifilesink: Make minimum distance between keyframes in next-file=key-frame mode configurable

Previously this was hardcoded to 10s, which is not necessarily the
desired behaviour.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1292>

qtmux: Fix deadlock in gst_qt_mux_prepare_moov_recovery

Regression from 5766731bd4200c3a374522a749386f740347661a

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1288>

egl: handle configless contexts

With EGL_KHR_no_config_context, EGL contexts may just not
have an EGLConfig to give you. Deal with it.

Fixes: #858
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1289>

examples: win32-videooverlay-playbin: Add a repeat option

... in order to test pipeline/element reuse cases.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1275>

codecs: mpeg2: Drain before a new_sequence get signalled

The decoder may need to re-allocate the output buffer, it is easier if all
pictured have been outputed.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1013>

v4l2codecs: gstv4l2codecsmpeg2dec: implement a render delay

The v4l2 backend support delayed output for performance reasons.
It is then possible to use render delays to queue multiple requests
simultaneously, thus increasing performance.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1013>

codecs: gstmpeg2decoder: add support for render delay

Some decoding APIs support delayed output for performance reasons.
One example would be to request decoding for multiple frames and
then query for the oldest frame in the output queue.
This also increases throughput for transcoding and improves seek
performance when supported by the underlying backend.

Introduce support in the mpeg2 base class, so that backends that
support render delays can actually implement it.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1013>

v4l2codecs: Implement a MPEG2 V4L2 decoder element

Implement a MPEG2 V4L2 decoder element based on the previous h264
implementation.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1013>

v4l2codecs: mpeg2: update to the new uAPI

The mpeg2 stateless api has undergone changes as it is being
destage. Update the v4l2-controls header to match.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1013>

sys: va: GstVaMpeg2Dec: use slice sc_offset and size

Seeing as how GstMpeg2Slice will now record the start code offset
as well as its size with the above field taken into account, the
manual computation in this class is not needed.

Remove it.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1013>

codecs: GstMpeg2Slice: add field for sc_offset and size

Downstream might need the start code offset when decoding.

Previously this computation would be scattered in multiple sites. This
is error prone, so move it to the base class. Subclasses can access
slice->sc_offset directly without computing the address themselves
knowing that the size will also take the start code into account.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1013>

v4l2codecs: Handle allocator creation failure

If `VIDIOC_REQBUFS` doesn't return enough buffers the allocator creation
function can fail and return `NULL`. Handle this by generating an error
and returning instead of segfaulting.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1220>

v4l2codecs: Fix segfault when destroying non-detached allocator

The GstV4l2CodecAllocator dispose function clears `self->decoder` but
the finalize function then tries to use it if the allocator has no been
detached yet.

Fix by detaching in the dispose function before we clear
`self->decoder`.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1220>