va: Document device-path property.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3361>

va: Reenable plugin cache.

This basically a revert of commit 9bb560e7

Also it adds vaav1dec.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3361>

vavp8dec, vampeg2dec: Fix type name.

The camel case name is wrong when registering the type.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3361>

ci: sprinkle some date -R to easier see how long build steps take

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3354>

ci: use 8 parallel jobs for the integration test suite

The runners have a lower core count than the previous
generation so the integration tests run much slower
overall than the used to, so fix that by running more
in parallel again. The build time is measured in seconds
and much faster than it used to be, so should still be
a reasonable trade-off, and means re-tried jobs will
complete faster.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3354>

videoconvert, videoscaleconvert: fix element description

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3362>

allocator: Switch `allow-none` annotations to `nullable` / `optional`

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3364>

allocator: Copy allocator name in gst_allocator_register()

The parameter is not marked as `transfer full` and stays around in the
hash table, so we will have to copy it ourselves.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3364>

tracers: latency: document the 'reported' flag

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3359>

vacaps: Sort raw caps according their efficiency.

Caps negotiations considers template caps order. This patch sets the
correct order according the feture caps: VAMemory, DMABuf and
SystemMemory.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3349>

Back to development

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3358>

subprojects: pin gst-plugins-rs for GStreamer 1.21.2 release

Release 1.21.2

Update ChangeLogs for 1.21.2

hotdoc: gst-hotdoc-plugins-scanner is not needed for libraries

Meson >= 0.64.0 does not allow any more to add executables into
hotdoc.generate_doc(..., dependencies: ...) and it should not be needed
any way.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3352>

srt: Add a property to disable automatic reconnect

This adds a new boolean property `auto-reconnect`, defaulting to `true`.

Setting it to `false` makes the elements (in caller mode) immediately
report an error to the application instead of trying to reconnect.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3326>

discoverer: Annotate `DiscovererInfo::get_audio_streams()` and others with the concrete stream info type

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3353>

vabasetransform: Fail if cannot import the input buffer.

Otherwise got a buffer double free.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3348>

mfvideosrc: Add support for DirectShow capture filter

Adding DirecShow video capture filter mode, in addition
to existing MediaFoundation and WinRT(UWP) mode, to support
DirectShow only filters (not KS driver compatible)
such as custom virtual camera filters.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3350>

tests: gstinfo: Test set_threshold_from_string's new reset behavior

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/605>

tests: gstinfo: Make logging tests pass when GST_DEBUG is set

Use gst_debug_set_threshold_from_string's new reset behavior to undo
GST_DEBUG and ensure the logging tests have a known configuration.

`gst_debug_set_threshold_from_string ("LOG", TRUE)` has the same effect
as `gst_debug_set_threshold_from_string ("", TRUE)` followed by
`gst_debug_set_default_threshold (GST_LEVEL_LOG)`.

Don't bother remembering the default log level set when the test
started. It will get reset by the next test, anyway.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/605>

info: Reset patterns with set_threshold_from_string

TLDR: Make `gst_set_threshold_from_string ("", TRUE)` reset *all*
threshold settings, including those set by previous invocations of
`gst_debug_set_threshold_from_string`.

The docs say:

    @reset: %TRUE to clear all previously-set debug levels before setting
        new thresholds

What actually happens is it sets the default threshold to `ERROR`,
leaves the patterns in place and calls
`gst_debug_category_reset_threshold` on each category.

In effect, any category that is matched by a pattern gets reset to that
threshold if the app changed it by directly invoking
`gst_debug_category_set_threshold`. All other categories are reset to
`ERROR`.

In my opinion this parameter currently has little value, as the same
effect can be achieved by including `ERROR` (without a pattern) in the
string, as in `"foo*:WARNING,*bar:INFO,ERROR"`.

What I actually expect it to do is reset *all* threshold settings,
including those set by previous invocations of
`gst_debug_set_threshold_from_string`, starting off with a clean slate
for the patterns provided with the call.

Otherwise there is no API to do this, besides:

  - Painfully removing patterns one-by-one via
    `gst_debug_unset_threshold_for_name` *if* you know what the patterns
    are.
  - Adding a `*:FOO` pattern to affect all categories, which makes the
    default threshold useless and practically leaks all the old
    patterns.

In my opinion this also makes it fit better into the layers of threshold
config, which is:

1. Temporary:
  - `gst_debug_category_set_threshold`
  - `gst_debug_category_reset_threshold`
2. Patterns:
  - `gst_debug_set_threshold_for_name`
  - `gst_debug_unset_threshold_for_name`
  - `gst_debug_set_threshold_from_string`
  - `GST_DEBUG`
3. Default:
  - `gst_debug_set_default_threshold`

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/605>

rtmp2: Improve error messages

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1403>

rtmp2/connection: Pass triggering GError in 'error' signal

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1403>

rtmp2/connection: Pass triggering GError to _emit_error

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1403>

rtmp2/connection: Discern reasons for cancelling all commands

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1403>

rtmp2/connection: Handle EOF like error

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1403>

rtmp2/client: Make sure 'salt' is not NULL

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1403>

rtmp2/client: Make sure 'reason' is not NULL

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1403>

rtmp2/client: Make sure 'desc' is not NULL

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1403>

rtmp2/client: Make sure 'code' is not NULL

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1403>

webrtcbin: Don't duplicate enum string values

Some were leaked when debugging was enabled. Instead just directly use the
static strings as-is.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3347>

gstpad: Avoid race in (un)setting EOS flag on sinkpads

The scenario is the following:

* Thread 1 is pushing an EOS event on a sinkpad
* Thread 2 is pushing a STREAM_START event on the same sinkpad before Thread 1
returns. Note : It starts pushing the event after Thread 1 took the object lock.

There is a potential race between:

* The moment Thread 1 sets the EOS flag once it has finished sending the
event (via store_sticky_event). When it does that it has both the STREAM and
OBJECT lock

* The moment Thread 2 sends the STREAM_START event (Which should release that
EOS status), but removing the EOS flag is only done while holding the OBJECT
lock and not the STREAM_LOCK, which means it could be re-set by Thread 1 before
it then checks again the EOS flag (without the STREAM lock taken).

The EOS flag unsetting by STREAM_START should be done with the STREAM lock
taken, otherwise it will be racy.

Fixes https://gitlab.freedesktop.org/gstreamer/gstreamer/-/issues/1452

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3320>

videocodectestsink: Add YUV422 support

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3331>

gst-inspect: print doc urls for Rust plugins

We have documentation for them now after all.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3346>

qtdemux: use unsigned int types to store result of QT_UINT32

In a few cases throughout qtdemux, the results of QT_UINT32 were being
stored in a signed integer, which could cause subtle bugs in the case of
an integer overflow, even allowing the the result to equal a negative
number!

This patch prevents this by simply storing the results of this function
call properly in an unsigned integer type. Additionally, we fix up the
length checking with stsd parsing to prevent cases of child atoms
exceeding their parent atom sizes.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3344>

d3d11videosink: Always clear back buffer on resize

Swapchain may not need to be resized if the size of backbuffer
is equal to the previous size. Then previously rendered frame will be stay
on the screen. Do clear back buffer whenever resize() is called

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3341>

subparse: Fix non-closed tag handling.

Unclear what the goal was, but we could end up reading way past the next_tag.

Instead just move everything from after the end tag ('>') to the next_tag.

Fixes https://bugs.chromium.org/p/oss-fuzz/issues/detail?id=53040

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3340>

avutils: Add missing space in doc

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3330>

avviddec: Initialize std_compliance to default

This was missed in the initial patch.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3330>

qsv: Fix encoding error when input memory belongs to other GPU

Copy frame if VA display of input buffer is different from that
of encoder, like we've been doing on Windows

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3335>

identity: fix "handoff" signal docs

The docs list an extra pad argument, which doesn't
match the actual signal function signature. Probably
a copy'n'paste mistake when copying things from fakesink.

Fixes #1546

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3337>

va: baseenc: Do not import the VA surface from other display.

Fixes: https://gitlab.freedesktop.org/gstreamer/gstreamer/-/issues/1167
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3322>

va: compositor: Do not use allocator and pool from other display.

Just like the va decoder, the compositor should not use allocator
and pool from other display. Also, when importing the input buffer,
if it is from other display, we should fallback to memory copy.

Fixes: https://gitlab.freedesktop.org/gstreamer/gstreamer/-/issues/1167
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3322>

va: basetransform: Do not use allocator and pool from other display.

Just like the va decoder, the vpp and deinterlace should not use allocator
and pool from other display. Also, when importing the input buffer, if it
is from other display, we should fallback to memory copy.

Fixes: https://gitlab.freedesktop.org/gstreamer/gstreamer/-/issues/1167
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3322>

va: basedec: Do not use allocator and pool from other display.

The command line such as:
  gst-launch-1.0 -vf filesrc location=1.264 ! h264parse !
  vah264dec ! varenderD129postproc ! fakesink
The decoder here gets the allocation proposal from the vpp which is
on another GPU device. The allocator and pool belong to other display
and should not be used in the decoder.

Fixes: https://gitlab.freedesktop.org/gstreamer/gstreamer/-/issues/1167
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3322>

libs: va: Add API to peek the va buffer's display.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3322>

libs: va: Add API to peek the va memory's display.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3322>

libs: va: Add API to peek the allocator's display.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3322>

libs: va: Fix code style in gstvaallocator.h

Replace all tabs with spaces.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3322>

d3dvideosink: Update plugin description

Direct3D is too generic (there are various version numbers).
Clarify that this element is using the Direct3D9 API

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3333>

qt: initialize GError properly in gst_qt_get_gl_wrapcontext()

Spotted by Claus Stovgaard.

Fixes #1545

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3327>

gst-env: import exit from sys

On windows the Cmd Prompt for VS 2019 complains
that exit is not defined.

  File "C:/data/gstreamer/gst-env.py", line 622, in <module>
    exit(subprocess.call(args, close_fds=False, env=env))

NameError: name 'exit' is not defined
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3324>

qtmux: Add durations to raw audio buffers from the raw audio adapter in prefill mode

This ensures that a duration can also be calculated and stored for the
last buffer at EOS.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3321>

qtmux: Release object lock before posting an error message

GST_ELEMENT_ERROR() also takes the object lock and this would then
deadlock.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3321>

gst-validate-1.0: Don't leak arguments and context

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3319>

validate-scenario: Don't leak temporary path value

It's only use in the structure iterator

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3319>

validate-scenario: Don't leak temporary local variables structures

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3319>

lv2: Don't leak plugin information on registration

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3319>

ladspa: Don't leak plugin information on registration

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3319>

fdkaacenc: Properly terminate GEnumValue table

It should be terminated with a NULL entry, otherwise we just stray into the
realms of cryptographic libraries^W^W random memory usage.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3319>

libav: avcfg: Avoid brittle comparision

Subtracting a gint from another (or a guint from another) has no guarantees that
it will result in a gint.

Therefore do the actual comparision instead.

Also use the *actual* type for comparing flags (the field value types are different)

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3319>

libav: avcfg: Don't leak duplicate enumvalue entries

If we discard the duplicates, we also need to free the allocated strings.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3319>

imagesequencesrc; Fix leaks

* The path was leaked
* The custom buffer was never freed

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3319>

qtdemux: Fix cenc-related leaks

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3319>

deinterlace: Don't leak metas

There is no correlation between the frame being NULL and the metas not being
present.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3319>

validate: plug some leaks

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3319>

adaptivedemux2: Fix collection leaks

* The collection on the period was never unreffed
* The collection in the message handler was never unreffed

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3319>

parsebin: Fix leak

Release the reference on the collection which was given to us when extracting it
from the event/message.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3319>

mssdemux2: Update for adaptivedemux2 refactoring

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3314>

adaptivedemux2: Move stream_seek() to the Stream class

Move the last stream specific vfunc from the demux
class to the stream class.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3314>

adaptivedemux2: Refactor stream methods into the stream

Unlike the legacy elements, GstAdaptiveDemuxStream is a GObject now,
so a bunch of things that were actually stream methods on the
parent demux object can directly become stream methods now.

Move the stream class out to a header of its own.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3314>

hlsdemux2/m3u8: Implement EXT-X-GAP parsing

Read the EXT-X-GAP tag and set is_gap on the segment.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3314>

hlsdemux2/m3u8: Refactor parsing for readability

Small readability improvements in the parsing code

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3314>

adaptivedemux2/downloadhelper: Remove return val for download_request_add_buffer()

The function can't actually fail, and the only caller
was ignoring the result anyway, so remove the return value.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3314>

adaptivedemux2/downloadhelper: Add debug output of response headers

Dump the HTTP response headers at TRACE level

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3314>

adaptivedemux2/downloadhelper: Don't mark transfer as complete/error if cancelled.

If the state of the download request was reset to UNSENT,
it was cancelled. Don't update the state to COMPLETE or ERRORED
in on_read_ready().

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3314>

adaptivedemux2/downloadhelper: Ignore spurious read failure

Sometimes g_input_stream_read_all_finish() can return
0 bytes, but still succeed (return TRUE) and have more
data available later. Only finish the transfer
if it returns 0 bytes *and* FALSE with no error.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3314>

adaptivedemux2/downloadhelper: Fix function name

Fix a typo in the name of function download_request_despatch_progress()

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3314>

adaptivedemux2: Remove scheduler_lock mutex

Remove the old unused scheduler_lock

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3314>

adaptivedemux2: Hold tracks lock accessing input_period

The input_period is protected by the TRACKS_LOCK,
so make sure to hold that when accessing it.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3314>

adaptivedemux2: Add state checks and clean up obsolete variables

The cancelled flag was only set in the stream finalize()
method, after all activity on the stream has stopped anyway.
Replace uses of cancelled with checks on the stream state.

Remove the replaced flag, which was checked but never set
to TRUE anywhere any more.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3314>

gstpad: Fix non-serialized sticky event push

With non-serialized sticky events, such as GST_EVENT_INSTANT_RATE, we both want
to store the event (for later re-linking) *AND* push the event in a non-blocking
way.

We therefore must *not* propagate pending sticky events if the event is "sticky
or serialized" but only if it's "serialized"

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3254>

srt: Remove callers for which srt_bstats fails

This keeps them from accumulating in the element and in the stats while
the sink is not being fed, as long as we at least periodically grab
stats.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3156>

srt: Use simpler list operations for callers

Avoid `g_list_append` and `g_list_remove` (which have to scan the list)
and replace them with `g_list_prepend` and `g_list_delete_link`.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3156>

srt: Clean up poll/sock lifecycle

Make sure `srtobject->poll_id` is never invalid as long as `srtobject`
exists. Only remove our caller socket from it when the socket becomes
invalid.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3156>

srt: Clean up error handling

- Make the srt_epoll_wait loops more uniform.

- Error only via GError when possible; let the element send the error
  message. Avoids a second error message.

- Return 0 when cancelled. Avoids an error message from the element.

- Don't send an error message from send_headers when we're a server
  sink.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3156>

srt: Simplify socket stats

Don't hide stats depending on whether we're a sending or receiving
socket. While we're here, add some more debug logs.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3156>

srt: Replace stats accumulation with naive byte counting

srt_bstats cannot be used to get the stats of closed connections, so the
best we can do is keep the running count ourselves.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3156>

directshow: Validate enumerated AM_MEDIA_TYPE struct

The video info struct can be null if format is undefined

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3323>

vpx: Complete high bitdepth vp9 en/decoding support

Adding 12bits variant formats to en/decoder, and high bitdepth
4:4:4 (except for GBR) encoding support

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3298>

vpx: Define formats for compatibility

ifdef for enum values never work. Instead, define new enum type
and use it

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3298>

examples: d3d11: Fix GCC build error/warning

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3318>

qsvjpegenc: Add support for YUY2 format

Now GstD3D11 defines YUY2 format

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3316>

qsv: Add JPEG decoder

Because DXVA does not define JPEG decoding, we need this
vendor specific API for Windows

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3316>

d3d11: Add support for YUY2 format

YUY2 is a primary YUV 4:2:2 format for DXVA.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3316>

doc: avviddec: Update plugin cache

This is to add std-compliance property and GstFFMpegVidDec base class.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3311>

avviddec: Introduce a class for shared properties

Without a parent class,the documentation would need to be duplicated for
every CODECs. This patch adds an abstract class in between GstVideoDecoder
and the element.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3311>

avviddec: Expose std-compliance option

Default compliance mode (i.e., FF_COMPLIANCE_NORMAL) might not
strictly follow specification. This property will allow user
to specifiy expected compliance mode.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3311>

parse: do delayed set only if the target child was not found and fail otherwise

When using the child proxy notation (child::property=value) it may
happen that the target child does not exist at the time of parsing
(i.e: decodebin creates the encoder according to the contents of the
stream). On this cases, we want to delay the setting of the property
to later, when new elements are added. Previous logic performed a
delayed set even if the target child was found but the property
was not found in it. This should be treated as a failure because,
unlike missing elements, properties should not appear dynamically.
By not failing, typos in property names may go unnoticed to the end
user.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2908>