video: Fix NV12_16L32 size calculation

The subsampling of the second plane was not taken into account, resulting in a
16bit per pixel buffers instead of 12.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2512>

basetransform: fix critical if transform_caps() returned NULL

klass->transform_caps() may return NULL, which was raising this
critical:

  GStreamer-CRITICAL **: 12:23:56.243: gst_caps_is_subset: assertion 'subset != NULL' failed

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2528>

adaptivedemux2: fix plugin/element init

In case of per features registration such as the
customizable gstreamer-full library, each
element should check that the soup library can be loaded to
facilitate the element registration.

Initialize the debug categories properly

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2348>

directshow: Fix for uninitialized debug category warning

A method in the code is used in another place (device provider)

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2519>

va: Register base classes as abstract.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2521>

va: encoder: Use entrypoint type.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2521>

plugin: add Apache 2 license to known licenses

the licence in gstreamer/subprojects/gstreamer/gst/gstplugin.c
currently is defined to be one of:
LGPL GPL QPL GPL/QPL MPL BSD MIT/X11 0BSD Proprietary
The open source project for the kinesis plugin is using an
Apache 2.0 license. Because "Apache 2.0" is not one of the
supported licenses it automatically falls back to Proprietary.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2514>

va264enc: Fix typo.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2508>

va: encoder: Fix regression since commit 2042c2d4f0.

There's no need to compare with the open entrypoint to query the
encoder. This commit removes the checks to the configured entrypoint
with the parameter.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2508>

rtpptdemux: Don't GST_FLOW_ERROR when ignoring invalid packets

https://bugzilla.gnome.org/show_bug.cgi?id=741398 changed
rtpptdemux in 2014 to not post a GST_ELEMENT_ERROR on the
bus when dropping an invalid (non-RTP) packet, but still
returned GST_FLOW_ERROR upstream - so the pipeline still
stops, but now without a useful bus error.

Return GST_FLOW_OK instead, so the pipeline keeps
running. Some old telephony equipment can send invalid
packets before the real RTP traffic starts.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2520>

mfvideoencoder: Handle dynamic property update

Re-init MFT on property update to apply new encoding options

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2037>

mediafoundation: Run indent

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2037>

mfvideoenc: Remove redundant bitwise or operation

MFT_ENUM_FLAG_SORTANDFILTER_APPROVED_ONLY (0xc0) covers
MFT_ENUM_FLAG_SORTANDFILTER (0x40)

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2037>

mfvideoencoder: Early terminate d3d11 interop

Shared texture will work only for the same GPU

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2037>

mfvideoencoder: Check HRESULT code as well for GPU sync

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2037>

mfvideoencoder: Don't specify keyed mutex misc flag

Current implemetation copies textures using non-keyed mutex way.
D3D11_RESOURCE_MISC_SHARED_KEYEDMUTEX is pointless

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2037>

jack: Always use jack_free as specified by the docs

Fixes a crash on Windows due to a CRT mismatch. The JACK installation
still uses MSVCRT, and we the Universal CRT for both MinGW and MSVC.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2497>

jack: Add support for detecting libjack on Windows

No source code changes were necessary to get the plugin working on
Windows with MSVC.

Run QJackCtl and audiotestsrc ! jackaudiosink just works.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2497>

vacompositor: Pass video info pointer than copying it.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2496>

vacompositor: Remove useless consts.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2496>

d3d11decoder: Work around Intel DXVA driver crash

Intel DXVA driver crashes sometimes (from GPU thread) if
ID3D11VideoDecoder is released while there are outstanding view objects.
To make sure the object life cycle, holds an ID3D11VideoDecoder refcount
in GstD3D11Memory object.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2504>

ci: Make sure to always update subprojects

The command was not run in Windows jobs, better keep subprojects
handling in a single script.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2509>

qsvencoder: Bind internal texture to render target

... so that memory allocator can initialize texture with black color

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2502>

d3d11memory: Clear YUV texture with black color

We prefer black color as an initial texture color and
Direct3D11 runtime will initialize texture with zeros (except for alpha)
which is fine for RGB formats. But UV components of YUV texture
requires manual clear for black color.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2502>

tests: vaapi: remove unused header

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2503>

tests: vaapi: fix memleak in vaapioverlay test

Need to unmap the frame to ensure vaapioverlay can destruct/dispose.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2503>

tests: va: add simple vacompositor test

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2481>

msdkh265enc: Add intra refresh parameters

Note that for msdkh265enc, intra refresh parameters are only supported
under low-power mode. If these parameters are set without enabling lowpower,
they will be simply ignored.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2308>

msdkh264enc: Add intra refresh parameters

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2308>

tests: examples: Stop duplicating keyboard interaction handler code

Extract the code from QSV and use it for D3D11 and NVCODEC examples

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2468>

webrtc: fix build with older libnice versions

1) check for right macro name when checking for NICE_VERSION_CHECK

2) if libnice version is 0.1.18.1 this should not satisfy
   a NICE_VERSION_CHECK(0,1,19).

Fixes build with libnice 0.1.18.1 subproject checkout.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2499>

v4l2: Reset transfer in gst_v4l2_object_acquire_format()

get_colorspace() checks input caps transfer when mapping V4L2_XFER_FUNC_709
back to V4L2_COLORSPACE_BT2020 and GST_VIDEO_TRANSFER_BT2020_12. After
receiving source change event, decoder will G_FMT and S_FMT again. So need
to reset transfer when acquiring format to avoid using the old transfer.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2475>

amc: Don't map input buffer with write mode

When copying data out of a GStreamer buffer, we don't
need to map it writable.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2492>

webrtc: Use new libnice API to get the candidate relay address

Corresponding libnice API added in:
https://gitlab.freedesktop.org/libnice/libnice/-/merge_requests/229 (0.1.19)
https://gitlab.freedesktop.org/libnice/libnice/-/merge_requests/232 (0.1.20)

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1998>

webrtc: Ensure the NICE_CHECK_VERSION macro is available

This macro was introduced in libnice 0.1.19.1, so until we bump our libnice
dependency to 0.1.20 we have to vendor the macro.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1998>

webrtc: Refactor ICECandidateStats freeing logic to a dedicated function

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1998>

webrtcbin: Document IceCandidateStats and RTCIceCandidatePairStats

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1998>

webrtcstats: Update unit test for outbound rtp stats

"remote-id" is not guaranteed to present after commit 1deb034e3d873f577a38bc483e6256b0346ee874.
Thus, we should not fail the test if "remote-id" is not found.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1998>

webrtcstats: Improve selected candidate pair stats by adding ICE candidate info

The implementation follows w3.org specs:
* https://www.w3.org/TR/webrtc-stats/#icecandidate-dict*
* https://www.w3.org/TR/webrtc-stats/#candidatepair-dict*

Corresponding unit tests are also added.

Rebased and updated from
https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1462

Fixes #1207

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1998>

va: H264enc: Add encoder for VAEntrypointEncSliceLP.

We name vah264lpenc to support H264's VAEntrypointEncSliceLP encoder.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2472>

va: encoder: Do not reset the frame count when reconfig().

We should not reset the input/output_frame_count when some configure
changes. For example, the if resolution changes, the current way just
resets the frame count and make the PTS of the output buffer restart
from the original PTS of the first frame. That causes a lot of QOS
event and drop all the new frames.
We should only reset them when encoder start().

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2489>

va: encoder: change reset_state() to a virtual function of base class.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2489>

playbin3: Configure combiner on pad-added if needed

When collection is updated, decodebin3 exposes pad first and then
streams-selected message is posted.
The condition can cause a situation where playbin3 links non-existing
combiner/playsink pads (since streams-selected is not posted yet) with
new decodebin output pad. This commit will re-check selected/active
streams condition on pad-added and reconfigure output if needed.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2482>

cutter: Include running/stream-time in messages

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2490>

gstreamer-full-static: autopep on script

Pass autopep on generate_init_static_plugins.py

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1637>

gstreamer-full-static: Be more strict with plugin param

Check that the plugins are correctly separated by ";"
and do not accept "," or ":" inside.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1637>

vacompositor: check caps and info return values

Check return values for get caps and video info from caps.

This was flagged by static analysis.

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

soup: Fix plugin/element init

In case of per features registration such as the
customizable gstreamer-full library, each
element should check that the soup library can be loaded to
facilitate the element registration.

Initialize the debug category properly

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2349>

d3d11decoder: Don't print error log on negotiation failure

It may not be a critical error and even expected when shutting down
pipeline or flushing.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2483>

d3d11screencapture: Fix missing/outdated cursor shape

d3d11screencapture can miss a cursor shape to draw or draw an outdated cursor shape.
 - AcquireNextFrame only provides cursor shape when there is one update
 - current d3d11screencapture skips cursor shape when mouse is not drawn

So, if a gstreamer application uses d3d11screencapture with cursor initially not drawn
"show-cursor"=false and then switches this property to true, the cursor will not be
actually drawn until AcquireNextFrame provides a new cursor shape.
This commit makes d3d11screencapture always update the cursor shape information, even
if the mouse is not drawn. d3d11screencapture will always have the latest cursor shape
when requested to draw it.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2485>

meson: use better zlib dependency fallback

zlib is required, and if it isn't found it is checked several ways and
then forced via subproject(). This code was added in commit
b93e37592a3ccc0eaece1c8fef2d362b1e5fe685, to account for systems where
zlib doesn't have pkg-config files installed.

But Meson already does dependency fallback, and also, since 0.54.0, does
the in-between checks for find_library('z') and has_header('zlib.h') via
the "system" type dependency. Simplify dependency lookup by marking it
as required, which also makes sure that the console log doesn't
confusingly list "not found".

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2484>

va: Avoid unused variable if logging is disabled.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2478>

v4l2videodec: Fix missing handling of resolution-change

The pool process function may poll and get the resolution-change event
whenever it is not possible to share our buffers. This typically happen
when downstream does not support GstVideoMeta.

Not handling this would cause the decoder thread to exit silently and the
pipeline to stall.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2457>

v4l2videodec: Downgrade to info resolution-change trace

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2457>

structure: Fix serializing with new format inside arrays/lists

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1554>

va: Keep include's code style

Headers are in alphabetical order.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2465>

vafilter: Fix logging of unsupported alpha blending.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2477>

webrtcbin: reuese the same fec/rtx/red payload types for the same media payload

WHen bundling, if multiple medias are used with the same media payload, then
each of the fec/rtx/red additions would add a distinct payload.  This could
very easily overflow the available payload space.

Instead, track the relationship between the media payload value and
the relevant fec/rtx/red payload values and reuse them whenever
necessary, even when bundling.

e.g.

...
a=group:BUNDLE video0 video1
m=video 9 UDP/SAVPF 96 97
a=mid:video0
a=rtpmap:96 VP8/90000
a=rtpmap:97 rtx/90000
a=fmtp:97 apt=96
...
m=video 9 UDP/SAVPF 96 97
a=mid:video1
a=rtpmap:96 VP8/90000
a=rtpmap:97 rtx/90000
a=fmtp:97 apt=96
...

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2474>

shout2: fix compiler warnings and bump req to libshout >= 2.4.2

Fix compiler warnings with latest libshout version (2.4.6).

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2473>

shout2send: Adding send-title-info and user-agent options

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2420>

tests: examples: Fix typo in QSV encoder example

Fixing width adjustment

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2467>

qsv: Add AV1 encoder element

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2467>

qsvencoder: Move common property to baseclass

adapter-luid and device-path are common for all codecs

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2467>

va: baseenc: No need to check the frame pointer in handle_frame().

No need to check the NULL pointer of "frame" in error handling, it
is assured non-NULL by base class and we already use it everywhere
in this function.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2471>

va: baseenc: include the "vacompat.h" for old version glib.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2463>

va: vpp: enable compositor

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2332>

va: vpp: add compositor

New implementation of a VA-API compositor with blend feature.

Various chunks of code was taken from vavpp.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2332>

va: Apply the va base encoder to the vah264enc.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2330>

va: Add the GstVaBaseEnc common object for all va encoders.

Just like the GstVaBaseDec and GstVaBaseTransform, this base encoder
will handle all common logic for all va encoders.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2330>

d3d11device: Enhance format enumeration and debugging

* Enhance debug log to print human readable D3D11_FORMAT_SUPPORT flags
  value, instead of packed numeric flagset value.
* Only device supported format will be added to format table.
  Depending on device feature level (i.e., D3D9 feature devices),
  16bits formats will not be supported. Although there might be formats
  we deinfed but not supported, it will not be a major issue in practice
  since our D3D11 implementation does not support legacy devices already
  (known limitation) and also old d3dvideosink will be promoted in that case.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2441>

ci: run gst-inspect after the build

useful to quick inspect the elements built, later on we can
start verifying the list of elements against a known good list.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2458>

ci: run ninja install post build

Make sure no weird regressions creep in.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2458>

pbutils: Add GST_PBUTILS_CAPS_DESCRIPTION_FLAG_METADATA and ONVIF XML Timed MetaData

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2453>

qtdemux: Add support for ONVIF XML Timed MetaData

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2453>

qtdemux: Add parsing/dumping of nmhd / metx boxes

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2453>

qtdemux: Parse styp box for informational purposes

And include some more details in the debug logs for the ftyp box too.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2453>

v4l2: set default resolution if caps has no such information

Output may attemp to set the width and height to zero values if
caps has no such information, which will cause capture get invalid
dimensions. Then decoder reports negotiation failure.

So need to set default resolution if caps has no such information.
Real values can be set again until source change event is signaled.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2400>

jpegparse: Rewrite element.

Now it uses the JPEG parser in libgstcodecparsers, while the whole
code is simplified by relying more in baseparser class for tag
handling.

The element now signals chroma-format and default framerate is 0/1,
which is for still-images.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1473>

tests: jpegparse: Mark data as static.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1473>

jpegformat: Use codecparsers' for JPEG markers.

Instead of repeating JPEG markers definition, this patch uses those
defined in gstcodecparsers library.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1473>

jpegparse: Headers cleanup

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1473>

rtsp-media: Correct logic on GstRTSPStreamBlocking message reception

We must take into account the receiving streams as well when calculating
the expected number of the received GstRTSPStreamBlocking messages.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2429>

datachannel: Notify low buffered amount according to spec

Quoting
https://www.w3.org/TR/webrtc/#dom-rtcdatachannel-bufferedamountlowthreshold

The bufferedAmountLowThreshold attribute sets the threshold at which the
bufferedAmount is considered to be low. When the bufferedAmount decreases from
above this threshold to **equal** or below it, the bufferedamountlow event fires.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2448>

gstreamer/gst/gstconfig.h.in: Add support for LoongArch

While current and future LoongArch machines that are supposed to run
GStreamer all support unaligned accesses, there might be future
lower-end cores (e.g. the embedded product line) without such support,
and we may not want to penalize these use cases.

So, mark LoongArch as not supporting unaligned accesses for now, and
hope the compilers do a good job optimizing them. We can always flip
switch later.

Suggested-by: CHEN Tao <redeast_cn@outlook.com>
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2443>

rtcpbuffer: Allow padding on first reduced size packets

It is valid to have the padding set to 1 on the first packet and it
happens very often from TWCC packets coming from libwebrtc. This means
that we were totally ignoring many TWCC packets.

Fix test that checked that a first packet with padding was not valid and
instead test a single twcc packet with padding to check precisely what
this patch was about.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2422>

tsmux: Make sure to set srcpad caps under all conditions before outputting the first buffer

Fixes https://gitlab.freedesktop.org/gstreamer/gstreamer/-/issues/1218

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2438>

ci: use the correct vs toolcahin version for cerbero

This seems to be a leftover from when we upgraded from vs15
to vs16.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2395>

ci: use LONGPATH for the windows container

Sometimes the CI might create paths that are bigger than the 260
character limit.

ex:
```
error: unable to create file subprojects/gst-integration-testsuites/testsuites/validate/h264/parse.trickmode_predicted.seek_trickmode_predicted/flow-expectations/log-parse-src-expected: Filename too long
```

https://docs.microsoft.com/en-us/windows/win32/fileio/maximum-file-path-limitation

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2395>

ci: remove windows-rust jobs

These are moved now into gstreamer-rs repository.

https://gitlab.freedesktop.org/gstreamer/gstreamer-rs/-/merge_requests/701

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2395>

ci: re-enable artifacts upload for the windows jobs

Hopefully the timeout issues won't affect the new runner

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2395>

ci: Install Media Foundations on the windows container

In order to do this, we need to switch to the Server container
image, instead of the ServerCore we were using till now.

This image also only has Server2022 base variants.

This is a continuation of:

https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1783

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2395>

ges/videourisource: handle non-1/1 PAR source videos

The automatic scaling done by framepositioner does not account for the
par of the video source.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2423>

d3d11device: Don't print ERROR log on open failure

Device open can be failed in various reasons, but it's unlikely
our problem

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2433>

sdpdemux: Release request pads from rtpbin when freeing a stream

Otherwise the pads of the rtpbin stay around forever and are leaked.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2437>

splitmuxsrc: Re-queue sticky events after probing.

When processing the first event after probing the
file and being activated, requeue sticky events
as there's no requirement that demuxers send tag
and other events again after a seek - that's
why they're sticky.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2432>

deinterlace: Clean up error handling in chain and _push_history

- Consistently unref the chained buffer at the end of the chain
  function, if we're not handing it off to `gst_pad_push`. This avoids a
  few buffer leaks in the error paths in `_chain` and `_push_history`.
- When mapping the video frame fails, return a flow error instead of
  crashing.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2428>

splitmuxsink: When flushing, exit handle_mq_input quickly

If we just break the loop, we might run into the `gop != NULL` assert
that follows it. Rather, exit immediately with flushing flow.

Also use this flushing mechanism when we release a pad. This avoids
having an extra flag.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1030>

splitmuxsink: Avoid deadlock on release, harder

Unlock after broadcasting and wait for the pad to be free before
relocking the muxer, giving the input probe a chance to react to our
broadcast.

Improves the fix from
https://gitlab.freedesktop.org/gstreamer/gst-plugins-good/-/merge_requests/838.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1030>

appsink: Fix race condition on caps handling

Background:

Whenever a caps event is received by appsink, the caps are stored in the
same internal queue as buffers. Only when enough buffers have been
popped from the queue to reach the caps, `priv->sample` gets its caps
updated to match, so that they are correct for the following buffers.

Note that as far as upstream elements are concerned, the caps of appsink
are updated immediately when the CAPS event is sent. Samples pulled from
appsink retain the old caps until a later buffer -- one that was sent by
upstream elements after the new caps -- is pulled.

The race condition:

When a flush is received, appsink clears the entire internal queue. The
caps of `priv->sample` are not updated as part of this process, and
instead remain as those of the sample that was last pulled by the user.

This leaves open a race condition where:
1. Upstream sends a new caps event, and possibly some buffers for the
   new caps.
2. Upstream sends a flush (possibly from a different thread).
3. Upstream sends a new buffer for the new caps. Since as far as
   upstream is concerned, appsink caps are the new caps already, no new
   CAPS event is sent.
4. The appsink user pulls a sample, having not pulled before enough
   samples to reach the buffers sent in step 1.

Bug: the pulled sample has the old caps instead of the new caps.

Fixing the race condition:

To avoid this problem, when a buffer is received after a flush,
`priv->sample`'s caps should be updated with the current caps before the
buffer is added to the internal queue.

Interestingly, before this patch, appsink already had code for this, in
gst_app_sink_render_common():

    /* queue holding caps event might have been FLUSHed,
     * but caps state still present in pad caps */
    if (G_UNLIKELY (!priv->last_caps &&
            gst_pad_has_current_caps (GST_BASE_SINK_PAD (psink)))) {
      priv->last_caps = gst_pad_get_current_caps (GST_BASE_SINK_PAD (psink));
      gst_sample_set_caps (priv->sample, priv->last_caps);
      GST_DEBUG_OBJECT (appsink, "activating pad caps %" GST_PTR_FORMAT,
          priv->last_caps);
    }

This code assumes `priv->last_caps` is reset when a flush is received,
which makes sense, but unfortunately, there was no code in the flush
code path resetting it.

This patch adds such code, therefore fixing the race condition. A unit
test demonstrating the bug and testing its behavior with the fix has
also been added.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2413>

videoaggregator: unref temporary caps

The "possible_caps" needs unref after finished using to
avoid memory leak.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2430>